<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#ecca99">
    <p><font size="+1">Gary,<br>
      </font></p>
    <p><font size="+1">You're minimizing my example, and frankly,  I
        don't believe DL models will "characterize some aspects of
        Psychology reasonably well"...  and before getting to your
        netflix bingeing.. one of the serious problems with DL models,
        say comparing to neural recordings, using correlation matrices
        there seem to be some correspondence to make.. but sometimes
        little more than a roscharch test.    In general, I think it
        will be hard to show good correspondence to decision processing,
        episodic memory, compound stimulus conditioning,  and various
        perceptual illusions and transformations.    In general the DL
        focus on classification and translation has created models very
        unlikely to easily model cognitive and perceptual phenomena.   
        Models like Grossberg are curated to account for specific
        effects and over a lifetime have done a better job at making
        sense of psychological/neural phenomena then any other neural
        models I know about, whether one subscribes to the details of
        the modeling is another issue.  <br>
      </font></p>
    <p><font size="+1">So in the perceptual/cognition abstraction task I
        discussed it is gob-smacking that JUST ADDING LAYERS solves this
        really critical failure of backpropagation, barely noted by most
        of the neural network community focused on better benchmarks.</font></p>
    <p><font size="+1">As to Netflix titles.   I agree, cognitive models
        should be adaptable and responsive to updates that create more
        predictive outcomes for the agent using them.   This in no ways
        means the cognitive model must be symbolic or rule based.   This
        was something that was true in the 1980s and perhaps truer
        today.</font></p>
    <p><font size="+1">This is clearly a critical aspect of GPT models..
        what are the cognitive models that they are building or are they
        just high dimensional, phrase-structure blobs that do similarity
        analysis and return a nearby phrase-structure response, which
        happens to sound good.<br>
      </font></p>
    <p><font size="+1">Steve<br>
      </font></p>
    <div class="moz-cite-prefix">On 2/7/22 5:07 PM, Gary Marcus wrote:<br>
    </div>
    <blockquote type="cite"
      cite="mid:BB0190E0-193B-4AFC-9AE1-A7C0961E8D60@nyu.edu">
      <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
      Stephen,
      <div class=""><br class="">
      </div>
      <div class="">I don’t doubt for a minute that deep learning can
        characterize some aspects of psychology reasonably well; but
        either it needs to expands its borders or else be used  in
        conjunction with other techniques. Take for example the name of
        the new Netflix show</div>
      <div class=""><br class="">
      </div>
      <div class=""><span style="margin: 0px; padding: 0px; border: 0px;
          font-family: "Open Sans", sans-serif; font-style:
          italic; font-stretch: inherit; line-height: inherit;
          font-size: 16px; vertical-align: baseline; caret-color:
          rgb(51, 51, 51); color: rgb(51, 51, 51);" class="">The Woman
          in the House Across the Street from the Girl in the Window</span></div>
      <div class=""><span style="margin: 0px; padding: 0px; border: 0px;
          font-family: "Open Sans", sans-serif; font-style:
          italic; font-stretch: inherit; line-height: inherit;
          font-size: 16px; vertical-align: baseline; caret-color:
          rgb(51, 51, 51); color: rgb(51, 51, 51);" class=""><br
            class="">
        </span></div>
      <div class="">Most of us can infer, compositionally, from that
        unusually long noun phrase, that the title is a description of
        particular person, that the title is not a complete sentence,
        and that the woman in question lives in a house; we also infer
        that there is a second, distinct person (likely a child) across
        the street, and so forth. We can also use some knowledge of
        pragmatics to infer that the woman in question is likely to be
        the protagonist in the show. Current systems still struggle with
        that sort of thing. </div>
      <div class=""><br class="">
      </div>
      <div class="">We can then watch the show (I watched a few minutes
        of Episode 1) and quickly relate the title to the protagonist’s
        mental state, start to develop a mental model of the
        protagonist’s relation to her new neighbors, make inferences
        about whether certain choices appear to be “within character”,
        empathize with character or question her judgements, etc, all
        with respect to a mental model that is rapidly encoded and
        quickly modified.</div>
      <div class=""><br class="">
      </div>
      <div class="">I think that an understanding of how people build
        and modify such models would be extremely valuable (not just for
        fiction for everyday reality), but I don’t see how deep learning
        in its current form gives us much purchase on that. There is
        plenty of precedent for the kind of mental processes I am
        sketching (e.g  Walter Kintsch’s work on text comprehension;
        Kamp/Kratzer/Heim work on discourse representation, etc) from
        psychological and linguistic perspectives, but almost no current
        contact in the neural network community with these well-attested
        psychological processes. </div>
      <div class=""><br class="">
      </div>
      <div class="">Gary<br class="">
        <div><br class="">
          <blockquote type="cite" class="">
            <div class="">On Feb 7, 2022, at 6:01 AM, Stephen José
              Hanson <<a href="mailto:jose@rubic.rutgers.edu"
                class="" moz-do-not-send="true">jose@rubic.rutgers.edu</a>>
              wrote:</div>
            <br class="Apple-interchange-newline">
            <div class="">
              <meta http-equiv="Content-Type" content="text/html;
                charset=UTF-8" class="">
              <div text="#000000" bgcolor="#ecca99" class="">
                <p class=""><font class="" size="+1">Gary,</font></p>
                <p class=""><font class="" size="+1">This is one of the
                    first posts of yours, that I can categorically agree
                    with!</font></p>
                <p class=""><font class="" size="+1">I think building
                    cognitive models through *some* training regime or
                    focused sampling or architectures or something but
                    not explicit, for example. <br class="">
                  </font></p>
                <p class=""><font class="" size="+1">The other
                    fundamental cognitive/perceptual capability in this
                    context is the ability of Neural Networks to do what
                    Shepard (1970; Garner 1970s), had modeled as
                    perceptual separable processing (finding parts) and
                    perceptual integral process (finding covariance and
                    structure).</font></p>
                <p class=""><font class="" size="+1">Shepard argued
                    these fundamental perceptual processes were
                    dependent on development and learning.   <br
                      class="">
                  </font></p>
                <p class=""><font class="" size="+1">A task was created
                    with double dissociation of a categorization
                    problem.  In one case: separable ( in effect,
                    uncorrelated  features in the stimulus) were
                    presented in categorization task that required you
                    pay attention to at least 2 features at the same
                    time to categorize correctly ("condensation").  in
                    the other case: integral stimuli (in effect
                    correlated features in stimuli) were presented  in a
                    categorization task that required you to ignore the
                    correlation and do categorize on 1 feature at a time
                    ("filtration").   This produced a  result that
                    separable stimuli were more quickly learned in
                    filtration tasks then integral stimuli in
                    condensation tasks.  Non-intuitively,  Separable
                    stimuli are learned more slowly in condensation
                    tasks then integral stimuli then in filtration
                    tasks.   In other words attention to feature
                    structure could cause improvement in learning or
                    interference.   Not that surprising.. however--<br
                      class="">
                  </font></p>
                <p class=""><font class="" size="+1">In the 1980s NN
                    with single layers (Backprop) *could not* replicate
                    this simple problem indicating that the cognitive
                    model was somehow inadequate.     Backprop simply
                    learned ALL task/stimuli parings at the same rate,
                    ignoring the subtle but critical difference.  It
                    failed.<br class="">
                  </font></p>
                <p class=""><font class="" size="+1">Recently we
                    (<a class="moz-txt-link-freetext"
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.frontiersin.org_articles_10.3389_fpsyg.2018.00374_full-3F-26utm-5Fsource-3DEmail-5Fto-5Fauthors-5F-26utm-5Fmedium-3DEmail-26utm-5Fcontent-3DT1-5F11.5e1-5Fauthor-26utm-5Fcampaign-3DEmail-5Fpublication-26field-3D-26journalName-3DFrontiers-5Fin-5FPsychology-26id-3D284733&d=DwMDaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=UoNnjRqVBL_CqkrYum3TEgMe4-81VubghRckfDUNtQ8tcpm40eCYjC7e9nor929C&s=U5bZxiHTjY26KZ_tX3LaoOk3ok6HI2wTaC5QOUqGbqc&e="
                      moz-do-not-send="true">https://www.frontiersin.org/articles/10.3389/fpsyg.2018.00374/full?&utm_source=Email_to_authors_&utm_medium=Email&utm_content=T1_11.5e1_author&utm_campaign=Email_publication&field=&journalName=Frontiers_in_Psychology&id=284733</a>)
                    were able to show that JUST BY ADDING LAYERS the DL 
                    does match to human performance.</font></p>
                <p class=""><font class="" size="+1">What are the layers
                    doing?   We offer an possible explanation that needs
                    testing.    Layers, appear to create a type of
                    buffer that allows the network to "curate", feature
                    detectors that are spatially distant from the input
                    (conv layer, for example), this curation comes in 
                    various attention forms (something in that will
                    appear in a new paper--not enough room here), which
                    appears to qualitatively change the network
                    processing states, and cognitive capabilities.    
                    Well, that's the claim.  <br class="">
                  </font></p>
                <p class=""><font class="" size="+1">The larger point,
                    is that apparently architectures interact with
                    learning rules, in ways that can cross this
                    symbolic/neural river of styx, without falling into
                    it.<br class="">
                  </font></p>
                <p class=""><font class="" size="+1">Steve<br class="">
                  </font></p>
                <p class=""><font class="" size="+1"><br class="">
                  </font></p>
                <div class="moz-cite-prefix">On 2/5/22 10:38 AM, Gary
                  Marcus wrote:<br class="">
                </div>
                <blockquote type="cite"
                  cite="mid:537DF004-25CE-45A2-8155-D7E6018F4EE5@nyu.edu"
                  class="">
                  <meta http-equiv="content-type" content="text/html;
                    charset=UTF-8" class="">
                  <div dir="ltr" class="">There is no magic in
                    understanding, just computation that has been
                    realized in the wetware of humans and that
                    eventually can be realized in machines. But
                    understanding is not (just) learning.</div>
                  <div dir="ltr" class=""><br class="">
                  </div>
                  <div dir="ltr" class="">
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;">Understanding </span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">incorporates (or
                        works in tandem with) learning - but also,
                        critically, in tandem with inference, <i
                          class="">and the development and maintenance
                          of cognitive models</i>.</span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class=""> </span> <span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">Part of developing
                        an understanding of cats in general is to learn
                        long term-knowledge about their properties, both
                        directly (e.g., through observation) and
                        indirectly (eg through learning facts</span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class=""> </span> <span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">about animals in
                        general that can be extended to cats), often
                        through inference (if all animals have DNA, and
                        a cat is an animal, it must also have DNA).</span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class=""> </span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">  </span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">The understanding
                        of a particular cat also involves direct
                        observation, but also inference (eg</span><span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class=""> </span> <span
                        style="font-family: UICTFontTextStyleBody;
                        font-size: 17.46px;" class="">one might surmise
                        that the reason that Fluffy is running about the
                        room is that Fluffy suspects there is a mouse
                        stirring somewhere nearby). </span><span
                        class="s3" style="font-family:
                        UICTFontTextStyleEmphasizedBody; font-weight:
                        bold; font-size: 17.46px;">But all of that, I
                        would say, is subservient to the construction of
                        cognitive models that can be routinely updated </span><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;">(e.g.,
                        Fluffy is currently in the living room,
                        skittering about, perhaps looking for a mouse).</span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;"><br
                          class="">
                      </span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;"> In
                        humans, those dynamic, relational models, which
                        form part of an understanding, can support
                        inference (if Fluffy is in the living room, we
                        can infer that Fluffy is not outside, not lost,
                        etc). Without such models - which I think
                        represent a core part of understanding - AGI is
                        an unlikely prospect.</span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      min-height: 22.9px; caret-color: rgb(0, 0, 0);"
                      class=""><span class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;"><br
                          class="">
                      </span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;">Current
                        neural networks, as it happens, are better at
                        acquiring long-term knowledge (cats have
                        whiskers) than they are at dynamically updating
                        cognitive models in real-time. LLMs like GPT-3
                        etc lack the kind of dynamic model that I am
                        describing. To a modest degree they can
                        approximate it on the basis of large samples of
                        texts, but their ultimate incoherence stems from
                        the fact that they do not have robust internal
                        cognitive models that they can update on the
                        fly. </span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      min-height: 22.9px; caret-color: rgb(0, 0, 0);"
                      class=""><span class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;"></span><br
                        class="">
                    </div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;">Without
                        such cognitive models you can still capture some
                        aspects of understanding (eg predicting that
                        cats are likely to be furry), but things fall
                        apart quickly; inference is never reliable, and
                        coherence is fleeting.</span></div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      min-height: 22.9px; caret-color: rgb(0, 0, 0);"
                      class=""><span class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;"></span><br
                        class="">
                    </div>
                    <div style="margin: 0px; font-stretch: normal;
                      font-size: 17.5px; line-height: normal;
                      caret-color: rgb(0, 0, 0);" class=""><span
                        class="s1" style="font-family:
                        UICTFontTextStyleBody; font-size: 17.46px;">As a
                        final note, one of the most foundational
                        challenges in constructing adequate cognitive
                        models of the world is to have a clear
                        distinction between individuals and kinds; as I
                        emphasized 20 years ago (in The Algebraic Mind),
                        this has always been a weakness in neural
                        networks, and I don’t think that the type-token
                        problem has yet been solved. </span></div>
                  </div>
                  <div dir="ltr" class=""><br class="">
                  </div>
                  <div dir="ltr" class="">Gary</div>
                  <div dir="ltr" class=""><br class="">
                  </div>
                  <div dir="ltr" class=""><br class="">
                    <blockquote type="cite" class="">On Feb 5, 2022, at
                      01:31, Asim Roy <a class="moz-txt-link-rfc2396E"
                        href="mailto:ASIM.ROY@asu.edu"
                        moz-do-not-send="true"><ASIM.ROY@asu.edu></a>
                      wrote:<br class="">
                      <br class="">
                    </blockquote>
                  </div>
                  <blockquote type="cite" class="">
                    <div dir="ltr" class="">
                      <meta http-equiv="Content-Type"
                        content="text/html; charset=UTF-8" class="">
                      <meta name="Generator" content="Microsoft Word 15
                        (filtered medium)" class="">
                      <!--[if !mso]><style>v\:* {behavior:url(#default#VML);}
o\:* {behavior:url(#default#VML);}
w\:* {behavior:url(#default#VML);}
.shape {behavior:url(#default#VML);}
</style><![endif]-->
                      <style class="">@font-face
        {font-family:"Cambria Math";
        panose-1:2 4 5 3 5 4 6 3 2 4;}@font-face
        {font-family:Calibri;
        panose-1:2 15 5 2 2 2 4 3 2 4;}p.MsoNormal, li.MsoNormal, div.MsoNormal
        {margin:0in;
        font-size:11.0pt;
        font-family:"Calibri",sans-serif;}a:link, span.MsoHyperlink
        {mso-style-priority:99;
        color:blue;
        text-decoration:underline;}span.EmailStyle18
        {mso-style-type:personal-reply;
        font-family:"Calibri",sans-serif;
        color:windowtext;}.MsoChpDefault
        {mso-style-type:export-only;
        font-family:"Calibri",sans-serif;}div.WordSection1
        {page:WordSection1;}</style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]-->
                      <div class="WordSection1">
                        <p class="MsoNormal">All,<o:p class=""></o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal">I think the broader
                          question was “understanding.” Here are two
                          Youtube videos showing simple robots
                          “learning” to walk. They are purely physical
                          systems. Do they “understand” anything – such
                          as the need to go around an obstacle, jumping
                          over an obstacle, walking up and down stairs
                          and so on? By the way, they “learn” to do
                          these things on their own, literally
                          unsupervised, very much like babies. The basic
                          question is: what is “understanding” if not
                          “learning?” Is there some other mechanism
                          (magic) at play in our brain that helps us
                          “understand?” <o:p class=""></o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.youtube.com_watch-3Fv-3Dgn4nRCC9TwQ&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=Knv_0zpl6J7FTpxevgUOS8qJpyvPOjpOXdLYhyOr6PnKQiWgHaftEAfPvwWb_IAB&s=zdQA6enDajD46kwz-nti6FBklz-72dzlA9NLEzRW1TY&e="
                            moz-do-not-send="true" class="">https://www.youtube.com/watch?v=gn4nRCC9TwQ</a><o:p
                            class=""></o:p></p>
                        <p class="MsoNormal"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.youtube.com_watch-3Fv-3D8sO7VS3q8d0&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=Knv_0zpl6J7FTpxevgUOS8qJpyvPOjpOXdLYhyOr6PnKQiWgHaftEAfPvwWb_IAB&s=PRhn1hhcfzNtbKXIZpOAM4lyyMp39202wE7Uu4MWg5M&e="
                            moz-do-not-send="true" class="">https://www.youtube.com/watch?v=8sO7VS3q8d0</a><o:p
                            class=""></o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal">Asim Roy<o:p class=""></o:p></p>
                        <p class="MsoNormal">Professor, Information
                          Systems<o:p class=""></o:p></p>
                        <p class="MsoNormal">Arizona State University<o:p
                            class=""></o:p></p>
                        <p class="MsoNormal"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__lifeboat.com_ex_bios.asim.roy&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=oDRJmXX22O8NcfqyLjyu4Ajmt8pcHWquTxYjeWahfuw&e="
                            target="_blank" moz-do-not-send="true"
                            class="">Lifeboat Foundation Bios: Professor
                            Asim Roy</a><o:p class=""></o:p></p>
                        <p class="MsoNormal"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__isearch.asu.edu_profile_9973&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=jCesWT7oGgX76_y7PFh4cCIQ-Ife-esGblJyrBiDlro&e="
                            target="_blank" moz-do-not-send="true"
                            class="">Asim Roy | iSearch (asu.edu)</a><o:p
                            class=""></o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <div style="border:none;border-top:solid #E1E1E1
                          1.0pt;padding:3.0pt 0in 0in 0in" class="">
                          <p class="MsoNormal"><b class="">From:</b> Ali
                            Minai <a class="moz-txt-link-rfc2396E"
                              href="mailto:minaiaa@gmail.com"
                              moz-do-not-send="true"><minaiaa@gmail.com></a>
                            <br class="">
                            <b class="">Sent:</b> Friday, February 4,
                            2022 11:38 PM<br class="">
                            <b class="">To:</b> Asim Roy <a
                              class="moz-txt-link-rfc2396E"
                              href="mailto:ASIM.ROY@asu.edu"
                              moz-do-not-send="true"><ASIM.ROY@asu.edu></a><br
                              class="">
                            <b class="">Cc:</b> Gary Marcus <a
                              class="moz-txt-link-rfc2396E"
                              href="mailto:gary.marcus@nyu.edu"
                              moz-do-not-send="true"><gary.marcus@nyu.edu></a>;
                            Danko Nikolic <a
                              class="moz-txt-link-rfc2396E"
                              href="mailto:danko.nikolic@gmail.com"
                              moz-do-not-send="true"><danko.nikolic@gmail.com></a>;
                            Brad Wyble <a class="moz-txt-link-rfc2396E"
                              href="mailto:bwyble@gmail.com"
                              moz-do-not-send="true"><bwyble@gmail.com></a>;
                            <a class="moz-txt-link-abbreviated"
                              href="mailto:connectionists@mailman.srv.cs.cmu.edu"
                              moz-do-not-send="true">connectionists@mailman.srv.cs.cmu.edu</a>;
                            AIhub <a class="moz-txt-link-rfc2396E"
                              href="mailto:aihuborg@gmail.com"
                              moz-do-not-send="true"><aihuborg@gmail.com></a><br
                              class="">
                            <b class="">Subject:</b> Re: Connectionists:
                            Stephen Hanson in conversation with Geoff
                            Hinton<o:p class=""></o:p></p>
                        </div>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <div class="">
                          <div class="">
                            <p class="MsoNormal">Asim<o:p class=""></o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal"><o:p class=""> </o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal">Of course there's
                              nothing magical about understanding, and
                              the mind has to emerge from the physical
                              system, but our AI models at this point
                              are not even close to realizing how that
                              happens. We are, at best, simulating a
                              superficial approximation of a few parts
                              of the real thing. A single, integrated
                              system where all the aspects of
                              intelligence emerge from the same deep,
                              well-differentiated physical substrate is
                              far beyond our capacity. Paying more
                              attention to neurobiology will be
                              essential to get there, but so will paying
                              attention to development - both physical
                              and cognitive - and evolution. The
                              configuration of priors by evolution is
                              key to understanding how real intelligence
                              learns so quickly and from so little. This
                              is not an argument for using genetic
                              algorithms to design our systems, just for
                              understanding the tricks evolution has
                              used and replicating them by design.
                              Development is more feasible to do
                              computationally, but hardly any models
                              have looked at it except in a superficial
                              sense. Nature creates basic intelligence
                              not so much by configuring functions by
                              explicit training as by tweaking,
                              modulating, ramifying, and combining
                              existing ones in a multi-scale
                              self-organization process. We then learn
                              much more complicated things (like playing
                              chess) by exploiting that substrate, and
                              using explicit instruction or learning by
                              practice. The fundamental lesson of
                              complex systems is that complexity is
                              built in stages - each level exploiting
                              the organization of the level below it. We
                              see it in evolution, development, societal
                              evolution, the evolution of technology,
                              etc. Our approach in AI, in contrast, is
                              to initialize a giant, naive system and
                              train it to do something really
                              complicated - but really specific - by
                              training the hell out of it. Sure, now we
                              do build many systems on top of
                              pre-trained models like GPT-3 and BERT,
                              which is better, but those models were
                              again trained by the same none-to-all
                              process I decried above. Contrast that
                              with how humans acquire language, and how
                              they integrate it into their *entire*
                              perceptual, cognitive, and behavioral
                              repertoire, not focusing just on this or
                              that task. The age of symbolic AI may have
                              passed, but the reductionistic mindset has
                              not. We cannot build minds by chopping it
                              into separate verticals.<o:p class=""></o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal"><o:p class=""> </o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal">FTR, I'd say that the
                              emergence of models such as GLOM and
                              Hawkins and Ahmed's "thousand brains" is a
                              hopeful sign. They may not be "right", but
                              they are, I think, looking in the right
                              direction. With a million miles to go!<o:p
                                class=""></o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal"><o:p class=""> </o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal">Ali<o:p class=""></o:p></p>
                          </div>
                          <div class="">
                            <p class="MsoNormal"><o:p class=""> </o:p></p>
                          </div>
                          <div class="">
                            <div class="">
                              <div class="">
                                <div class="">
                                  <div class="">
                                    <div class="">
                                      <div class="">
                                        <div class="">
                                          <div class="">
                                            <div class="">
                                              <div class="">
                                                <div class="">
                                                  <div class="">
                                                    <div class="">
                                                      <div class="">
                                                        <p
                                                          class="MsoNormal"><b
                                                          class="">Ali
                                                          A. Minai,
                                                          Ph.D.</b><br
                                                          class="">
                                                          Professor and
                                                          Graduate
                                                          Program
                                                          Director<br
                                                          class="">
                                                          Complex
                                                          Adaptive
                                                          Systems Lab<br
                                                          class="">
                                                          Department of
                                                          Electrical
                                                          Engineering
                                                          & Computer
                                                          Science<o:p
                                                          class=""></o:p></p>
                                                      </div>
                                                      <div class="">
                                                        <p
                                                          class="MsoNormal">828
                                                          Rhodes Hall<o:p
                                                          class=""></o:p></p>
                                                      </div>
                                                      <div class="">
                                                        <p
                                                          class="MsoNormal">University
                                                          of Cincinnati<br
                                                          class="">
                                                          Cincinnati, OH
                                                          45221-0030<o:p
                                                          class=""></o:p></p>
                                                      </div>
                                                      <div class="">
                                                        <p
                                                          class="MsoNormal"><br
                                                          class="">
                                                          Phone: (513)
                                                          556-4783<br
                                                          class="">
                                                          Fax: (513)
                                                          556-7326<br
                                                          class="">
                                                          Email: <a
                                                          href="mailto:Ali.Minai@uc.edu"
target="_blank" moz-do-not-send="true" class="">Ali.Minai@uc.edu</a><br
                                                          class="">
                                                                    <a
href="mailto:minaiaa@gmail.com" target="_blank" moz-do-not-send="true"
                                                          class="">minaiaa@gmail.com</a><br
                                                          class="">
                                                          <br class="">
                                                          WWW: <a
href="https://urldefense.com/v3/__http:/www.ece.uc.edu/*7Eaminai/__;JQ!!IKRxdwAv5BmarQ!Jd2XhTzWg6HDp9IPjlyNv847sUdhGDNfsnqZQ0gy1_mu-CfyUdpBMswhfqdbZTo$"
target="_blank" moz-do-not-send="true" class="">
https://eecs.ceas.uc.edu/~aminai/</a><o:p class=""></o:p></p>
                                                      </div>
                                                    </div>
                                                  </div>
                                                </div>
                                              </div>
                                            </div>
                                          </div>
                                        </div>
                                      </div>
                                    </div>
                                  </div>
                                </div>
                              </div>
                            </div>
                            <p class="MsoNormal"><o:p class=""> </o:p></p>
                          </div>
                        </div>
                        <p class="MsoNormal"><o:p class=""> </o:p></p>
                        <div class="">
                          <div class="">
                            <p class="MsoNormal">On Fri, Feb 4, 2022 at
                              2:42 AM Asim Roy <<a
                                href="mailto:ASIM.ROY@asu.edu"
                                moz-do-not-send="true" class="">ASIM.ROY@asu.edu</a>>
                              wrote:<o:p class=""></o:p></p>
                          </div>
                          <blockquote
                            style="border:none;border-left:solid #CCCCCC
                            1.0pt;padding:0in 0in 0in
                            6.0pt;margin-left:4.8pt;margin-right:0in"
                            class="">
                            <div class="">
                              <div class="">
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">First
                                  of all, the brain is a physical
                                  system. There is no “magic” inside the
                                  brain that does the “understanding”
                                  part. Take for example learning to
                                  play tennis. You hit a few balls -
                                  some the right way and some wrong –
                                  but you fairly quickly learn to hit
                                  them right most of the time. So there
                                  is obviously some simulation going on
                                  in the brain about hitting the ball in
                                  different ways and “learning” its
                                  consequences. What you are calling
                                  “understanding” is really these
                                  simulations about different scenarios.
                                  It’s also very similar to augmentation
                                  used to train image recognition
                                  systems where you rotate images,
                                  obscure parts and so on, so that you
                                  still can say it’s a cat even though
                                  you see only the cat’s face or
                                  whiskers or a cat flipped on its back.
                                  So, if the following questions relate
                                  to “understanding,” you can easily
                                  resolve this by simulating such
                                  scenarios when “teaching” the system.
                                  There’s nothing “magical” about
                                  “understanding.” As I said, bear in
                                  mind that the brain, after all, is a
                                  physical system and “teaching” and
                                  “understanding” is embodied in that
                                  physical system, not outside it. So
                                  “understanding” is just part of
                                  “learning,” nothing more.<o:p class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                    style="background-color: yellow;"
                                    class="">DANKO:</span><o:p class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                    style="background-color: yellow;"
                                    class="">What would happen to the
                                    hat if the hamster rolls on its
                                    back? (Would the hat fall off?)</span><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                    style="background-color: yellow;"
                                    class="">What would happen to the
                                    red hat when the hamster enters its
                                    lair? (Would the hat fall off?)</span><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                    style="background-color: yellow;"
                                    class="">What would happen to that
                                    hamster when it goes foraging?
                                    (Would the red hat have an influence
                                    on finding food?)</span><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                    style="background-color: yellow;"
                                    class="">What would happen in a
                                    situation of being chased by a
                                    predator? (Would it be easier for
                                    predators to spot the hamster?)</span><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Asim
                                  Roy<o:p class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Professor,
                                  Information Systems<o:p class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Arizona
                                  State University<o:p class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__lifeboat.com_ex_bios.asim.roy&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=oDRJmXX22O8NcfqyLjyu4Ajmt8pcHWquTxYjeWahfuw&e="
                                    target="_blank"
                                    moz-do-not-send="true" class="">Lifeboat
                                    Foundation Bios: Professor Asim Roy</a><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__isearch.asu.edu_profile_9973&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=jCesWT7oGgX76_y7PFh4cCIQ-Ife-esGblJyrBiDlro&e="
                                    target="_blank"
                                    moz-do-not-send="true" class="">Asim
                                    Roy | iSearch (asu.edu)</a><o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                    class=""></o:p></p>
                                <div class="">
                                  <div
                                    style="border:none;border-top:solid
                                    windowtext 1.0pt;padding:3.0pt 0in
                                    0in 0in;border-color:currentcolor
                                    currentcolor" class="">
                                    <p class="MsoNormal"
                                      style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><b
                                        class="">From:</b> Gary Marcus
                                      <<a
                                        href="mailto:gary.marcus@nyu.edu"
                                        target="_blank"
                                        moz-do-not-send="true" class="">gary.marcus@nyu.edu</a>>
                                      <br class="">
                                      <b class="">Sent:</b> Thursday,
                                      February 3, 2022 9:26 AM<br
                                        class="">
                                      <b class="">To:</b> Danko Nikolic
                                      <<a
                                        href="mailto:danko.nikolic@gmail.com"
                                        target="_blank"
                                        moz-do-not-send="true" class="">danko.nikolic@gmail.com</a>><br
                                        class="">
                                      <b class="">Cc:</b> Asim Roy <<a
                                        href="mailto:ASIM.ROY@asu.edu"
                                        target="_blank"
                                        moz-do-not-send="true" class="">ASIM.ROY@asu.edu</a>>;
                                      Geoffrey Hinton <<a
                                        href="mailto:geoffrey.hinton@gmail.com"
                                        target="_blank"
                                        moz-do-not-send="true" class="">geoffrey.hinton@gmail.com</a>>;
                                      AIhub <<a
                                        href="mailto:aihuborg@gmail.com"
                                        target="_blank"
                                        moz-do-not-send="true" class="">aihuborg@gmail.com</a>>;
                                      <a
                                        href="mailto:connectionists@mailman.srv.cs.cmu.edu"
                                        target="_blank"
                                        moz-do-not-send="true" class="">connectionists@mailman.srv.cs.cmu.edu</a><br
                                        class="">
                                      <b class="">Subject:</b> Re:
                                      Connectionists: Stephen Hanson in
                                      conversation with Geoff Hinton<o:p
                                        class=""></o:p></p>
                                  </div>
                                </div>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                    class=""></o:p></p>
                                <p class="MsoNormal"
                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Dear
                                  Danko,<o:p class=""></o:p></p>
                                <div class="">
                                  <p class="MsoNormal"
                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                      class=""></o:p></p>
                                </div>
                                <div class="">
                                  <p class="MsoNormal"
                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Well
                                    said. I had a somewhat similar
                                    response to Jeff Dean’s 2021 TED
                                    talk, in which he said (paraphrasing
                                    from memory, because I don’t
                                    remember the precise words) that the
                                    famous 200 Quoc Le unsupervised
                                    model [<a
href="https://urldefense.com/v3/__https:/static.googleusercontent.com/media/research.google.com/en/*archive/unsupervised_icml2012.pdf__;Lw!!IKRxdwAv5BmarQ!PFl2URDWVshfy1BPSwAMXKYyn1wszxpN4EPzShAm3sX83AOt05MQX07oVyVLEqo$"
                                      target="_blank"
                                      moz-do-not-send="true" class="">https://static.googleusercontent.com/media/research.google.com/en//archive/unsupervised_icml2012.pdf</a>]
                                    had learned the concept of a ca. In
                                    reality the model had clustered
                                    together some catlike images based
                                    on the image statistics that it had
                                    extracted, but it was a long way
                                    from a full,
                                    counterfactual-supporting concept of
                                    a cat, much as you describe below. <o:p
                                      class=""></o:p></p>
                                </div>
                                <div class="">
                                  <p class="MsoNormal"
                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                      class=""></o:p></p>
                                </div>
                                <div class="">
                                  <p class="MsoNormal"
                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">I
                                    fully agree with you that the reason
                                    for even having a semantics is as
                                    you put it, "to 1) learn with a few
                                    examples and 2) apply the knowledge
                                    to a broad set of situations.” GPT-3
                                    sometimes gives the appearance of
                                    having done so, but it falls apart
                                    under close inspection, so the
                                    problem remains unsolved.<o:p
                                      class=""></o:p></p>
                                </div>
                                <div class="">
                                  <div class="">
                                    <p class="MsoNormal"
                                      style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                        class=""></o:p></p>
                                  </div>
                                  <div class="">
                                    <p class="MsoNormal"
                                      style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Gary<o:p
                                        class=""></o:p></p>
                                    <div class="">
                                      <p class="MsoNormal"
                                        style="mso-margin-top-alt:auto;margin-bottom:12.0pt"><o:p
                                          class=""> </o:p></p>
                                      <blockquote
                                        style="margin-top:5.0pt;margin-bottom:5.0pt"
                                        class="">
                                        <div class="">
                                          <p class="MsoNormal"
                                            style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">On
                                            Feb 3, 2022, at 3:19 AM,
                                            Danko Nikolic <<a
                                              href="mailto:danko.nikolic@gmail.com"
                                              target="_blank"
                                              moz-do-not-send="true"
                                              class="">danko.nikolic@gmail.com</a>>
                                            wrote:<o:p class=""></o:p></p>
                                        </div>
                                        <p class="MsoNormal"
                                          style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                            class=""></o:p></p>
                                        <div class="">
                                          <div class="">
                                            <p class="MsoNormal"
                                              style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">G.
                                              Hinton wrote: "I believe
                                              that any reasonable person
                                              would admit that if you
                                              ask a neural net to draw a
                                              picture of a hamster
                                              wearing a red hat and it
                                              draws such a picture, it
                                              understood the request."<o:p
                                                class=""></o:p></p>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">I
                                                would like to suggest
                                                why drawing a hamster
                                                with a red hat does not
                                                necessarily imply
                                                understanding of the
                                                statement "hamster
                                                wearing a red hat".<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">To
                                                understand that "hamster
                                                wearing a red hat" would
                                                mean inferring, in
                                                newly emerging
                                                situations of this
                                                hamster, all the
                                                real-life
                                                implications that the
                                                red hat brings to the
                                                little animal.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <div class="">
                                                <p class="MsoNormal"
                                                  style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">What
                                                  would happen to the
                                                  hat if the hamster
                                                  rolls on its back?
                                                  (Would the hat fall
                                                  off?)<o:p class=""></o:p></p>
                                              </div>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">What
                                                would happen to the red
                                                hat when the hamster
                                                enters its lair? (Would
                                                the hat fall off?)<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">What
                                                would happen to that
                                                hamster when it goes
                                                foraging? (Would the red
                                                hat have an influence on
                                                finding food?)<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">What
                                                would happen in a
                                                situation of being
                                                chased by a predator?
                                                (Would it be easier for
                                                predators to spot the
                                                hamster?)<o:p class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">...and
                                                so on.<o:p class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Countless
                                                many questions can be
                                                asked. One has
                                                understood "hamster
                                                wearing a red hat" only
                                                if one can answer
                                                reasonably well many of
                                                such real-life relevant
                                                questions. Similarly, a
                                                student has understood
                                                materias in a class only
                                                if they can apply the
                                                materials in real-life
                                                situations (e.g.,
                                                applying Pythagora's
                                                theorem). If a student
                                                gives a correct answer
                                                to a multiple choice
                                                question, we don't know
                                                whether the student
                                                understood the material
                                                or whether this was just
                                                rote learning (often, it
                                                is rote learning). <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">I
                                                also suggest that
                                                understanding also comes
                                                together with effective
                                                learning: We store new
                                                information in such a
                                                way that we can recall
                                                it later and use it
                                                effectively  i.e., make
                                                good inferences in newly
                                                emerging situations
                                                based on this knowledge.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">In
                                                short: Understanding
                                                makes us humans able to
                                                1) learn with a few
                                                examples and 2) apply
                                                the knowledge to a broad
                                                set of situations. <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">No
                                                neural network today has
                                                such capabilities and we
                                                don't know how to give
                                                them such capabilities.
                                                Neural networks need
                                                large amounts of
                                                training examples that
                                                cover a large variety of
                                                situations and then
                                                the networks can only
                                                deal with what the
                                                training examples have
                                                already covered. Neural
                                                networks cannot
                                                extrapolate in that
                                                'understanding' sense.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">I
                                                suggest that
                                                understanding truly
                                                extrapolates from a
                                                piece of knowledge. It
                                                is not about satisfying
                                                a task such as
                                                translation between
                                                languages or drawing
                                                hamsters with hats. It
                                                is how you got the
                                                capability to complete
                                                the task: Did you only
                                                have a few examples that
                                                covered something
                                                different but related
                                                and then you
                                                extrapolated from that
                                                knowledge? If yes, this
                                                is going in the
                                                direction of
                                                understanding. Have you
                                                seen countless examples
                                                and then interpolated
                                                among them? Then perhaps
                                                it is not understanding.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">So,
                                                for the case of drawing
                                                a hamster wearing a red
                                                hat, understanding
                                                perhaps would have taken
                                                place if the following
                                                happened before that:<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">1)
                                                first, the network
                                                learned about hamsters
                                                (not many examples)<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">2)
                                                after that the network
                                                learned about red hats
                                                (outside the context of
                                                hamsters and without
                                                many examples) <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">3)
                                                finally the network
                                                learned about drawing
                                                (outside of the context
                                                of hats and hamsters,
                                                not many examples)<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">After
                                                that, the network is
                                                asked to draw a hamster
                                                with a red hat. If it
                                                does it successfully,
                                                maybe we have started
                                                cracking the problem of
                                                understanding.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Note
                                                also that this requires
                                                the network to learn
                                                sequentially without
                                                exhibiting catastrophic
                                                forgetting of the
                                                previous knowledge,
                                                which is possibly also a
                                                consequence of human
                                                learning by
                                                understanding.<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Danko<o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                            <div class="">
                                              <div class="">
                                                <div class="">
                                                  <div class="">
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Dr. Danko
                                                      Nikolić<br
                                                        class="">
                                                      <a
href="https://urldefense.proofpoint.com/v2/url?u=http-3A__www.danko-2Dnikolic.com&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=HwOLDw6UCRzU5-FPSceKjtpNm7C6sZQU5kuGAMVbPaI&e="
                                                        target="_blank"
moz-do-not-send="true" class="">www.danko-nikolic.com</a><br class="">
                                                      <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.linkedin.com_in_danko-2Dnikolic_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=b70c8lokmxM3Kz66OfMIM4pROgAhTJOAlp205vOmCQ8&e="
                                                        target="_blank"
moz-do-not-send="true" class="">https://www.linkedin.com/in/danko-nikolic/</a><o:p
                                                        class=""></o:p></p>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">--- A
                                                        progress usually
                                                        starts with an
                                                        insight ---<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                  </div>
                                                </div>
                                              </div>
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                  class=""></o:p></p>
                                            </div>
                                          </div>
                                          <div
                                            id="gmail-m_3776411903040420401DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2"
                                            class="">
                                            <p class="MsoNormal"
                                              style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                class=""></o:p></p>
                                            <table
                                              class="MsoNormalTable"
                                              style="border:none;border-top:solid
                                              windowtext
                                              1.0pt;border-color:currentcolor
                                              currentcolor"
                                              cellpadding="0" border="1">
                                              <tbody class="">
                                                <tr class="">
                                                  <td
                                                    style="width:41.25pt;border:none;padding:9.75pt
                                                    .75pt .75pt .75pt"
                                                    class="" width="55">
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.avast.com_sig-2Demail-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dlink-26utm-5Fcampaign-3Dsig-2Demail-26utm-5Fcontent-3Dwebmail&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=Ao9QQWtO62go0hx1tb3NU6xw2FNBadjj8q64-hl5Sx4&e="
                                                        target="_blank"
moz-do-not-send="true" class=""><span style="text-decoration:none"
                                                          class=""><img
style="width:.4791in;height:.302in" id="_x0000_i1026"
src="https://ipmcdn.avast.com/images/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif"
data-unique-identifier="" moz-do-not-send="true" class="" width="46"
                                                          height="29"
                                                          border="0"></span></a><o:p
                                                        class=""></o:p></p>
                                                  </td>
                                                  <td
                                                    style="width:352.5pt;border:none;padding:9.0pt
                                                    .75pt .75pt .75pt"
                                                    class="" width="470">
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;line-height:13.5pt"><span
style="font-size:10.0pt;font-family:"Arial",sans-serif;color:#41424E"
                                                        class="">Virus-free.
                                                        <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.avast.com_sig-2Demail-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dlink-26utm-5Fcampaign-3Dsig-2Demail-26utm-5Fcontent-3Dwebmail&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=Ao9QQWtO62go0hx1tb3NU6xw2FNBadjj8q64-hl5Sx4&e="
target="_blank" moz-do-not-send="true" class=""> <span
                                                          style="color:#4453EA"
                                                          class="">www.avast.com</span></a>
                                                      </span><o:p
                                                        class=""></o:p></p>
                                                  </td>
                                                </tr>
                                              </tbody>
                                            </table>
                                          </div>
                                          <p class="MsoNormal"
                                            style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                              class=""></o:p></p>
                                          <div class="">
                                            <div class="">
                                              <p class="MsoNormal"
                                                style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">On
                                                Thu, Feb 3, 2022 at 9:55
                                                AM Asim Roy <<a
                                                  href="mailto:ASIM.ROY@asu.edu"
                                                  target="_blank"
                                                  moz-do-not-send="true"
                                                  class="">ASIM.ROY@asu.edu</a>>
                                                wrote:<o:p class=""></o:p></p>
                                            </div>
                                            <blockquote
                                              style="border:none;border-left:solid
                                              windowtext
                                              1.0pt;padding:0in 0in 0in
6.0pt;margin-left:4.8pt;margin-top:5.0pt;margin-right:0in;margin-bottom:5.0pt;border-color:currentcolor
                                              currentcolor currentcolor
                                              rgb(204,204,204)" class="">
                                              <div class="">
                                                <div class="">
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Without
                                                    getting into the
                                                    specific dispute
                                                    between Gary and
                                                    Geoff, I think with
                                                    approaches similar
                                                    to GLOM, we are
                                                    finally headed in
                                                    the right direction.
                                                    There’s plenty of
                                                    neurophysiological
                                                    evidence for
                                                    single-cell
                                                    abstractions and
                                                    multisensory neurons
                                                    in the brain, which
                                                    one might claim
                                                    correspond to
                                                    symbols. And I think
                                                    we can finally
                                                    reconcile the
                                                    decades old dispute
                                                    between Symbolic AI
                                                    and Connectionism.<o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
style="background-color: yellow;" class="">GARY: (Your GLOM, which as
                                                      you know I praised
                                                      publicly, is in
                                                      many ways an
                                                      effort to wind up
                                                      with encodings
                                                      that effectively
                                                      serve as symbols
                                                      in exactly that
                                                      way, guaranteed to
                                                      serve as
                                                      consistent
                                                      representations of
                                                      specific
                                                      concepts.)</span><o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
style="background-color: yellow;" class="">GARY: I have <i class="">never</i>
                                                      called for
                                                      dismissal of
                                                      neural networks,
                                                      but rather for
                                                      some hybrid
                                                      between the two
                                                      (as you yourself
                                                      contemplated in
                                                      1991); the point
                                                      of the 2001 book
                                                      was to
                                                      characterize
                                                      exactly where
                                                      multilayer
                                                      perceptrons
                                                      succeeded and
                                                      broke down, and
                                                      where symbols
                                                      could complement
                                                      them.</span><o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Asim
                                                    Roy<o:p class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Professor,
                                                    Information Systems<o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Arizona
                                                    State University<o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__lifeboat.com_ex_bios.asim.roy&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=oDRJmXX22O8NcfqyLjyu4Ajmt8pcHWquTxYjeWahfuw&e="
                                                      target="_blank"
                                                      moz-do-not-send="true"
                                                      class="">Lifeboat
                                                      Foundation Bios:
                                                      Professor Asim Roy</a><o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__isearch.asu.edu_profile_9973&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=jCesWT7oGgX76_y7PFh4cCIQ-Ife-esGblJyrBiDlro&e="
                                                      target="_blank"
                                                      moz-do-not-send="true"
                                                      class="">Asim Roy
                                                      | iSearch
                                                      (asu.edu)</a><o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                      class=""></o:p></p>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                      class=""></o:p></p>
                                                  <div class="">
                                                    <div
                                                      style="border:none;border-top:solid
                                                      windowtext
                                                      1.0pt;padding:3.0pt
                                                      0in 0in
                                                      0in;border-color:currentcolor
                                                      currentcolor"
                                                      class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><b class="">From:</b>
                                                        Connectionists
                                                        <<a
                                                          href="mailto:connectionists-bounces@mailman.srv.cs.cmu.edu"
target="_blank" moz-do-not-send="true" class="">connectionists-bounces@mailman.srv.cs.cmu.edu</a>>
                                                        <b class="">On
                                                          Behalf Of </b>Gary
                                                        Marcus<br
                                                          class="">
                                                        <b class="">Sent:</b>
                                                        Wednesday,
                                                        February 2, 2022
                                                        1:26 PM<br
                                                          class="">
                                                        <b class="">To:</b>
                                                        Geoffrey Hinton
                                                        <<a
                                                          href="mailto:geoffrey.hinton@gmail.com"
target="_blank" moz-do-not-send="true" class="">geoffrey.hinton@gmail.com</a>><br
                                                          class="">
                                                        <b class="">Cc:</b>
                                                        AIhub <<a
                                                          href="mailto:aihuborg@gmail.com"
target="_blank" moz-do-not-send="true" class="">aihuborg@gmail.com</a>>;
                                                        <a
                                                          href="mailto:connectionists@mailman.srv.cs.cmu.edu"
target="_blank" moz-do-not-send="true" class="">connectionists@mailman.srv.cs.cmu.edu</a><br
                                                          class="">
                                                        <b class="">Subject:</b>
                                                        Re:
                                                        Connectionists:
                                                        Stephen Hanson
                                                        in conversation
                                                        with Geoff
                                                        Hinton<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                  </div>
                                                  <p class="MsoNormal"
                                                    style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                      class=""></o:p></p>
                                                  <div class="">
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Dear Geoff,
                                                        and interested
                                                        others,<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">What, for
                                                        example, would
                                                        you make of a
                                                        system that
                                                        often drew the
                                                        red-hatted
                                                        hamster you
                                                        requested, and
                                                        perhaps a fifth
                                                        of the time gave
                                                        you utter
                                                        nonsense?  Or
                                                        say one that you
                                                        trained to
                                                        create birds but
                                                        sometimes output
                                                        stuff like this:<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><image001.png><o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">One could <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">a. avert
                                                        one’s eyes and
                                                        deem the
                                                        anomalous
                                                        outputs
                                                        irrelevant<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">or<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">b. wonder if
                                                        it might be
                                                        possible that
                                                        sometimes the
                                                        system gets the
                                                        right answer for
                                                        the wrong
                                                        reasons (eg
                                                        partial
                                                        historical
                                                        contingency),
                                                        and wonder
                                                        whether another
                                                        approach might
                                                        be indicated.<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Benchmarks
                                                        are harder than
                                                        they look; most
                                                        of the field has
                                                        come to
                                                        recognize that.
                                                        The Turing Test
                                                        has turned out
                                                        to be a lousy
                                                        measure of
                                                        intelligence,
                                                        easily gamed. It
                                                        has turned out
                                                        empirically that
                                                        the Winograd
                                                        Schema Challenge
                                                        did not measure
                                                        common sense as
                                                        well as Hector
                                                        might have
                                                        thought. (As it
                                                        happens, I am a
                                                        minor coauthor
                                                        of a very recent
                                                        review on this
                                                        very topic: <a
href="https://urldefense.com/v3/__https:/arxiv.org/abs/2201.02387__;!!IKRxdwAv5BmarQ!INA0AMmG3iD1B8MDtLfjWCwcBjxO-e-eM2Ci9KEO_XYOiIEgiywK-G_8j6L3bHA$"
target="_blank" moz-do-not-send="true" class="">https://arxiv.org/abs/2201.02387</a>)
                                                        But its conquest
                                                        in no way means
                                                        machines now
                                                        have common
                                                        sense; many
                                                        people from many
                                                        different
                                                        perspectives
                                                        recognize that
                                                        (including,
                                                        e.g., Yann
                                                        LeCun, who
                                                        generally tends
                                                        to be more
                                                        aligned with you
                                                        than with me).<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">So: on the
                                                        goalpost of the
                                                        Winograd schema,
                                                        I was wrong, and
                                                        you can quote
                                                        me; but what you
                                                        said about me
                                                        and machine
                                                        translation
                                                        remains your
                                                        invention, and
                                                        it is
                                                        inexcusable that
                                                        you simply
                                                        ignored my 2019
                                                        clarification.
                                                        On the essential
                                                        goal of trying
                                                        to reach meaning
                                                        and
                                                        understanding, I
                                                        remain unmoved;
                                                        the problem
                                                        remains
                                                        unsolved. <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">All of the
                                                        problems LLMs
                                                        have with
                                                        coherence,
                                                        reliability,
                                                        truthfulness,
                                                        misinformation,
                                                        etc stand
                                                        witness to that
                                                        fact. (Their
                                                        persistent
                                                        inability to
                                                        filter out toxic
                                                        and insulting
                                                        remarks stems
                                                        from the same.)
                                                        I am hardly the
                                                        only person in
                                                        the field to see
                                                        that progress on
                                                        any given
                                                        benchmark does
                                                        not inherently
                                                        mean that the
                                                        deep underlying
                                                        problems have
                                                        solved. You,
                                                        yourself, in
                                                        fact, have
                                                        occasionally
                                                        made that
                                                        point. <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">With respect
                                                        to embeddings:
                                                        Embeddings are
                                                        very good for
                                                        natural language
                                                        <i class="">processing</i>;
                                                        but NLP is not
                                                        the same as NL<i
                                                          class="">U</i>
                                                        – when it comes
                                                        to <i class="">understanding</i>,
                                                        their worth is
                                                        still an open
                                                        question.
                                                        Perhaps they
                                                        will turn out to
                                                        be necessary;
                                                        they clearly
                                                        aren’t
                                                        sufficient. In
                                                        their extreme,
                                                        they might even
                                                        collapse into
                                                        being symbols,
                                                        in the sense of
                                                        uniquely
                                                        identifiable
                                                        encodings, akin
                                                        to the ASCII
                                                        code, in which a
                                                        specific set of
                                                        numbers stands
                                                        for a specific
                                                        word or concept.
                                                        (Wouldn’t that
                                                        be ironic?)<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">(Your GLOM,
                                                        which as you
                                                        know I praised
                                                        publicly, is in
                                                        many ways an
                                                        effort to wind
                                                        up with
                                                        encodings that
                                                        effectively
                                                        serve as symbols
                                                        in exactly that
                                                        way, guaranteed
                                                        to serve as
                                                        consistent
                                                        representations
                                                        of specific
                                                        concepts.)<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Notably
                                                        absent from your
                                                        email is any
                                                        kind of apology
                                                        for
                                                        misrepresenting
                                                        my position.
                                                        It’s fine to say
                                                        that “many
                                                        people thirty
                                                        years ago once
                                                        thought X” and
                                                        another to say
                                                        “Gary Marcus
                                                        said X in 2015”,
                                                        when I didn’t. I
                                                        have
                                                        consistently
                                                        felt throughout
                                                        our interactions
                                                        that you have
                                                        mistaken me for
                                                        Zenon Pylyshyn;
                                                        indeed, you once
                                                        (at NeurIPS
                                                        2014) apologized
                                                        to me for having
                                                        made that error.
                                                        I am still not
                                                        he. <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Which maybe
                                                        connects to the
                                                        last point; if
                                                        you read my
                                                        work, you would
                                                        see thirty years
                                                        of arguments <i
                                                          class="">for</i> neural
                                                        networks, just
                                                        not in the way
                                                        that you want
                                                        them to exist. I
                                                        have ALWAYS
                                                        argued that
                                                        there is a role
                                                        for them;
                                                         characterizing
                                                        me as a person
                                                        “strongly opposed
                                                        to neural
                                                        networks” misses
                                                        the whole point
                                                        of my 2001 book,
                                                        which was
                                                        subtitled
                                                        “Integrating
                                                        Connectionism
                                                        and Cognitive
                                                        Science.”<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">In the last
                                                        two decades or
                                                        so you have
                                                        insisted (for
                                                        reasons you have
                                                        never fully
                                                        clarified, so
                                                        far as I know)
                                                        on abandoning
                                                        symbol-manipulation,
                                                        but the reverse
                                                        is not the case:
                                                        I have <i
                                                          class="">never</i>
                                                        called for
                                                        dismissal of
                                                        neural networks,
                                                        but rather for
                                                        some hybrid
                                                        between the two
                                                        (as you yourself
                                                        contemplated in
                                                        1991); the point
                                                        of the 2001 book
                                                        was to
                                                        characterize
                                                        exactly where
                                                        multilayer
                                                        perceptrons
                                                        succeeded and
                                                        broke down, and
                                                        where symbols
                                                        could complement
                                                        them. It’s a
                                                        rhetorical trick
                                                        (which is what
                                                        the previous
                                                        thread was
                                                        about) to
                                                        pretend
                                                        otherwise.<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Gary<o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <p
                                                        class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                    </div>
                                                    <div class="">
                                                      <blockquote
                                                        style="margin-top:5.0pt;margin-bottom:5.0pt"
                                                        class="">
                                                        <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;margin-bottom:12.0pt">On Feb 2, 2022, at
                                                          11:22,
                                                          Geoffrey
                                                          Hinton <<a
href="mailto:geoffrey.hinton@gmail.com" target="_blank"
                                                          moz-do-not-send="true"
                                                          class="">geoffrey.hinton@gmail.com</a>>
                                                          wrote:<o:p
                                                          class=""></o:p></p>
                                                      </blockquote>
                                                    </div>
                                                    <blockquote
                                                      style="margin-top:5.0pt;margin-bottom:5.0pt"
                                                      class="">
                                                      <div class="">
                                                        <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><o:p
                                                          class=""></o:p></p>
                                                        <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Embeddings
                                                          are just
                                                          vectors of
                                                          soft feature
                                                          detectors and
                                                          they are very
                                                          good for NLP. 
                                                          The quote on
                                                          my webpage
                                                          from Gary's
                                                          2015 chapter
                                                          implies the
                                                          opposite.<o:p
                                                          class=""></o:p></p>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">A few decades
                                                          ago, everyone
                                                          I knew then
                                                          would have
                                                          agreed that
                                                          the ability to
                                                          translate a
                                                          sentence into
                                                          many different
                                                          languages was
                                                          strong
                                                          evidence that
                                                          you understood
                                                          it.<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                        </div>
                                                      </div>
                                                    </blockquote>
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;margin-bottom:12.0pt"> <o:p class=""></o:p></p>
                                                    <blockquote
                                                      style="margin-top:5.0pt;margin-bottom:5.0pt"
                                                      class="">
                                                      <div class="">
                                                        <div class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">But once
                                                          neural
                                                          networks could
                                                          do that, their
                                                          critics moved
                                                          the goalposts.
                                                          An exception
                                                          is Hector
                                                          Levesque who
                                                          defined the
                                                          goalposts more
                                                          sharply by
                                                          saying that
                                                          the ability to
                                                          get pronoun
                                                          references
                                                          correct in
                                                          Winograd
                                                          sentences is a
                                                          crucial test.
                                                          Neural nets
                                                          are improving
                                                          at that but
                                                          still have
                                                          some way to
                                                          go. Will Gary
                                                          agree that
                                                          when they can
                                                          get pronoun
                                                          references correct
                                                          in Winograd
                                                          sentences they
                                                          really do
                                                          understand? Or
                                                          does he want
                                                          to reserve the
                                                          right to
                                                          weasel out of
                                                          that too?<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Some people,
                                                          like Gary,
                                                          appear to be
                                                          strongly opposed
                                                          to neural
                                                          networks
                                                          because they
                                                          do not fit
                                                          their
                                                          preconceived
                                                          notions of how
                                                          the mind
                                                          should work.<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">I believe
                                                          that any
                                                          reasonable
                                                          person would
                                                          admit that if
                                                          you ask a
                                                          neural net to
                                                          draw a picture
                                                          of a hamster
                                                          wearing a red
                                                          hat and it
                                                          draws such a
                                                          picture, it
                                                          understood the
                                                          request.<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Geoff<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          </div>
                                                          </div>
                                                        </div>
                                                        <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                        <div class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">On Wed, Feb
                                                          2, 2022 at
                                                          1:38 PM Gary
                                                          Marcus <<a
href="mailto:gary.marcus@nyu.edu" target="_blank" moz-do-not-send="true"
                                                          class="">gary.marcus@nyu.edu</a>>
                                                          wrote:<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <blockquote
                                                          style="border:none;border-left:solid
                                                          windowtext
                                                          1.0pt;padding:0in
                                                          0in 0in
6.0pt;margin-left:4.8pt;margin-top:5.0pt;margin-right:0in;margin-bottom:5.0pt;border-color:currentcolor
                                                          currentcolor
                                                          currentcolor
                                                          rgb(204,204,204)"
                                                          class="">
                                                          <div class="">
                                                          <div class="">
                                                          <div class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Dear
                                                          AI Hub, cc:
                                                          Steven Hanson
                                                          and Geoffrey
                                                          Hinton, and
                                                          the larger
                                                          neural network
                                                          community,</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">There
                                                          has been a lot
                                                          of recent
                                                          discussion on
                                                          this list
                                                          about framing
                                                          and scientific
                                                          integrity.
                                                          Often the
                                                          first step in
                                                          restructuring
                                                          narratives is
                                                          to bully and
                                                          dehumanize
                                                          critics. The
                                                          second is to
                                                          misrepresent
                                                          their
                                                          position.
                                                          People in
                                                          positions of
                                                          power are
                                                          sometimes
                                                          tempted to do
                                                          this.</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">The
                                                          Hinton-Hanson
                                                          interview that
                                                          you just
                                                          published is a
                                                          real-time
                                                          example of
                                                          just that. It
                                                          opens with a
                                                          needless and
                                                          largely
                                                          content-free
                                                          personal
                                                          attack on a
                                                          single scholar
                                                          (me), with the
                                                          explicit
                                                          intention of
                                                          discrediting
                                                          that person.
                                                          Worse, the
                                                          only
                                                          substantive
                                                          thing it says
                                                          is false.</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Hinton
                                                          says “In 2015
                                                          he [Marcus]
                                                          made a
                                                          prediction
                                                          that computers
                                                          wouldn’t be
                                                          able to do
                                                          machine
                                                          translation.”</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">I
                                                          never said any
                                                          such thing. </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">What
                                                          I predicted,
                                                          rather, was
                                                          that
                                                          multilayer
                                                          perceptrons,
                                                          as they
                                                          existed then,
                                                          would not (on
                                                          their own,
                                                          absent other
                                                          mechanisms) <i
                                                          class="">understand</i> language.
                                                          Seven years
                                                          later, they
                                                          still haven’t,
                                                          except in the
                                                          most
                                                          superficial
                                                          way.   </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">I
                                                          made no
                                                          comment
                                                          whatsoever
                                                          about machine
                                                          translation,
                                                          which I view
                                                          as a separate
                                                          problem,
                                                          solvable to a
                                                          certain degree
                                                          by
                                                          correspondance
                                                          without
                                                          semantics. </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">I
                                                          specifically
                                                          tried to
                                                          clarify
                                                          Hinton’s
                                                          confusion in
                                                          2019, but,
                                                          disappointingly,
                                                          he has
                                                          continued to
                                                          purvey
                                                          misinformation
                                                          despite that
                                                          clarification.
                                                          Here is what I
                                                          wrote
                                                          privately to
                                                          him then,
                                                          which should
                                                          have put the
                                                          matter to
                                                          rest:</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">You
                                                          have taken a
                                                          single out of
                                                          context quote
                                                          [from 2015]
                                                          and
                                                          misrepresented
                                                          it. The quote,
                                                          which you have
                                                          prominently
                                                          displayed at
                                                          the bottom on
                                                          your own web
                                                          page, says:</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal;min-height:22.9px"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:.75in;font-stretch:normal"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Hierarchies
                                                          of features
                                                          are less
                                                          suited to
                                                          challenges
                                                          such as
                                                          language,
                                                          inference, and
                                                          high-level
                                                          planning. For
                                                          example, as
                                                          Noam Chomsky
                                                          famously
                                                          pointed out,
                                                          language is
                                                          filled with
                                                          sentences you
                                                          haven't seen
                                                          before. Pure
                                                          classifier
                                                          systems don't
                                                          know what to
                                                          do with such
                                                          sentences. The
                                                          talent of
                                                          feature
                                                          detectors --
                                                          in  identifying
                                                          which member
                                                          of some
                                                          category
                                                          something
                                                          belongs to --
                                                          doesn't
                                                          translate into
                                                          understanding
novel  sentences, in which each sentence has its own unique meaning. </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal;min-height:22.9px"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">It
                                                          does <i
                                                          class="">not</i> say
                                                          "neural nets
                                                          would not be
                                                          able to deal
                                                          with novel
                                                          sentences"; it
                                                          says that
                                                          hierachies of
                                                          features
                                                          detectors (on
                                                          their own, if
                                                          you read the
                                                          context of the
                                                          essay) would
                                                          have trouble <i
                                                          class="">understanding </i>novel sentences.
                                                           </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal;min-height:22.9px"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Google
                                                          Translate does
                                                          yet not <i
                                                          class="">understand</i> the
                                                          content of the
                                                          sentences is
                                                          translates. It
                                                          cannot
                                                          reliably
                                                          answer
                                                          questions
                                                          about who did
                                                          what to whom,
                                                          or why, it
                                                          cannot infer
                                                          the order of
                                                          the events in
                                                          paragraphs, it
                                                          can't
                                                          determine the
                                                          internal
                                                          consistency of
                                                          those events,
                                                          and so forth.</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Since
                                                          then, a number
                                                          of scholars,
                                                          such as the
                                                          the
                                                          computational
                                                          linguist Emily
                                                          Bender, have
                                                          made similar
                                                          points, and
                                                          indeed current
                                                          LLM
                                                          difficulties
                                                          with
                                                          misinformation,
                                                          incoherence
                                                          and
                                                          fabrication
                                                          all follow
                                                          from these
                                                          concerns.
                                                          Quoting from
                                                          Bender’s
                                                          prizewinning
                                                          2020 ACL
                                                          article on the
                                                          matter with
                                                          Alexander
                                                          Koller, <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__aclanthology.org_2020.acl-2Dmain.463.pdf&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=xnFSVUARkfmiXtiTP_uXfFKv4uNEGgEeTluRFR7dnUpay2BM5EiLz-XYCkBNJLlL&s=K-Vl6vSvzuYtRMi-s4j7mzPkNRTb-I6Zmf7rbuKEBpk&e="
target="_blank" moz-do-not-send="true" class="">https://aclanthology.org/2020.acl-main.463.pdf</a>,
                                                          also
                                                          emphasizing
                                                          issues of
                                                          understanding
                                                          and meaning:</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div
                                                          style="margin-left:27.0pt;font-stretch:normal"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><i class=""><span
style="font-size:13.0pt;font-family:"Times New Roman",serif"
                                                          class="">The
                                                          success of the
                                                          large neural
                                                          language
                                                          models on many
                                                          NLP tasks is
                                                          exciting.
                                                          However, we
                                                          find that
                                                          these
                                                          successes
                                                          sometimes lead
                                                          to hype in
                                                          which these
                                                          models are
                                                          being
                                                          described as
                                                          “understanding”
                                                          language or
                                                          capturing
                                                          “meaning”. In
                                                          this position
                                                          paper, we
                                                          argue that a
                                                          system trained
                                                          only on form
                                                          has a priori
                                                          no way to
                                                          learn meaning.
                                                          .. a clear
                                                          understanding
                                                          of the
                                                          distinction
                                                          between form
                                                          and meaning
                                                          will help
                                                          guide the
                                                          field towards
                                                          better science
                                                          around natural
                                                          language
                                                          understanding. </span></i><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Her
                                                          later article
                                                          with Gebru on
                                                          language
                                                          models
                                                          “stochastic
                                                          parrots” is in
                                                          some ways an
                                                          extension of
                                                          this point;
                                                          machine
                                                          translation
                                                          requires
                                                          mimicry, true
                                                          understanding
                                                          (which is what
                                                          I was
                                                          discussing in
                                                          2015) requires
                                                          something
                                                          deeper than
                                                          that. </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Hinton’s
                                                          intellectual
                                                          error here is
                                                          in equating
                                                          machine
                                                          translation
                                                          with the
                                                          deeper
                                                          comprehension
                                                          that robust
                                                          natural
                                                          language
                                                          understanding
                                                          will require;
                                                          as Bender and
                                                          Koller
                                                          observed, the
                                                          two appear not
                                                          to be the
                                                          same. (There
                                                          is a longer
                                                          discussion of
                                                          the relation
                                                          between
                                                          language
                                                          understanding
                                                          and machine
                                                          translation,
                                                          and why the
                                                          latter has
                                                          turned out to
                                                          be more
                                                          approachable
                                                          than the
                                                          former, in my
                                                          2019 book with
                                                          Ernest Davis).</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">More
                                                          broadly,
                                                          Hinton’s
                                                          ongoing
                                                          dismissiveness
                                                          of research
                                                          from
                                                          perspectives
                                                          other than his
                                                          own (e.g.
                                                          linguistics)
                                                          have done the
                                                          field a
                                                          disservice. </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">As
                                                          Herb Simon
                                                          once observed,
                                                          science does
                                                          not have to be
                                                          zero-sum.</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt"
                                                          class=""> </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Sincerely,</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Gary
                                                          Marcus</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">Professor
                                                          Emeritus</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-size:13.0pt;font-family:"Times
                                                          New
                                                          Roman",serif"
                                                          class="">New
                                                          York
                                                          University</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;margin-bottom:12.0pt"> <o:p class=""></o:p></p>
                                                          <blockquote
                                                          style="margin-top:5.0pt;margin-bottom:5.0pt"
                                                          class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;margin-bottom:12.0pt">On Feb 2, 2022, at
                                                          06:12, AIhub
                                                          <<a
                                                          href="mailto:aihuborg@gmail.com"
target="_blank" moz-do-not-send="true" class="">aihuborg@gmail.com</a>>
                                                          wrote:<o:p
                                                          class=""></o:p></p>
                                                          </blockquote>
                                                          </div>
                                                          <blockquote
                                                          style="margin-top:5.0pt;margin-bottom:5.0pt"
                                                          class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><o:p
                                                          class=""></o:p></p>
                                                          <div class="">
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">Stephen
                                                          Hanson in
                                                          conversation
                                                          with Geoff
                                                          Hinton<o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">In the latest
                                                          episode of
                                                          this video
                                                          series for <a
href="https://urldefense.proofpoint.com/v2/url?u=http-3A__AIhub.org&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=xnFSVUARkfmiXtiTP_uXfFKv4uNEGgEeTluRFR7dnUpay2BM5EiLz-XYCkBNJLlL&s=eOtzMh8ILIH5EF7K20Ks4Fr27XfNV_F24bkj-SPk-2A&e="
target="_blank" moz-do-not-send="true" class=""> AIhub.org</a>, Stephen
                                                          Hanson talks
                                                          to  Geoff
                                                          Hinton about
                                                          neural
                                                          networks,
                                                          backpropagation,
overparameterization, digit recognition, voxel cells, syntax and
                                                          semantics,
                                                          Winograd
                                                          sentences, and
                                                          more.<o:p
                                                          class=""></o:p></p>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto">You can watch
                                                          the
                                                          discussion,
                                                          and read the
                                                          transcript,
                                                          here:<br
                                                          class=""
                                                          clear="all">
                                                          <o:p class=""></o:p></p>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__aihub.org_2022_02_02_what-2Dis-2Dai-2Dstephen-2Dhanson-2Din-2Dconversation-2Dwith-2Dgeoff-2Dhinton_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=OY_RYGrfxOqV7XeNJDHuzE--aEtmNRaEyQ0VJkqFCWw&e="
target="_blank" moz-do-not-send="true" class="">https://aihub.org/2022/02/02/what-is-ai-stephen-hanson-in-conversation-with-geoff-hinton/</a><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-family:"Arial",sans-serif"
                                                          class="">About
                                                          AIhub: </span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-family:"Arial",sans-serif"
                                                          class="">AIhub
                                                          is a
                                                          non-profit
                                                          dedicated to
                                                          connecting the
                                                          AI community
                                                          to the public
                                                          by providing
                                                          free,
                                                          high-quality
                                                          information
                                                          through <a
href="https://urldefense.proofpoint.com/v2/url?u=http-3A__AIhub.org&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=xnFSVUARkfmiXtiTP_uXfFKv4uNEGgEeTluRFR7dnUpay2BM5EiLz-XYCkBNJLlL&s=eOtzMh8ILIH5EF7K20Ks4Fr27XfNV_F24bkj-SPk-2A&e="
target="_blank" moz-do-not-send="true" class=""> AIhub.org</a> (<a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__aihub.org_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=IKFanqeMi73gOiS7yD-X_vRx_OqDAwv1Il5psrxnhIA&e="
target="_blank" moz-do-not-send="true" class="">https://aihub.org/</a>).
                                                          We help
                                                          researchers
                                                          publish the
                                                          latest AI
                                                          news,
                                                          summaries of
                                                          their work,
                                                          opinion
                                                          pieces,
                                                          tutorials and
                                                          more.  We are
                                                          supported by
                                                          many leading
                                                          scientific
                                                          organizations
                                                          in AI, namely
                                                          <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__aaai.org_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=wBvjOWTzEkbfFAGNj9wOaiJlXMODmHNcoWO5JYHugS0&e="
target="_blank" moz-do-not-send="true" class=""> AAAI</a>, <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__neurips.cc_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=3-lOHXyu8171pT_UE9hYWwK6ft4I-cvYkuX7shC00w0&e="
target="_blank" moz-do-not-send="true" class=""> NeurIPS</a>, <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__icml.cc_imls_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=JJyjwIpPy9gtKrZzBMbW3sRMh3P3Kcw-SvtxG35EiP0&e="
target="_blank" moz-do-not-send="true" class=""> ICML</a>, <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.journals.elsevier.com_artificial-2Dintelligence&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=eWrRCVWlcbySaH3XgacPpi0iR0-NDQYCLJ1x5yyMr8U&e="
target="_blank" moz-do-not-send="true" class=""> AIJ</a>/<a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.journals.elsevier.com_artificial-2Dintelligence&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=eWrRCVWlcbySaH3XgacPpi0iR0-NDQYCLJ1x5yyMr8U&e="
target="_blank" moz-do-not-send="true" class="">IJCAI</a>, <a
href="https://urldefense.proofpoint.com/v2/url?u=http-3A__sigai.acm.org_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=7rC6MJFaMqOms10EYDQwfnmX-zuVNhu9fz8cwUwiLGQ&e="
target="_blank" moz-do-not-send="true" class=""> ACM SIGAI</a>,
                                                          EurAI/AICOMM,
                                                          <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__claire-2Dai.org_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=66ZofDIhuDba6Fb0LhlMGD3XbBhU7ez7dc3HD5-pXec&e="
target="_blank" moz-do-not-send="true" class=""> CLAIRE</a> and <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.robocup.org__&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yl7-VPSvMrHWYKZFtKdFpThQ9UTb2jW14grhVOlAwV21R4FwPri0ROJ-uFdMqHy1&s=bBI6GRq--MHLpIIahwoVN8iyXXc7JAeH3kegNKcFJc0&e="
target="_blank" moz-do-not-send="true" class=""> RoboCup</a>.</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          <div class="">
                                                          <p
                                                          class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><span
                                                          style="font-family:"Arial",sans-serif"
                                                          class="">Twitter:
                                                          @aihuborg</span><o:p
                                                          class=""></o:p></p>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </blockquote>
                                                          </div>
                                                          </div>
                                                          </blockquote>
                                                        </div>
                                                      </div>
                                                    </blockquote>
                                                  </div>
                                                </div>
                                              </div>
                                            </blockquote>
                                          </div>
                                          <div
                                            id="gmail-m_3776411903040420401DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2"
                                            class="">
                                            <p class="MsoNormal"
                                              style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                                class=""></o:p></p>
                                            <table
                                              class="MsoNormalTable"
                                              style="border:none;border-top:solid
                                              windowtext
                                              1.0pt;border-color:currentcolor
                                              currentcolor"
                                              cellpadding="0" border="1">
                                              <tbody class="">
                                                <tr class="">
                                                  <td
                                                    style="width:41.25pt;border:none;padding:9.75pt
                                                    .75pt .75pt .75pt"
                                                    class="" width="55">
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"><a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.avast.com_sig-2Demail-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dlink-26utm-5Fcampaign-3Dsig-2Demail-26utm-5Fcontent-3Dwebmail&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=Ao9QQWtO62go0hx1tb3NU6xw2FNBadjj8q64-hl5Sx4&e="
                                                        target="_blank"
moz-do-not-send="true" class=""><span style="text-decoration:none"
                                                          class=""><img
style="width:.4791in;height:.302in" id="_x0000_i1025"
src="https://ipmcdn.avast.com/images/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif"
data-unique-identifier="" moz-do-not-send="true" class="" width="46"
                                                          height="29"
                                                          border="0"></span></a><o:p
                                                        class=""></o:p></p>
                                                  </td>
                                                  <td
                                                    style="width:352.5pt;border:none;padding:9.0pt
                                                    .75pt .75pt .75pt"
                                                    class="" width="470">
                                                    <p class="MsoNormal"
style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;line-height:13.5pt"><span
style="font-size:10.0pt;font-family:"Arial",sans-serif;color:#41424E"
                                                        class="">Virus-free.
                                                        <a
href="https://urldefense.proofpoint.com/v2/url?u=https-3A__www.avast.com_sig-2Demail-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dlink-26utm-5Fcampaign-3Dsig-2Demail-26utm-5Fcontent-3Dwebmail&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=waSKY67JF57IZXg30ysFB_R7OG9zoQwFwxyps6FbTa1Zh5mttxRot_t4N7mn68Pj&s=Ao9QQWtO62go0hx1tb3NU6xw2FNBadjj8q64-hl5Sx4&e="
target="_blank" moz-do-not-send="true" class=""> <span
                                                          style="color:#4453EA"
                                                          class="">www.avast.com</span></a>
                                                      </span><o:p
                                                        class=""></o:p></p>
                                                  </td>
                                                </tr>
                                              </tbody>
                                            </table>
                                          </div>
                                        </div>
                                      </blockquote>
                                    </div>
                                    <p class="MsoNormal"
                                      style="mso-margin-top-alt:auto;mso-margin-bottom-alt:auto"> <o:p
                                        class=""></o:p></p>
                                  </div>
                                </div>
                              </div>
                            </div>
                          </blockquote>
                        </div>
                      </div>
                    </div>
                  </blockquote>
                </blockquote>
                <div class="moz-signature">-- <br class="">
                  <span
                    id="cid:part51.0AC1D259.61E87F5D@rubic.rutgers.edu"><signature.png></span></div>
              </div>
            </div>
          </blockquote>
        </div>
        <br class="">
      </div>
    </blockquote>
    <div class="moz-signature">-- <br>
      <img src="cid:part61.D989DC19.5F87F962@rubic.rutgers.edu"
        border="0"></div>
  </body>
</html>