<div dir="ltr">I would argue that humans also have "patches of competence", or expertise, and vast gulfs between them just like ChatGPT. But what we have that ChatGPT does not, is a means to approximate a metric of when we are operating in an area of expertise (and for some those approximations are looser than others..). <div><br></div><div>Even more importantly, we understand Q&A through a epistemic/moral perspective which includes:</div><div>- that there is such a thing as being wrong</div><div>- that there is such a thing as value </div><div>- that there is value in being right</div><div>- Sometimes there is value in being wrong</div><div>- etc...</div><div><br></div><div>So I think that even once we develop some kind of causal reasoning engine to allow a model to generalize somewhat accurately outside of its patches, the harder question is to give the model an understanding of these deeper aspects of interaction. Otherwise it will always be easy to run them in circles, break them out of their sandboxes, and prompt them to give stupid answers. </div><div><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Mar 9, 2023 at 3:37 PM Dietterich, Thomas <<a href="mailto:tgd@oregonstate.edu">tgd@oregonstate.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="msg2512655426513871632">
<div lang="EN-US">
<div class="m_2512655426513871632WordSection1">
<p class="MsoNormal">ChatGPT’s errors reveal that its “understanding” of the world is not systematic but rather consists of patches of competence separated by regions of incompetence and incoherence. ChatCPT would be much stronger if it could fill in the gaps
between those patches by appealing to general causal models. This raises two questions: (a) how could a system learn such causal models and (b) how could we test a system to determine whether it had succeeded.
<u></u><u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<p class="MsoNormal">Researchers in the area of symbolic regression and causal discovery have an answer to (a): Learn a symbolic (e.g., DAG or differential equation) model. If we seek a fully connectionist account, how can we encourage/enforce systematicity
in the knowledge acquired through deep learning? Regarding (b), does anyone have pointers to work on testing whether a connectionist system has acquired a systematic understanding of a causal relationship? This must go beyond testing (x,y) points, perhaps
by verifying (Lipschitz) continuity?<u></u><u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<p class="MsoNormal">--Tom<u></u><u></u></p>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console"">Thomas G. Dietterich, Distinguished Professor Voice: 541-737-5559<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console"">School of Electrical Engineering FAX: 541-737-1300<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console""> and Computer Science URL: <a href="http://eecs.oregonstate.edu/~tgd" target="_blank">eecs.oregonstate.edu/~tgd</a><u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console"">US Mail: 1148 Kelley Engineering Center
<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console"">Office: 2067 Kelley Engineering Center<u></u><u></u></span></p>
<p class="MsoNormal"><span style="font-size:10pt;font-family:"Lucida Console"">Oregon State Univ., Corvallis, OR 97331-5501<u></u><u></u></span></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div style="border-right:none;border-bottom:none;border-left:none;border-top:1pt solid rgb(225,225,225);padding:3pt 0in 0in">
<p class="MsoNormal"><b>From:</b> Connectionists <<a href="mailto:connectionists-bounces@mailman.srv.cs.cmu.edu" target="_blank">connectionists-bounces@mailman.srv.cs.cmu.edu</a>>
<b>On Behalf Of </b>Stefan C Kremer<br>
<b>Sent:</b> Wednesday, March 8, 2023 17:44<br>
<b>To:</b> Terry Sejnowski <<a href="mailto:terry@salk.edu" target="_blank">terry@salk.edu</a>><br>
<b>Cc:</b> Connectionists List <<a href="mailto:connectionists@cs.cmu.edu" target="_blank">connectionists@cs.cmu.edu</a>><br>
<b>Subject:</b> Re: Connectionists: Chomsky's apple<u></u><u></u></p>
</div>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
<p><span style="color:rgb(215,63,9)">[This email originated from outside of OSU. Use caution with links and attachments.]</span><u></u><u></u></p>
<div>
<p class="MsoNormal">One of the criticisms against John Searle’s argument (<a href="https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2FChinese_room&data=05%7C01%7Ctgd%40oregonstate.edu%7C784658ebd66d4fa3f07308db2071adfb%7Cce6d05e13c5e4d6287a84c4a2713c113%7C0%7C0%7C638139444863338994%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=dah%2FU3oilsuzZ%2BzWZOtJVUwodjpPRjsjC3pkyPU9Ih4%3D&reserved=0" target="_blank">https://en.wikipedia.org/wiki/Chinese_room</a>)
has always been that it wouldn’t be possible to construct a book comprehensive enough to answer all the queries, or that it would take too long to produce an output. Chat GPT shows that we have at least approached that limitation (perhaps not truly overcome
it…yet). <u></u><u></u></p>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">The question posed by Searle (and answered with a “yes” by Chomsky in his thinking about counterfactuals, causal explanation, and thinking) is: is there a difference between understanding and simulated understanding?<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">I don’t know how we could ever answer this question (an to me that’s the important point), but it seems that Searle’s thought experiment becomes more relevant, now that a feasible implementation can be constructed, than when it was originally
proposed.<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
<div>
<p class="MsoNormal">Stefan<u></u><u></u></p>
</div>
<div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<p class="MsoNormal"><br>
<br>
<u></u><u></u></p>
<blockquote style="margin-top:5pt;margin-bottom:5pt">
<div>
<p class="MsoNormal">On Mar 8, 2023, at 12:25 PM, Terry Sejnowski <<a href="mailto:terry@salk.edu" target="_blank">terry@salk.edu</a>> wrote:<u></u><u></u></p>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
<div>
<div>
<p class="MsoNormal">CAUTION: This email originated from outside of the University of Guelph. Do not click links or open attachments unless you recognize the sender and know the content is safe. If in doubt, forward suspicious emails to
<a href="mailto:IThelp@uoguelph.ca" target="_blank">IThelp@uoguelph.ca</a><br>
<br>
<br>
I have always been impressed with Chomsky's ability to use plausible<br>
arguments to make his case even when they were fallacious.<br>
<br>
<a href="https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html#commentsContainer" target="_blank">https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html#commentsContainer</a><br>
<br>
"Here’s an example. Suppose you are holding an apple in your hand. Now you let the apple go. You observe the result and say, “The apple falls.” That is a description. A prediction might have been the statement “The apple will fall if I open my hand.” Both are
valuable, and both can be correct. But an explanation is something more: It includes not only descriptions and predictions but also counterfactual conjectures like “Any such object would fall,” plus the additional clause “because of the force of gravity” or
“because of the curvature of space-time” or whatever. That is a causal explanation: “The apple would not have fallen but for the force of gravity.” That is thinking."<br>
<br>
His apple example sounded convincing, so I tested GPT on it:<br>
<br>
Me: If I hold an apple in my hand and let it go what will happen?<br>
<br>
GPT: The apple will fall to the ground due to the force of gravity.<br>
<br>
Me: What about any such object?<br>
<br>
GPT: Any object released from a person's hand will fall to the ground due to the force of gravity.<br>
<br>
Me: What would happen if there wasn't a force of gravity?<br>
<br>
GPT: If there were no force of gravity, objects would not be able to stay on the ground and would float away in all directions<br>
<br>
Chomsky defined thinking by the correct answers to these questions. <br>
<br>
Alas, the tide has turned.<br>
<br>
Terry<br>
<br>
-----<u></u><u></u></p>
</div>
</div>
</blockquote>
</div>
<p class="MsoNormal"><u></u> <u></u></p>
</div>
</div>
</div>
</div>
</div></blockquote></div><br clear="all"><div><br></div><span class="gmail_signature_prefix">-- </span><br><div dir="ltr" class="gmail_signature"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr">Brad Wyble<br>Professor<span style="font-size:12.8px"> of Psychology</span> <br>Penn State University<div><br></div></div></div></div></div></div>