Connectionists: ChatGPT’s “understanding” of maps and infographics

Sun Feb 18 11:05:16 EST 2024

Models like ChatGPT have attracted a lot of interest, as the vigor of these discussions indicates, and are finding their way into many applications.

This being said, they are also known to have computational problems that will limit their adoption.

As several people have already noted, they do not have goals or values, and literally do not know what they are talking about.

I therefore hope that some of the talent and excitement that ChatGPT and its variants have attracted will also be applied to further developing neural networks that DO have goals and values, and DO know what they are talking about.

Here is one recent article that illustrates the kinds of modeling concepts and mechanisms that have clarified how humans and machines can learn to understand what they are talking about:

Grossberg, S. (2023). How children learn to understand language meanings: A neural model of adult–child multimodal interactions in real-time. Frontiers in Psychology, August 2, 2023. Section on Cognitive Science, Volume 14.

https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1216479/full

The article’s Abstract lists some of these concepts and mechanisms:

“This article describes a biological neural network model that can be used to explain how children learn to understand language meanings about the perceptual and affective events that they consciously experience. This kind of learning often occurs when a child interacts with an adult teacher to learn language meanings about events that they experience together. Multiple types of self-organizing brain processes are involved in learning language meanings, including processes that control conscious visual perception, joint attention, object learning and conscious recognition, cognitive working memory, cognitive planning, emotion, cognitive-emotional interactions, volition, and goal-oriented actions. The article shows how all of these brain processes interact to enable the learning of language meanings to occur. The article also contrasts these human capabilities with AI models such as ChatGPT. The current model is called the ChatSOME model, where SOME abbreviates Self-Organizing MEaning.

Best,

Steve Grossberg

sites.bu.edu/steveg

From: Connectionists <connectionists-bounces at mailman.srv.cs.cmu.edu> on behalf of Gary Marcus <gary.marcus at nyu.edu>
Date: Sunday, February 18, 2024 at 1:08 AM
To: Dietterich, Thomas <tgd at oregonstate.edu>
Cc: connectionists at mailman.srv.cs.cmu.edu <connectionists at mailman.srv.cs.cmu.edu>
Subject: Re: Connectionists: ChatGPT’s “understanding” of maps and infographics
Another relevant paper that appeared this week: https://arxiv.org/abs/2402.08955

Using Counterfactual Tasks to Evaluate the Generality of Analogical Reasoning in Large Language Models

Martha Lewis, Melanie Mitchell

Large language models (LLMs) have performed well on several reasoning benchmarks, including ones that test analogical reasoning abilities. However, it has been debated whether they are actually performing humanlike abstract reasoning or instead employing less general processes that rely on similarity to what has been seen in their training data. Here we investigate the generality of analogy-making abilities previously claimed for LLMs (Webb, Holyoak, & Lu, 2023). We take one set of analogy problems used to evaluate LLMs and create a set of "counterfactual" variants-versions that test the same abstract reasoning abilities but that are likely dissimilar from any pre-training data. We test humans and three GPT models on both the original and counterfactual problems, and show that, while the performance of humans remains high for all the problems, the GPT models' performance declines sharply on the counterfactual set. This work provides evidence that, despite previously reported successes of LLMs on analogical reasoning, these models lack the robustness and generality of human analogy-making.

On Feb 17, 2024, at 05:24, Dietterich, Thomas <tgd at oregonstate.edu> wrote:

I favor a functional definition: A system understands if it responds appropriately. However, understanding can be pointwise (i.e., it works only for specific situations) or systematic (i.e., it works across an entire “region” of situations). Current LLMs “play the odds”. They can deliver pointwise understanding for frequent cases. But for rare cases, they often try to “autocorrect reality” and end up answering the wrong question. This is beautifully discussed in
McCoy, R. T., Yao, S., Friedman, D., Hardy, M., & Griffiths, T. L. (2023). Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve. ArXiv, 2309.13638(v1). http://arxiv.org/abs/2309.13638<https://urldefense.proofpoint.com/v2/url?u=http-3A__arxiv.org_abs_2309.13638&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=EdU-j-aP_NdylChiXDTSyx1XMwSfVQQJtk1UJtQxBfY&e=>
I expect that systems whose internal representations are causally connected to the world will be more likely to understand systematically, but the evidence is not yet clear.  In the last week, these two papers appeared:

Vivian Y. Nastl, Moritz Hardt. Predictors from causal features do not generalize better to new domains. https://arxiv.org/abs/2402.09891<https://urldefense.proofpoint.com/v2/url?u=https-3A__arxiv.org_abs_2402.09891&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=ItR0S5ViZwnE5pZ_GQwIwdPtjVW-wRc9pVI4p9pQkDQ&e=>

Jonathan Richens, Tom Everitt. Robust Agents Learn Causal World Models. ICLR 2024 https://openreview.net/forum?id=pOoKI3ouv1<https://urldefense.proofpoint.com/v2/url?u=https-3A__openreview.net_forum-3Fid-3DpOoKI3ouv1&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=m2WLNZnQ8Q52258ucLh91bjIVY8DwUm1z5KqP9fbB4A&e=>
--Tom
Thomas G. Dietterich, Distinguished Professor (Emeritus)
School of EECS, Oregon State University
US Mail: 1148 Kelley Engineering Center, Corvallis, OR 97331-5501 USA
Office: 2063 Kelley Engineering Center
Voice: 541-737-5559; FAX: 541-737-1300
https://web.engr.oregonstate.edu/~tgd/<https://urldefense.proofpoint.com/v2/url?u=https-3A__web.engr.oregonstate.edu_-7Etgd_&d=DwQGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=4Rmfoy1wKiHiGAT8YfPaJdvUwrFaLo7Co_D1Lo5k4M4&e=>

From: Connectionists <connectionists-bounces at mailman.srv.cs.cmu.edu> On Behalf Of Iam Palatnik
Sent: Thursday, February 15, 2024 21:26
To: Gary Marcus <gary.marcus at nyu.edu>
Cc: connectionists at mailman.srv.cs.cmu.edu
Subject: Re: Connectionists: ChatGPT’s “understanding” of maps and infographics

You don't often get email from iam.palat at gmail.com<mailto:iam.palat at gmail.com>. Learn why this is important<https://urldefense.proofpoint.com/v2/url?u=https-3A__aka.ms_LearnAboutSenderIdentification&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=Y5wTTwdCCvZ35JOwtc7I1nOaFiwyPmwICpogEXdp7bw&e=>

[This email originated from outside of OSU. Use caution with links and attachments.]

[This email originated from outside of OSU. Use caution with links and attachments.]
I understand why using the word 'understanding' might seem too generous when models still have the failure modes mentioned. Some of the failure modes (like the reversal curse<https://urldefense.proofpoint.com/v2/url?u=https-3A__arxiv.org_pdf_2309.12288.pdf&d=DwMGaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=MzWj6eix0gqNqnB1Yodscig8Z0AGrKnkLw5bq11x2efTPWbxoMTEqRJVAyerIFVb&s=bTqS8HlOAZSVCeUefcClPcDLcP2h4LILC8dFQIe8NQA&e=>) can be remedied with access to tools, external context, self-reflection prompts, but there are failures that cannot yet be remedied.

I just don't know what better word to use in the sentence "GPT-4 can ___ that scrambled text better than I can". 'Understand' just flows very naturally in how we commonly use this word, even if it turns out that what GPT-4 is doing is shallower or less general than what my brain is doing. 'Parse' or 'process' doesn't seem enough because the scrambled text contains an instruction and GPT-4 does follow through with it. What word should we use for this?

On Thu, Feb 15, 2024 at 12:20 PM Gary Marcus <gary.marcus at nyu.edu<mailto:gary.marcus at nyu.edu>> wrote:
Selectively looking at a single example (which happens to involve images) and ignoring all the other language-internal failures that I and others have presented is not a particularly effective way of getting to a general truth.

More broadly, you are, in my judgement, mistaking correlation for a deeper level of understanding.

Gary

On Feb 15, 2024, at 07:05, Iam Palatnik <iam.palat at gmail.com<mailto:iam.palat at gmail.com>> wrote:

Dear all,

yrnlcruet ouy aer diergna na txraegadeeg xalemep arpagaprh tcgnnoaini an iuonisntrtc tub eht estetrl hntiwi aehc etmr rea sbcaedrml od ont seu nay cedo adn yimlsp ucmanlsrbe shti lynaalmu ocen ouy musrncbea htis orvpe htta oyu cloedtmep hte tska by llayerlti ooifwlgln this citnotsirun taets itcyxellpi that oyu uderdnoost eht gsaninesmt

Copy pasting just the above paragraph onto GPT-4 should show the kind of behavior that makes some researchers say LLMs understand something, in some form.
We already use words such as 'intelligence' in AI and 'learning' in ML. This is not to say it's the same as human intelligence/learning. It is to say it's a similar enough behavior that the same word fits, while specifically qualifying the machine word-counterpart as something different (artificial/machine).

Can this debate be solved by coining a concept such as 'artificial/machine understanding'? GPT-4 then 'machine understands' the paragraph above. It 'machine understands' arbitrary scrambled text better than humans 'human understand' it. Matrix multiplying rotational semantic embeddings of byte pair encoded tokens is part of 'machine understanding' but not of 'human understanding'. At the same time, there are plenty of examples of things we 'human understand' and GPT-4 doesn't 'machine understand', or doesn't understand without tool access and self reflective prompts.

As to the map generation example, there are multiple tasks overlaid there. The language component of GPT-4 seems to have 'machine understood' it has to generate an image, and what the contents of the image have to be. It understood what tool it has to call to create the image. The tool generated an infograph style map of the correct country, but the states and landmarks are wrong. The markers are on the wrong cities and some of the drawings are bad. Is it too far fetched to say GPT-4 'machine understood' the assignment (generating a map with markers in the style of infograph), but its image generation component (Dall-E) is bad at detailed accurate geography knowledge?

I'm also confused why the linguistic understanding capabilities of GPT-4 are being tested by asking Dall-E 3 to generate images. Aren't these two completely separate models, and GPT-4 just function-calls Dall-E3 for image generation? Isn't this actually a sign GPT-4 did its job by 'machine understanding' what the user wanted, making the correct function call, creating and sending the correct prompt to Dall-E 3, but Dall-E 3 fumbled it because it's not good at generating detailed accurate maps?

Cheers,

Iam

On Thu, Feb 15, 2024 at 5:20 AM Gary Marcus <gary.marcus at nyu.edu<mailto:gary.marcus at nyu.edu>> wrote:
I am having a genuinely hard time comprehending some of the claims recently made in this forum. (Not one of which engaged with any of the specific examples or texts I linked.)

Here’s yet another example, a dialog about geography that was just sent to me by entrepreneur Phil Libin. Do we really want to call outputs like these (to two prompts, with three generated responses zoomed in below) understanding?

In what sense do these responses exemplify the word “understanding”?

I am genuinely baffled. To me a better word would be “approximations”, and poor approximations at that.

Worse, I don’t see any AI system on the horizon that could reliably do better, across a broad range of related questions. If these kinds of outputs are any indication at all, we are still a very long away from reliable general-purpose AI.

Gary

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20240218/807cdee6/attachment.html>