Connectionists: Yann Le Cun and Stephen Hanson on Yann's new vision of AI

[Stephen José Hanson]

Hi Park

So to be clear, its actually not heterogeneous in a representational sense.   That's what is often confabulated by critics.. as soon as a neural architecture.. oh I suppose like a brain.. it must be symbolic!  Billions of Neurons can't possibly be reasoning.

So to the extent Yann's system can monitor an error signal , a mismatch, a credit reconciliation.. backpropagation will still work fine over a set of homogeneous representaional elements, like neurons.

Steve (my last name is Hanson, I suspect I just made the same error with you!)


On 12/22/22 15:00, Park, Deok Gun wrote:
Thank you, Hanson, for sharing this.

It is inspiring to see a non-homogeneous cognitive architecture from mainstream researcher.   If anyone is pursuing similar approaches, please let me know so that I can learn from it.

I was pursuing similar direction myself.  My version also has a hierarchical (heterarchical) prediction memory and simulation device for mode 2 thinking.

Below is the link to the position paper.  Like Yann’s, this is more of a proposal without results.

https://arxiv.org/abs/2212.08626<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Farxiv.org%2Fabs%2F2212.08626&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264194506%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=sAUihmwXTzRIJDUp7LTSloLsRw%2BchaQvt1oiOn0%2Fxgc%3D&reserved=0>

If you find this paper interesting, I would be happy to have a seminar or meeting for discussion.  I would also appreciate your feedback about my direction.

Thank you for reading.

Regards

===================
Deokgun Park
Assistant Prof. at UTA CSE
===================

From: Connectionists <connectionists-bounces at mailman.srv.cs.cmu.edu><mailto:connectionists-bounces at mailman.srv.cs.cmu.edu> on behalf of Stephen José Hanson <jose at rubic.rutgers.edu><mailto:jose at rubic.rutgers.edu>
Date: Thursday, December 15, 2022 at 8:46 AM
To: connectionists at mailman.srv.cs.cmu.edu<mailto:connectionists at mailman.srv.cs.cmu.edu> <connectionists at mailman.srv.cs.cmu.edu><mailto:connectionists at mailman.srv.cs.cmu.edu>
Subject: Connectionists: Yann Le Cun and Stephen Hanson on Yann's new vision of AI

[External]

In the last year, Yann Le Cun has posted a new vision for AI, partly avoiding the ambiguous complexity of LLM, and AGI, he focuses on
explicit reasoning systems, going beyond the implicit classification systems which have been so successful with Deep Learning.  Like other AI leaders, he realized that  somethingthay must be scaled down before scaling up.  It would be better to get the reasoning sense of a non-human mammal, say a cat, then jump to human reasoning, Yann argues.

This also underscores, I think,  much of the problem with LLM--e.g. ChatBOTs, which probably  will be a concern for classicial lingustics, nonetheless are inscructable.

At the same time critics have emerged, because when one provides enough detail of a specific vision it is both courageous and risky.   Our favorite critic/troll, Gary Marcus wrote several critiques (see  the one below), and  Yann and Jake Browning tried to clarify, and deflate some of the Marcus arguments, as many of them  appear to almost perversely misunderstand the difference between complex Nerual Network architectures and symbolic systems from the 1970s (which mostly didn't work).

Here is the conversation in whole focused entirely on Yann's new AI vision.  Cheers and Happy Holidays--Steve

https://www.youtube.com/watch?v=4s2S2l2OvdM&t=32s<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D4s2S2l2OvdM%26t%3D32s&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264194506%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=woH84FLS7JXNThQYw9uhqLZdH8iLyCYkfQAboD7f28g%3D&reserved=0>

The  description of the paper and critques.

https://ai.facebook.com/blog/yann-lecun-advances-in-ai-research/<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fai.facebook.com%2Fblog%2Fyann-lecun-advances-in-ai-research%2F&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264350753%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=etkNIQNxE9WkuIVDnzkFyTbc8EkjJZN7Z1%2FSkxwlHwo%3D&reserved=0>

https://www.technologyreview.com/2022/06/24/1054817/yann-lecun-bold-new-vision-future-ai-deep-learning-meta/<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.technologyreview.com%2F2022%2F06%2F24%2F1054817%2Fyann-lecun-bold-new-vision-future-ai-deep-learning-meta%2F&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264350753%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=bdEu9EVczOkilXR0fubFeiUuGobJD%2BSCzr37g4lzEFA%3D&reserved=0>

https://garymarcus.substack.com/p/how-new-are-yann-lecuns-new-ideas<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgarymarcus.substack.com%2Fp%2Fhow-new-are-yann-lecuns-new-ideas&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264350753%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=wwKbuJyFKOn27u7aongD9LHulbfsVgzA6KM8CSfQx8g%3D&reserved=0>

https://www.noemamag.com/what-ai-can-tell-us-about-intelligence/<https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.noemamag.com%2Fwhat-ai-can-tell-us-about-intelligence%2F&data=05%7C01%7Cjose%40rubic.rutgers.edu%7C866a5fe7a0974c1eb0f408dae45737b4%7Cb92d2b234d35447093ff69aca6632ffe%7C1%7C0%7C638073361264350753%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C2000%7C%7C%7C&sdata=XZK7vQJDJRiuVxSPnDkYSD1oB9OUxacQzkWxpwFLaYw%3D&reserved=0>

*(due to a small mic issue, we have also closed captioned the conversation)

--

Stephen José Hanson

Professor, Psychology Department

Director, RUBIC (Rutgers University Brain Imaging Center)

Member, Executive Committee, RUCCS

--
Stephen José Hanson
Professor, Psychology Department
Director, RUBIC (Rutgers University Brain Imaging Center)
Member, Executive Committee, RUCCS
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20221222/d5ece55e/attachment.html>