Connectionists: Statistics versus Understanding in Generative AI.

[Danny Silver]

I have a tendancy to agree with Leslie Smith concerning the philosophical disjunction that is occurring in this discussion.

However, the distinction between an agent that statistically manipulates symbols (ie. can use language similar to humans) and an agent that has a grounded sense of concepts and relations (ie. knows about the world like humans) is not as obvious as it was 20 years ago.   Certainly, it has a lot to do with the sensory modalities the agent can perceive and its ability to integrate and act on information from those modalities.    We are getting closer to an understanding of how such modalities might be integrated.  Recent work on LLMs that can input from, and output to, various modalities (text, video, audio, tactile) show evidence that they develop complex embedded representations of concepts (conreps) that include, or are closely connected to, representations of symbols (symreps).  There is a lot more to do here in terms of how these sensory modalities are integrated in real-time (in terms of architecture and learning), but my sense (pardon the pun) is that many AI researchers are on the right path.

For more on symreps and conreps, please see our paper (Silver and Mitchell 2023) “The Roles of Symbols in Neural-based AI: They are Not What You Think<https://arxiv.org/abs/2304.13626>!”  from the Compendium of Neurosymbolic Artificial Intelligence 2023: p.1-28.  Our hypothesis and the conceptual architecture we present imply that symbols will remain critical to the future of intelligent systems NOT because they are the fundamental building blocks of thought, but because they are characterizations of subsymbolic processes that constitute thought.

==========================
Daniel L. Silver
Professor, Jodrey School of Computer Science
Data Scientist, Acadia Institute for Data Analytics
Acadia University,
Office 314, Carnegie Hall,
Wolfville, Nova Scotia Canada B4P 2R6
Cell: (902) 679-9315

acadiau.ca<http://www2.acadiau.ca/index.php>
Facebook<https://www.facebook.com/acadiauniversity>  Twitter<https://twitter.com/acadiau>  YouTube<https://www.youtube.com/user/AcadiaWebmaster>  LinkedIn<https://www.linkedin.com/company/acadia-university?trk=biz-companies-cym>  Flickr<https://www.flickr.com/photos/acadiauniversity/albums>
[id:image001.png at 01D366AF.7F868A70]


From: Connectionists <connectionists-bounces at mailman.srv.cs.cmu.edu> on behalf of Prof Leslie Smith <l.s.smith at cs.stir.ac.uk>
Date: Wednesday, February 14, 2024 at 1:38 AM
To: connectionists at mailman.srv.cs.cmu.edu <connectionists at mailman.srv.cs.cmu.edu>
Subject: Re: Connectionists: Statistics versus Understanding in Generative AI.
CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments.

Can I suggest that there's a deep philosophical disjunction here.

I'd say that it hinges on whether "Understanding" implies grounded-ness or
not. Does it require the understanding entity to have a model of the
environment into which the information can be placed, or can understanding
be achieved in an abstract system that only has access to a world of
textual data?

One one hand, there's a danger that we imply that understanding only
becomes possible in living systems, whatever the capability of the
synthetic system, and on the other hand there's a danger that we ascribe
understanding to systems that we really shouldn't. Does a thermostat
understand temperature?

This is a difficult question altogether: at what point to we ascribe
understanding to animals? They all behave in ways appropriate to their
environment, from the single celled Paramecium upwards, but we don't tend
to consider them to understand their environment until we consider much
higher animals (like mice or rats).

--Leslie Smith

Gary Marcus wrote:


--
Prof Leslie Smith (Emeritus)
Computing Science & Mathematics,
University of Stirling, Stirling FK9 4LA
Scotland, UK
Mobile: +44 7913 219013
Web: http://www.cs.stir.ac.uk/~lss
Blog: http://lestheprof.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20240214/df92f29c/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 7490 bytes
Desc: image001.png
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20240214/df92f29c/attachment.png>