Connectionists: AI at 50 videos...

[Danny Silver]

Dear Barrack .. I fully agree with your statements below,  but would add to it the following:

There is no question that the human nervous system has the ability to represent a simple symbol that is associated with a complex concept such as <cat>.   We can use the written word ‘cat' or spoken word "chat”  or alternatively an iconic image of a cat's face as a symbol to convey a message about a <cat>,  so our brains do learn representations for such.   But this distributed representation came into being primarily for communication purposes forced by and helping to form the social nature of our existence with each other.  Later in human development (as well as in several other animals), such symbolic representations (symreps) were found to be useful as an additional constraint on learning and reasoning about their more complex related concepts - which have a much more complex distributed representation (conreps) within the brain.

In our 2023 paper (https://arxiv.org/abs/2304.13626) Tom Mitchell and I present a Neural-Symbolic Hypothesis:  "Symbols are critical to intelligence NOT because they are the building blocks of thought, but because they are characterizations of thought that (1) allow us to explain our subsymbolic thinking to ourselves and others and (2) act as constraints on inference and learning about the world. Symbols explain our thinking and aid our thinking, but are not the foundation of our thinking."

We conjecture that internal agent “self-communication” using symreps meant for human-to-human communication, became key to human intelligence because: (1) it provides a second, more abstract level of representation and reasoning which can occur in parallel with subsymbolic reasoning, and (2) it places an additional constraint on learning where prior learning act as an inductive bias for learning new symbols and concepts. Shared symbols allow us to explain and justify, internally as well as externally, our decisions and actions. And what we learn is shaped and constrained by the “lexicon” of what we recognize as symbols.

For the readers digest version of the paper see the Neural-Symbolic Conference abstract at https://ceur-ws.org/Vol-3432/paper40.pdf

Best regards  ... Danny


==========================

Daniel L. Silver PhD, CIM

Professor Emeritus, 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 Barak A. Pearlmutter <barak at pearlmutter.net>
Date: Tuesday, August 25, 2026 at 12:25 PM
To: connectionists at mailman.srv.cs.cmu.edu <connectionists at mailman.srv.cs.cmu.edu>
Subject: Re: Connectionists: AI at 50 videos...

CAUTION: This email comes from outside Acadia. Verify the sender and use caution with any requests, links or attachments.

> Plate (2002): “Another equivalent property is that in a distributed representation one cannot interpret the meaning of activity on a single neuron in isolation: the meaning of activity on any particular neuron is dependent on the activity in other neurons (Thorpe, 1995).”

> Thorpe (1995, p. 550): “With a local representation, activity in individual units can be interpreted directly … with distributed coding individual units cannot be interpreted without knowing the state of other units in the network.”

> Elman (1995, p. 210): “These representations are distributed, which typically has the consequence that interpretable information cannot be obtained by examining activity of single hidden units.”

People blithely toss out statements like that, but local
uninterpretability does not follow from the definition of a
distributed representation.

It is very easy to construct distributed representations whose
components *can* be understood in isolation (the bits of a standard
binary representation of a natural number ≤2³², say) or whose
components cannot be understood in isolation (a code for a natural
number ≤2³² whose bits each have an associated set of half the numbers
in the domain picked at random, say).

The "big regions" distributed representation for points in 2D space
from the PDP books is locally interpretable.

Naturally, all other things being equal, given the choice we'd choose
distributed representations whose parts are interpretable in
isolation. Does the brain do that? Well, deep learning networks seem
to, and experimental data seems consistent with that notion.

--Barak Pearlmutter.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20260826/b411ff11/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 7489 bytes
Desc: image001.png
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20260826/b411ff11/attachment.png>