Does AI Understand Things? | Robert Wright & Gary Marcus

Information:

Subscribe to The Nonzero Newsletter at https://nonzero.substack.com
Bob's piece on Searle's "Chinese Room" argument.

0:00 Gary’s background and place in the AI world
1:17 Are Gary’s views on AI paradoxical?
8:48 Bob: Searle’s Chinese Room argument is dead
19:01 Have LLMs demonstrated "theory of mind"?
26:06 Arguing the semantics of an LLM's “semantic space”
31:36 Do LLM representations map onto the real world?
40:52 Can (and should) we slow down AI development?
51:43 Gary: I’ve never seen a field as myopic as AI today
54:25 The (symbolic?) future of AI

Robert Wright (Nonzero, The Evolution of God, Why Buddhism Is True) and Gary Marcus (https://garymarcus.substack.com/, Humans vs. Machines, The Algebraic Mind, Kluge). Recorded February 13, 2024.

Marcus at about 53:34:

The thing that I would say that's getting missing is we actually need to look at some other approaches. And this is taking all the oxygen. Right now all the funding, every graduate student goes into this because that's where the money is.

And I've never seen as intellectually narrow a field as I see right now in AI. There is one approach which is the Transformer architecture which was developed in 2017 and that is the only thing that like 95% of the people are using. 95% of the time all the funding from The Venture capitalists go there, and so forth. My view is in the long term it is not the answer to AI, that by 2030 we're going to look back and say they really were all in on that one and they should have looked at other things sooner. Like that's how I think it's going to turn out.

That's consistent with the argument I made almost three months ago in 3 Quarks Daily: Aye Aye, Cap’n! Investing in AI is like buying shares in a whaling voyage captained by a man who knows all about ships and little about whales.