AI megathread

[Laird]

What exactly am I taking on faith? We know how the inference process works. Where in that process do you see anything non-physical occurring?

Just to be clear, although you seem to have understood given your "like" of my last response, my point was that forming an opinion on something based on direct, daily, professional experience with that something is the complete opposite of taking it on faith. Implying otherwise deserves a facepalm.

It’s correct that I think their workings can be explained entirely in computational terms without considering any kind of idealism as an underlying feature of reality.

So they aren't mysterious?

There's an algorithm that perhaps has yet to been ascertained, in terms of a current "black box", but ultimately how LLMs work will be inline with any other program running on a Turing Machine.

There may be aspects of how an LLM works that parallel how human intelligence works, but there's no solution here to the Hard Problems of Subjectivity, Intentionality, Rationality. (My guess is humans don't have the kind of "jagged intelligence" LLMs have so the parallels will be minimal.)

Physicalism remains as much a dead end faith as ever.

An algorithm can be impressively designed by intelligent conscious human programmers ~ but that does not make the algorithm in itself "intelligent" or conscious or what-have-you.

Yeah any user of a system can be fooled. When I was young I thought video game AI had reached incredible levels...only to later actually learn all the tricks devs used to make the enemies seem intelligent. It's odd to me that professors who you and I post are apparently not equal to customers, article writers who need clicks, AI companies own hype machines about their unreleased models...

On Brian Cantwell Smith and the Promise of AI

Melanie Mitchell, Professor at the Santa Fe Institute, is the award-winning author of Artificial Intelligence: A Guide for Thinking Humans. She works in the fields of AI, cognitive science, and complex systems.

Today I had the bittersweet pleasure of participating in a symposium honoring the late philosopher Brian Cantwell Smith, a good friend whom I’d known for over 30 years. Before he died last year, Brian held an endowed chair at the University of Toronto: the Reid Hoffman Chair in Artificial Intelligence and the Human.

..It is our embedding and engagement, our ability to conceptualize the world and ourselves in it—what AI folks nowadays like to call “having a world model”—the thing that AI doesn’t currently have, the thing that allows we humans to figure out what to do in the novel situations we continually find ourselves in.

As Brian wrote,

“How we register the world...find it ontologically intelligible in such a way as to support our projects and practices—is in my judgment the most important task to which intelligence is devoted.”
He went on:

“AI needs to take on board one of the deepest intellectual realizations of the last 50 years...that taking the world to consist of discrete intelligible mesoscale objects is an achievement of intelligence, not a premise on top of which intelligence runs. AI needs to explain objects, properties, and relations, and the ability of creatures to find the world intelligible in terms of them; it cannot assume them.”

This was written before the advent of large language models, but it is excruciatingly true for them—they are built entirely upon the achievements of human intelligence. They are handed the world on a silver platter in the form of human written text...

So they aren't mysterious?

There's an algorithm that perhaps has yet to been ascertained, in terms of a current "black box", but ultimately how LLMs work will be inline with any other program running on a Turing Machine.

Agree. But it's worth remembering that humans can't solve the general halting problem either - only particular instances, just like computers can.

There may be aspects of how an LLM works that parallel how human intelligence works, but there's no solution here to the Hard Problems of Subjectivity, Intentionality, Rationality. (My guess is humans don't have the kind of "jagged intelligence" LLMs have so the parallels will be minimal.)

Indeed, there may be a parallel. When an LLM works through a problem step by step, the resulting reasoning often looks remarkably similar to the way humans articulate their reasoning.

"I have direct, daily, professional experience with this."

"Oh, so you're just taking it on faith."

🤦‍♂️

None of us are going to give the personal information necessary to verify claims like this.

I think everyone agrees there are leaps in what these LLMs can do.

The contention is whether the difference is in degree, or has there been a difference in kind.

It seems to me the difference remains one of degree, especially since I am doubtful there's intelligence without consciousness.

It seems to me the difference remains one of degree, especially since I am doubtful there's intelligence without consciousness.

There are 2 problems here. 1) There’s no agrees definition to what intelligence really means 2) It’s not possible to prove humans aren’t Turing machines too (in the very abstract sense).

There are 2 problems here. 1) There’s no agrees definition to what intelligence really means 2) It’s not possible to prove humans aren’t Turing machines too (in the very abstract sense).

2 seems to be a double negative.

I think it is very easy to see humans aren't Turing Machines, since a Turing Machine is an abstraction we've no reason to believe has thoughts unless we grant thoughts to its components.

2 seems to be a double negative.

I think it is very easy to see humans aren't Turing Machines, since a Turing Machine is an abstraction we've no reason to believe has thoughts unless we grant thoughts to its components.

I thought we were discussing “problem” solving human vs machine. Can a computer program solve Goldbach’s conjecture by finding a general proof? Certainly not today - but dare we say never?

Agree. But it's worth remembering that humans can't solve the general halting problem either - only particular instances, just like computers can.


Indeed, there may be a parallel. When an LLM works through a problem step by step, the resulting reasoning often looks remarkably similar to the way humans articulate their reasoning.

Can a computer, without human aid, even glean what the Halting Problem is?

As for similarity in articulation, it seems to me it's the human reasoning that creates the syntax that allows, in some cases, the program's ability to reach conclusions. I don't think theorem provers or Prolog programs are reasoning either.