Gort, Robert Wright, and A.I. @3QD [GT-5]

My latest 3 Quarks Daily article is up:

Klaatu barada nikto: Robert Wright’s The God Test

It’s a review of Robert Wright’s current book, The God Test. It turned out to be a tough slog, more than I’d anticipated.

I started blogging about it well ahead of time. I figured I’d write a half-dozen to ten or so blog posts ahead of time and then wrap things up with the formal review. I wrote some blog posts, but by Tuesday, August 4, less than a week before my article was due, I was still blogging about coming up with an approach and hadn’t gotten around to the various things I wanted to cover in the lead-up. Then the weekend arrived and I had to start writing. So I did, here and then. But things wouldn’t converge. Still, I managed to write something and post it on Sunday afternoon. But I wasn’t happy.

Of course, while sometimes I’m satisfied with my 3QD pieces, that’s not always so. This was just one of those times. But still.

On Monday, when the piece actually appeared, I thought I had one of those situations I often had as an undergraduate. I’d write the term paper, turn it in, and then it would hit me, “this is what I should have done.” The process of writing the piece had taken me to a different intellectual place, one where I saw what had to be done. I had such moments of illumination on Mondey, Tuesday as well. But none of them held.

And so, here I am, still not knowing how I could have done a better job of reviewing that book. On the one hand, I wanted to give it a good review. Wright’s a smart guy, I’ve followed him for a long time, I think his “cosmic perspective” is important and I think his views on international cooperation on AI development are important as well. On the other hand, I don’t think much of his account of how LLMs work, but I just didn’t want to get into that. Yes, I indicated one problem in the review, Wright’s casual treatment of semantic relationships. As I indicated in the review, Bert Idem has discussed other problems. But I see still other problems. And I just didn’t want to get into that morass. Going into that would have swamped the review. (Let me say, moreover, that I think Wright is unnecessarily credulous of the Silicon Valley view of the technology. He should have listened more carefully to Gary Marcus.)

It is what it is.

But I will say this. I am working on my own book in which I present my own “cosmic” view: Play: How to Stay Human in the AI Revolution. On this I agree with Wright. That is the perspective we need to take. But I approach it in somewhat different terms, terms which I’ve sketched out in a piece I published four years ago, Welcome to the Fourth Arena – The World is Gifted. Note that in the book length version I do not intend to even try to explain how LLMs function internally. The technology that will take us there – whatever and wherever it is – will necessarily involve different architectures. The last chapter of Play will sketch out the kind of research program I think we need to take us there. Here’s a sketch: Brave New World: Notes on the next 30 years in AI [Work in Progress].

As for explaining how LLMs work, that’s a different project. I don’t intend to work on that. But I am thinking about a book that would be useful to anyone who is interested in that subject. The book I’m imagining would be preliminary to such an account: Language, Memory, and Mind: A Supplement to The Computer and the Brain.