A brief appreciation of @NegarestaniReza's _Intelligence and Spirit_ from @urbanomicdotcom, and in particular, the final chapter on the philosophy of intelligence. https://amzn.to/3elGWI7 

I&S gets something really right, I think, which is the need to decouple "intelligence" from the ability to perform a task.

That's the metric of GPU era, driven in part by the "bake off" style of computer science conferences (our new algorithm classifies X Y points better, &c.)

It's the same thing that IQ people do: construct a task that, successfully completed, indicates the presence of intelligence. It boils down to pattern matching with priors, and another bake-off, although ironically, the IQ people want the least "ecological" tasks, AI the most.

An AI is smart if it can drive a car as well as a human; a human is smart if it can pattern match Ravens Progressive Matrices like a computer.

The most interesting work in cognitive science (by contrast) has moved well beyond the dead-end of utility theory.

The free energy principle has a lot of problems. Including some basic faults in the mathematics that have led more than just me to throw our hands up in frustration.

BUT, it's a huge advance because it's a (predictive) theory that makes the epistemic state its own utility.

[I talk a little about the general idea in a talk at @HarryDCrane's seminar. Standard rational choice makes behavior a function of P(x) (your beliefs about X), and U(x) (your desires for X); these new kinds of theories replace U(x) with an f(P(x)).]

One of the really nice things about I&S is his use of the Philebus, which (who knew!) takes up this very problem. What matters to a good life—knowledge or pleasure? (One lovely feature of the answer is, well, neither is the thing itself, but if you ask me, more the former.)

Out of this (in part), Reza proposes the agathosic test for intelligence: "can you make something better than yourself". You have to kind of practice saying that word, but I'm totally pumped to use it in a talk one day. ("Better" is punted—more on this.)

What's important about this is that Reza makes an attack on the identification problem.

We all know that intelligence is a thing. Unless you've hopelessly swallowed some pretty dark pills, you see it in an enormous range of human contexts.

Again, standard answers are (at best) correlational—and in as much as they correlate, they're construct-invalid. The better they predict intelligent behavior, the more clearly they don't capture it. http://simondedeo.com/?p=337 

But a serious attack on the question runs into the problem that we're intelligent agents ourselves. It's worse than birds attempting ornithology—we're forced to use the very thing we want to understand in the understanding task itself.

If that sounds a little recursive (and thus vulnerable to diagonalization, Godelian problems, &c.), I think it should. There are thoughts a computer can not think ("what are the general principles of a halting program", most famously)—so the project can't get off the ground.

Reza's answers take in a huge literature. There's some Type Theory (which IMO suggests one answer to the Godelian problem). But there's also some of that old pragmatist literature (Nelson Goodman), and even some Brandom on the other side of Oakland.

One thing I particularly like is the emphasis on the *artificiality* of intelligence. This helps get us away from a "forms of life" story that being smart is just being. The history (anthropology) of intelligence is a technological one—the Axial Age its Industrial Revolution.

I won't try to summarize Reza's answer—in part, I think, because it would be a really good way to show I don't understand it! That's not a mystical statement (this isn't the Tractatus), but rather a claim that he's provided some really good tools to think about the problem.

I honestly Reza has *the best book* on the philosophy of intelligence out there, and one that can be profitably read by non-philosophers who want to break out of a set of mental boxes that created by the superintelligence crowd.

I'm really bad at selling books. Let me try again.

A few years ago, Alva Noë came out with Strange Tools, on aesthetics. It's not a theory of beauty, but a theory of what art is doing in the world. Not socially (Bordieu), but "what the task is that, in striving for it, puts something in the art category". https://amzn.to/3a0Sc9m 

When you try to do science on "art", bad accounts of what it is mean that you skitter off to one side, and end up actually talking about class, or economics. So good ones really matter.

I think Reza has done something similar for intelligence.

We're in a deadlock right now, and the next generation of AI researchers is going to be disappointed by the ideas we have about what they were promised (AGI), why it's not here yet, and why the approaches are clearly not going to get us there.

Secretly, I think Reza's ideas will show up in the right circles in ten years, both among the practitioners, and their funders. So, very selfishly, it's a way to stay ahead of the curve.