Welcome to Welcome back to the tech policy podcast. I'm Corbyn Barthold.

Sayash Kapoor is here. He is a computer science PhD candidate at Princeton University's

Center for Information Technology Policy.

He is the co-author with Arvind Narayanan of the new book, A.I. Snake Oil,

what artificial intelligence can do, what it can't, and how to tell the difference.

Very excited to have him on. I have so many questions for him.

Sayash is a master debunker of AI hype. I'm going to guess that if he had his druthers,

we'd all talk less about the likelihood of AI enslaving or destroying humanity.

So I owe him a bit of an apology at the outset, because we're going to start by discussing

the likelihood of AI enslaving or destroying humanity.

Sash wrote a fantastic article with Arvind on the misleading precision of AI existential risk probabilities.

It drives me nuts when supposedly rigorous experts make highly unrigorous, but headline-grabbing, predictions,

such as that we have a one-and-six chance of extinction this century.

As Syash and Arvind Wright, experts vague intuitions and fears are being translated into pseudo-precise numbers.

Bingo.

So we're going to start with probabilistic snake oil, and then we'll move on to AI snake oil proper. Sayash, it is so great to have you on the show. Thank you so much for

having me. It's a real pleasure.

AI and existential risk. It's very popular to throw around your P-Doom in tech circles these days.

We get Eliyzer-Yudkowski saying his P-Doom is 95%. So he thinks there's a 95% chance that in some relatively near-term scenario, AI is going to kill us or something. Mark Andresens is

0% as those confident and precise numbers suggest, something rather unscientific seems to be going on

if we can get 95 and 0 and everything in between.

You guys devote a chapter in your book to these sorts of claims.

...