As AI’s capabilities improve, reports of “rogue” agents and misaligned systems are becoming more frequent. In response, some activists, academics, and politicians have called for a pause in development, warning of the dangers that superintelligent AI could pose. But Richard Socher, the CEO of AI startup Recursive as well as the AI search engine You.com, sees another path. In his new book, The Eureka Machine, Socher lays out his vision for how self-improving AI could lead to breakthroughs in science, medicine, economics, and more.

In conversation with Nicholas Thompson, CEO of The Atlantic, he details some of the ways we can mitigate its risks, explains why the limits of our own minds could hold AI back, and considers whether we’re projecting too many fears about human behavior onto an insensate technology.

Produced independently of The Atlantic’s editorial staff.

(00:00) Introduction

(02:25) Richard's childhood in East Germany and Ethiopia

(10:23) What Richard got wrong about AI hype and the DeepMind acquisition

(12:14) The idea behind The Eureka Machine

(17:18) How AIs can compete to make each other safer

(20:30) Are fears about synthetic data overblown?

(22:32) When hallucinations are actually helpful

(28:16) How human intelligence could hold AI back

(34:52) The metacognition tension

(39:13) The OpenAI-Hugging Face hack

(46:15) Would narrower AI goals have put us in a better spot?

(50:30) Richard’s debate with Yoshua Bengio: are we building a dominant species?

(54:29) What breakthrough will finally win over AI skeptics?

(59:26) The case for widely available superintelligence

(1:04:20) What Richard would do if he had an unlimited budget to invest in AI

Learn more about your ad choices. Visit podcastchoices.com/adchoices