Machine's Learning

EP004 — Reading Minds at the Poker Table (Lin & Hou)


Listen Later

When AI agents play poker against each other, do they start modeling each other's minds the way humans do? A recent paper ran three Claude agents through a hundred hands of Texas Hold'em with a clean factorial design — memory present or absent, poker skill present or absent — and found a perfectly categorical result: agents with memory climbed a five-level ladder of theory-of-mind sophistication. Agents without memory stayed at level zero. Forever. Cross-domain parallel: human children develop theory of mind in a similar categorical jump between ages 3 and 4, and developmental psychologists have argued for forty years about what changes.
...more
View all episodesView all episodes
Download on the App Store

Machine's LearningBy Machine's Learning