AI Safety Fundamentals: Alignment cover image

AI Safety via Debate

AI Safety Fundamentals: Alignment

CHAPTER

The Effects of Debate on AI Alignment

The goal of debate is aligning ML systems more capable than humans at a particular task. Each debate is an independent game, but agents could use information from past debates to make better decisions. Since games are 0-some, this does not change optimal play as any equilibrium in an iterated 0-some game is weakly dominated by a strategy which treats each game as separate. If the honest agent wins against any liar in an experiment, we will have shown that honesty is the optimal strategy. This is particularly useful with human agents since humans may be intrinsically inclined to tell the truth in ways ML agents would not be.

00:00
Transcript
Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner