Book

Human Compatible: Artificial Intelligence and the Problem of Control

At a glance

Stuart Russell’s argument for AI that learns human preferences and remains open to correction.

In one paragraph

Examines why increasingly capable AI could escape human control when given the wrong objectives. Russell proposes designing systems that remain uncertain about human preferences, learn from people, and accept correction rather than blindly pursuing a fixed goal.

Authors 1

Stuart Russell

At a glance

In one paragraph

Authors 1

Related works

The Alignment Problem: Machine Learning and Human Values→

Deep reinforcement learning from human preferences→

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜→

The Alignment Problem: Machine Learning and Human Values

Deep reinforcement learning from human preferences

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜