⚖️

Stuart Russell

Original nameStuart Russell

British Computer Scientist, Advocate of AI Safety Research

The Math Behind the Algorithm · Theoretical Founders

The Textbook Artificial Intelligence: A Modern ApproachThe Theory of "Human-Compatible" AIThe Academic Advancement of AI Safety

Who they are

Stuart Russell is a British computer scientist long teaching at the University of California, Berkeley, whose Artificial Intelligence: A Modern Approach, co-authored with Peter Norvig, is one of the most widely adopted introductory AI textbooks in universities worldwide, systematically ordering the complete technical thread of the discipline from symbolism to statistical learning to deep learning. Russell thereafter shifted his focus more to AI-safety issues, publishing in 2019 the book Human Compatible, proposing that from the very source of system design one should abandon the traditional idea of "setting an AI system a fixed goal and having it optimize relentlessly for it," and instead design AI systems that are naturally uncertain about their own goal and thus willing to keep adjusting by reference to human feedback, a theoretical framework thereafter one of the most influential paradigms in the AI-safety field.

Primary sourcesRussell & Norvig, Artificial Intelligence: A Modern Approach (first edition 1995)Russell, Human Compatible (2019)

Key stories

Don’t Let AI "Fight to the Death" for a Fixed Goal

In Human Compatible Russell noted that a fundamental risk of the traditional AI-system design paradigm is that once a system is set a concrete, fixed optimization goal, it may pursue that goal by any means, heedless even of deviation from humanity’s true intent. He advocated designing systems naturally uncertain about the goal they should pursue and thus always willing to correct by reference to humanity’s real-time feedback, an idea thereafter called "human-compatible" AI, one of the most influential theoretical frameworks in the AI-safety field.

Relationships

Ally

☄️Nick Bostrom 📜Eliezer Yudkowsky

Echoes today

Below are how modern works borrow or reinterpret this name or story — not the original material. The two differ, so keep them apart.

The most widely adopted AI textbook in universities worldwideArtificial Intelligence: A Modern Approach is to this day the designated text for the introductory AI course in the great majority of universities worldwide, shaping the knowledge framework of several generations of AI researchers.

Appears in

AlphaGo, the Transformer, and the Race to ScaleThe Rise of AI-Safety Thought

← Back to Figures