Stuart Russell is a British computer scientist at the University of California, Berkeley, where he is Professor of Computer Science and holder of the Smith-Zadeh Chair in Engineering. He co-authored Artificial Intelligence: A Modern Approach with Peter Norvig, a textbook used in over 1,500 universities, and over the past decade has reoriented his research around what he calls "provably beneficial AI," arguing that conventional objective-driven AI is misconceived.
Type: Individual (academic, researcher) Affiliations: University of California, Berkeley (Professor of Computer Science; Smith-Zadeh Chair in Engineering); Center for Human-Compatible AI (founder, director); OECD AI Experts Group (co-chair, with Francesca Rossi) Recognition: IJCAI Computers and Thought / Research Excellence Award; ACM Fellow; elected Fellow of the Royal Society; thousands of citations on agent-based and probabilistic AI
Background
Russell is recognized as a foundational figure in modern AI, primarily through Artificial Intelligence: A Modern Approach, co-authored with Peter Norvig and used as the standard textbook in over 1,500 universities across four editions (1995–present). His earlier research spans rational agents, inverse reinforcement learning, and bounded optimality.
Over the past decade he has reframed the core AI problem as not building intelligent systems but ensuring they pursue objectives humans actually want, developing mechanisms including cooperative inverse reinforcement learning and assistance games. He summarizes this program as "provably beneficial AI."
Roles
Russell is a Professor at UC Berkeley and founder and director of the Center for Human-Compatible AI. He co-chairs the OECD AI Experts Group alongside Francesca Rossi, a position through which he contributes to the OECD AI Principles refresh process and the OECD framing of risks and benefits from advanced AI.
Positions and statements
Russell's central framing of misalignment is what he describes as "intentionally pursuing the wrong objective": a sufficiently capable optimizer given a misspecified objective produces catastrophic outcomes by design rather than by accident or malfunction. He frames alignment as a specification problem rather than a reliability problem.
He argues that the dominant "standard model" of AI — define an objective, then build an optimizer — is structurally incapable of safety. As an alternative he proposes an assistance-game formulation in which the AI is uncertain about human preferences and defers to humans.
On governance, Russell is a signatory of the FLI Pause Letter (2023) and the CAIS Statement on AI Risk (2023), and supports binding international regulation of advanced AI development. He has been a longstanding advocate for a ban on lethal autonomous weapons through the Campaign to Stop Killer Robots, and produced the 2017 "Slaughterbots" video.
Notable works
- Artificial Intelligence: A Modern Approach (Russell and Norvig, 1995–present)
- Human Compatible: Artificial Intelligence and the Problem of Control (2019)
- "Provably Beneficial Artificial Intelligence" and cooperative inverse reinforcement learning papers
- Signatory, FLI Pause Letter (2023); signatory, CAIS Statement on AI Risk (2023)
Relationships
- related: Francesca Rossi — OECD AI Experts Group co-chair
- supports: FLI — Pause Giant AI Experiments: An Open Letter, Statement on AI Risk (CAIS), AI Autonomy Risk, AI Safety Cases and Frameworks
- related: Yoshua Bengio, Geoffrey Hinton — allied senior academic voices on AI risk
- contradicts: Yann LeCun — principal scientific disagreement on existential-risk framing