---
title: "Stuart Russell"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Stuart Russell"
aliases: []
first_seen: "2026-10-19T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["bounded optimality","potential-based reward shaping","artificial intelligence","Andrew Y. Ng","CoDA (co-evolution of data engines and action policies)"]
seek_code_commit: "d40ee39"
---


UC Berkeley computer scientist and AI theorist — co-author of the standard textbook *Artificial Intelligence: A Modern Approach* and a leading voice on long-term AI safety, who has also made two separate load-bearing theoretical contributions this vault now tracks independently.

Matters to this vault on two counts that predate any connection between them. First, Russell (with Eric Horvitz) is a co-originator of [[entity-bounded-optimality|bounded optimality]], a theory for designing optimal programs for agents with performance-limited hardware, which [[claim-resource-rational-analysis-descends-from-bounded-optimality-not-werbos-adp|resource-rational analysis cites as its AI antecedent]]. Second, Russell co-authored (with Andrew Y. Ng and Daishi Harada, ICML 1999) the classical potential-based-reward-shaping invariance theorem — that a fixed, bounded potential with appropriate terminal-state handling preserves the policies optimal under the original task reward — which the 2026 robot-learning paper CoDA invokes as the theoretical basis for its "Text2potential" dense-reward mechanism ([[claim-coda-shared-memory-couples-data-engine-policy-optimality-preserving-reward-shaping]]). The two contributions sit in different vault clusters (rational-agent theory and RL reward design) twenty-seven years apart, connected only by Russell himself.

## References
- [[claim-resource-rational-analysis-descends-from-bounded-optimality-not-werbos-adp]]
- [[claim-coda-shared-memory-couples-data-engine-policy-optimality-preserving-reward-shaping]]
- [[claim-anytime-algorithms-continuously-connected-not-stent-premature]]
- Captures: 10-inbox/raw/2026-10-18-outside-arxiv261022362-coda-co-evolution-of-data-engines.md

## Log
- 2026-10-21: A 2026 LLM test-time-compute position paper names Russell & Subramanian's 1995 [[entity-bounded-optimality|bounded optimality]] and Russell & Wefald's 1991 "rational metareasoning" as direct theoretical ancestors of present-day test-time-compute allocation — the same bounded-optimality framework resource-rational analysis already cites, now also carried forward into AI-systems research rather than only cognitive science ([[claim-anytime-algorithms-continuously-connected-not-stent-premature]]).