AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › People & Organizations

Shane Legg

medium confidence · updated 2026-09-18

Co-founder and Chief AGI Scientist of Google DeepMind. Originator, with Marcus Hutter, of the Legg-Hutter measure of universal intelligence, which formalizes intelligence as average performance across all computable tasks; a senior author on DeepMind's 2026 From AGI to ASI report.

Contents

Co-founder of DeepMind and its Chief AGI Scientist. See Google DeepMind. Since September 2026 he has also been a director and the managing editor of the DeepMind Institute, Google and Google DeepMind's publishing forum on artificial general intelligence (Source: axios.com).

Background

According to his biography on the DeepMind Institute's site, Legg studied for his PhD at the Dalle Molle Institute for Artificial Intelligence Research (IDSIA) in Switzerland, with a thesis proposing a formal definition of machine intelligence that received the Canadian Singularity Institute research prize. He then spent a year as a postdoctoral researcher at the Swiss Finance Institute building models of human decision-making and two years at the Gatsby Computational Neuroscience Unit at University College London. He is a Distinguished Alumnus of the University of Waikato and an Honorary Fellow of UCL, received the Digital Masters Award for Excellence in Technology in 2017, and was appointed Commander of the Order of the British Empire (CBE) in the 2019 Birthday Honours for services to investment and the science and technology sectors (Introducing the DeepMind Institute (Legg, Manyika and Hassabis, September 2026)).

DeepMind Institute

Legg is managing editor of the DeepMind Institute, launched September 16, 2026, and one of its three directors with James Manyika and Demis Hassabis. The three co-wrote its launch essay, which states that AI is approaching AGI, "a system that exhibits all the cognitive capabilities of the human brain," that remaining gaps are expected "to be closed soon," and that the risks include "the potential for loss of control in future self-improving systems" (Introducing the DeepMind Institute (Legg, Manyika and Hassabis, September 2026)). The statement describes the Institute as a platform for researchers from Google DeepMind, Google and the wider research community who "will not always agree," and states that answers about AGI "shouldn't be technologists alone who provide them" (Introducing the DeepMind Institute (Legg, Manyika and Hassabis, September 2026)). At launch he told Axios: "AI is going to drive all kinds of advances in science, and science is an international thing. If some scientific breakthrough happens in China or it happens in Japan or the U.S. or wherever — this is a global community advancing the frontier in science" (Source: axios.com).

The Legg-Hutter measure

Legg's doctoral work with Marcus Hutter produced a formal definition of machine intelligence as "the average performance of an agent across all computable tasks," weighted so that simpler tasks (lower Kolmogorov complexity) count for more. The measure is not computable in practice; its purpose is to supply a well-defined limit against which informal notions of general intelligence can be positioned, and it subsumes many of those informal notions as special cases.

Its policy relevance is that it makes intelligence a continuum rather than a threshold. DeepMind's From AGI to ASI (June 2026), on which Legg is a senior author, uses it for exactly this purpose: because the measure is smooth, "we do not need to very precisely define the Legg-Hutter score threshold of AGI and ASI," only to establish that a significant gap separates them. The report's third definitional tier, Universal AI, is the theoretical maximum of the measure, formalized through the AIXI agent of Legg's co-author Hutter.

Chain-of-thought monitorability

Legg is one of the 41 authors of the July 2025 position paper "Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety" (Korbak et al.), which recommended that frontier developers "consider the impact of development decisions on CoT monitorability" (Source: arxiv.org). Two other co-authors of that paper, Rohin Shah and Anca Dragan, wrote the DeepMind Institute's inaugural essay on the subject, published under Legg's editorship (The Case for Reasoning Transparency (Shah and Dragan, DeepMind Institute, September 2026)). See Chain-of-Thought Monitorability.

Positions

Legg has been associated with the view that AGI is a concrete engineering target on a decadal rather than indefinite horizon, and that its arrival should be planned for rather than debated in the abstract. From AGI to ASI reflects that posture in its structure: it brackets the question of when AGI arrives and asks what follows, treating the four pathways it maps — scaling, algorithmic paradigm shifts, recursive improvement, and multi-agent collectives — as parallel possibilities rather than competing predictions, and treating the impact of the frictions along them as open research questions rather than settled parameters (From AGI to ASI (Google DeepMind, June 2026)).

In a Financial Times interview at the Institute's launch, as reported by The Next Web, Legg said AI capabilities must not outrun safety controls, called it premature to declare that AGI had been achieved (following claims to that effect by OpenAI on September 4 and by Nvidia's Jensen Huang on September 7, 2026), said he remains "comfortable" with his forecast of a 50% chance of "minimal" AGI by 2028, and described Anthropic chief executive Dario Amodei's call to slow but not pause frontier releases as "interesting directionally" and "worth considering" (Source: thenextweb.com). See We Must Pace the Frontier (Amodei, September 2026).

Relationships

Categories: People & Organizations | technical | academic | company | frontier-models | global
What links here (15)