Shane Legg

Sources#

• From AGI to ASI

Summary#

Shane Legg is a co-founder of DeepMind and a long-standing theorist of machine intelligence. With Marcus Hutter (his PhD supervisor) he formalized the Legg–Hutter score — a definition of intelligence as an agent's expected performance averaged over all complexity-weighted computable tasks (Legg & Hutter 2007a; Legg's 2008 thesis Machine Super Intelligence). In this corpus he is the senior (final) author of the "From AGI to ASI" report (June 2026), the DeepMind document that opens the theory-of-superintelligence cluster in this wiki.

Role in the corpus#

Legg's intellectual fingerprint is on the report's foundational move: grounding informal AGI/ASI talk in the smooth Legg–Hutter intelligence continuum, so the authors can avoid sharp capability thresholds and instead reason about the gap between AGI and ASI. His framing — that there is a measurable continuum with Universal AI/AIXI as its incomputable endpoint — is what lets the report bound ASI from above with theory while extrapolating from below with today's systems.

Connections#

• Universal AI (AIXI) — co-originator of the universal intelligence measure that AIXI maximizes

• Artificial Superintelligence (ASI) — the continuum framing that makes the report's coarse AGI/ASI definitions workable is his

• Marcus Hutter — co-author of the Legg–Hutter score; the two anchor the Universal AI framework together

• Google DeepMind — co-founder; the lab behind the report

• AGI-to-ASI Pathways — senior author of the report that lays out the four pathways

• Safety Commitments That Cannot Bind the Actor Who States Them — what the alignment-scoping assumption actually costs the report — a friction it concedes in the same paragraph is missing from its friction table, biting on pathway 3 — and the finding that this corpus records no Legg timeline claim to test the question's premise against

Open Questions#

• The report assumes alignment is "solved to a sufficient degree" to focus on trajectories — how does Legg's AGI-timelines optimism square with that scoping choice? Partially answered (2026-08-19): Safety Commitments That Cannot Bind the Actor Who States Them answers the scoping half and corrects the premise. The corpus records no Legg timeline claim at all — the optimism is imported from outside the wiki and cannot be tested here (the only DeepMind-side short-timeline claim on record is Hassabis's "probably only a few short years away," on Frontier AI Standards Body), and the report itself declines to give one ("instead of focusing on one technological trajectory and timeline"). What is settled: the assumption is internally inconsistent with the report's own concession, in the same paragraph, that "alignment difficulties may act at least to some degree as a direct bottleneck to capability development itself" — a friction by the report's own definition, absent from Table 4, and it gates pathway 3 specifically, the one pathway with no historic data to fit. Retag toward #oq/source if a source in Legg's own voice on timelines is ingested.

Sources#

• From AGI to ASI — final author; Legg (2008), Legg & Hutter (2007a) cited as the basis for the intelligence measure