Marcus Hutter

Sources#

• From AGI to ASI

Summary#

Marcus Hutter is the originator of AIXI and the Universal AI framework — the formal, mathematically optimal model of a general agent (Hutter 2005, Universal Artificial Intelligence). He is a senior researcher at DeepMind and professor at the Australian National University, co-author with Shane Legg of the Legg–Hutter intelligence measure, and lead author of the 2024 textbook An Introduction to Universal Artificial Intelligence (Hutter et al. 2024), cited throughout the "From AGI to ASI" report as the authoritative reference. He is a co-author of that report.

Role in the corpus#

Hutter supplies the report's theoretical backbone. AIXI — an agent optimal on average over all computable environments under Solomonoff's universal prior — is what lets the report treat ASI as a region on a continuum approaching a well-understood limit, rather than an unbounded mystery. His framework also delivers the report's hard-limits message (Fundamental Limits of ASI): AIXI formalizes maximal data efficiency and inherits, via Kolmogorov's structure function, the result that an agent's own approximate (lossy-compression) performance can be fundamentally unpredictable. He has separately written on post-labor economics under AGI (Hutter 2026), cited in the report's discussion of deliberate slowdown, and on digital intelligences inhabiting compute-based virtual worlds (Hutter 2012).

Connections#

• Universal AI (AIXI) — creator of AIXI and the Universal AI framework; author of its authoritative textbook
• Fundamental Limits of ASI — AIXI formalizes the data-efficiency limit and the lossy-compression unpredictability result
• Shane Legg — co-author of the Legg–Hutter intelligence measure; together they anchor Universal AI
• Google DeepMind — senior researcher; co-author of the report
• Artificial Superintelligence (ASI) — UAI/AIXI is the incomputable endpoint above ASI on his and Legg's continuum

Open Questions#

• AIXI is incomputable and non-embedded; how far do recent fixes (amortized predictors, embedded/multi-agent AIXI) carry the theory toward practical relevance for real ASI?

Sources#

• From AGI to ASI — co-author; Hutter (2005), Hutter et al. (2024), Hutter (2012, 2026) cited