Google DeepMind Launches an Institute to Argue About AGI in Public

Google DeepMind has launched the DeepMind Institute (DMI), a new platform built specifically for interdisciplinary debate over artificial general intelligence — systems with the full range of human cognitive capabilities — and what its arrival means. It’s directed by DeepMind co-founder Shane Legg as managing/chief editor, alongside Google executive James Manyika and Google DeepMind CEO Demis Hassabis. Four founding essays anchor the launch: reasoning transparency (Rohin Shah and Anca Dragan), economic policy for AGI (Julian Jacobs and Alex Imas), a case for “new utopianism” (Stephen Cave), and Hassabis’s own framework for frontier AI. DMI is explicitly positioned as a conversation-starter, not an official Google policy stance — a venue meant to pull in researchers, policymakers, and voices from the humanities and arts, not just DeepMind’s own staff.

The timing matters because the industry doesn’t agree on whether AGI has already shown up. Forrester declared in September that “stage one” AGI arrived a year ahead of its own 2025 forecast — based on OpenAI’s Astra system jumping from roughly 1% accuracy to 62.7%, then 99.9% with an optimized memory harness, on the ARC-AGI-3 benchmark. Forrester’s own criteria score Astra as passing on tool-building, barely passing on self-learning, and weakest on collaboration, and the firm is already telling enterprise planners to retire “frontier model” language for “competent AGI” while building contingency plans for what it calls “stage two.”

Read together, the two moves describe the same moment from opposite ends: Forrester is trying to pin down AGI with a scoring rubric that keeps moving; DeepMind is opening a forum explicitly because it doesn’t think the definition, or the policy response, is settled yet. For anyone advising clients on AI capability roadmaps, that’s the real signal — the labs and the analysts building the yardsticks are still arguing about what’s being measured, which means capability claims built on either framework alone are provisional, not settled ground.