█

AI/TLDR

DeepMind Institute · 2026-11-19 · notable

What Comes After AGI? — DeepMind researcher maps four paths to ASI

In short

Tim Genewein of Google DeepMind argues that AI progress may not stop at human level. His essay names four paths to superintelligence and estimates the world's effective compute grows at least 10x a year.

DeepMind Institute · 2026-11-19 · 3 sources checked

Diagram of the continuum of general intelligence, from sub-AGI to human scale, superintelligence and a theoretical limit

A Google DeepMind researcher argues human-level AI is one point on a continuum, and lays out how AI could pass it.

What is it?

"What Comes After AGI?" is a November 19, 2026 essay by Google DeepMind research scientist Tim Genewein on the DeepMind Institute, a platform started by Google and Google DeepMind researchers. The essay is based on the technical report "From AGI to ASI" (with co-authors including Shane Legg and Marcus Hutter), and the site says it "should not be read as Google's official view".

How does it work?

Genewein treats intelligence as a continuum, with human-level AGI as one point far below the theoretical limit. Multiplying compute per dollar, AI hardware spending and algorithmic efficiency, he estimates global effective compute grows at least 10x per year. Four pathways could then push past human level: scaling models and data, new technical paradigms, better coordination between many agents, and recursive self-improvement.

Why does it matter?

The essay argues the four pathways reinforce each other, so limiting only one may not stop progress and could cause sudden jumps later. Its advice is "model, measure, forecast, and repeat", which it says needs frontier-model access, large evaluation compute, new benchmarks and work across labs and institutions.

Who is it for?

AI policy and safety researchers, forecasters

Sources · 3 outlets

Tags

  • google-deepmind
  • deepmind-institute
  • agi
  • asi
  • superintelligence
  • recursive-self-improvement
  • ai-forecasting
  • ai-safety

← All releases · Learn AI