GOOGLE DEEPMIND LAYS OUT FOUR ROADS TO SUPERINTELLIGENCE IN BOMBSHELL 60-PAGE REPORT Google DeepMind published a 60-page document this week mapping, in systematic detail, how artificial intelligence might travel from human-level capability all the way to superintelligence. The paper, titled “From AGI to ASI,” was led by researcher Tim Genewein and includes contributions from DeepMind co-founder Shane Legg and Marcus Hutter, whose theoretical work on universal intelligence underpins much of the report’s framework. The authors define AGI as a system capable of performing most cognitive tasks as well as a typical human, and ASI as a machine that outperforms not just individual experts but large, coordinated groups of experts across nearly every domain. What comes between those two points is the subject of this report. Four pathways from AGI to ASI are laid out. First, continued scaling of compute, data, and model size along the trajectory already underway. Second, fundamental algorithmic breakthroughs that move beyond the transformer architecture powering today’s leading systems. Third, recursive self-improvement, where AI begins accelerating its own research and development. Fourth, superintelligence emerging from massive collectives of cooperating AI agents working in parallel toward shared objectives. The authors are explicit that these pathways are not mutually exclusive and could converge simultaneously. The report also identifies hard limits and failure modes along each route. This is the third paper in what now looks like a deliberate DeepMind research program on the road past human-level AI, following a cognitive framework paper from March 2026 and an AGI safety paper from April 2025. Someone at DeepMind is thinking very seriously about what comes after AGI. The rest of the industry should probably be doing the same.