Humanoid Robotics: The emperor's new brain and the AlphaGo moment - four predictions shaping the next three years

Bernstein's report on humanoid robotics adopts a Bayesian approach to update the probability of hypotheses as new evidence emerges, yielding four predictions shaping the next three years. MODEL: the emperor's new brain — the World Model paradigm, especially the World Action Model (WAM) that imagines both actions and their effects on the environment, will be widely adopted, though visual realism does not equal physical correctness (the best 2026 World Models score only 22% on comprehensive physical understanding); robot brains will start to remember, and multi-modality (third-party/egocentric/palm video, human and robot action, force and tactile sensing) will be fused into the brain. DATA: the "AlphaGo moment" — manipulation is harder than locomotion because effective reward functions are elusive; solving the reinforcement learning problem for manipulation could push success rates from 60% to 99%, though breakthroughs may be a few years away;