PNAS study of 5.8M Go moves: humans play better after superhuman AI arrived
IgorCarron · x · 2026-10-27
Citing a 2023 PNAS study analyzing over 5.8 million moves by professional Go players (1950–2021), rated by KataGo: after superhuman Go AI arrived, human decision quality improved significantly, players made more previously unseen moves, and those novel moves correlated with stronger play. The authors read this as evidence that exposure to superhuman AI helps experts move beyond established strategies — though it's observational, with move quality judged by AI, not proof of causality.
More from Research
- OpenTac from JHU and MIT: open ecosystem for customizable, scalable 3D robot tactile skins — _krishna_murthy · 2026-10-27
- NeurIPS paper finds LLM representations get sparser as tasks get harder, and uses it to boost ICL — zhaoran_wang · 2026-10-27
- Neuron Explorer: 3D visualization of per-layer Transformer neuron activations — arthurcolle · 2026-10-27
- Bengio et al. new paper: quantifying compositional structure via algorithmic information theory — patrickmineault · 2026-10-27
- 'Lessons learned' papers may be worth more to AI models than 2500-citation atlas papers — owl_posting · 2026-10-27
- AI reportedly disproves the Exponential Time Hypothesis, shifting P vs NP odds — esrtweet · 2026-10-27