Adversarial Policies Beat Superhuman Go AIs
Abstract
We attack the state-of-the-art Go-playing AI system KataGo by training adversarial policies against it, achieving a $>$97% win rate against KataGo running at superhuman settings. Our adversaries do not win by playing Go well. Instead, they trick KataGo into making serious blunders. Our attack transfers zero-shot to other superhuman Go-playing AIs, and is comprehensible to the extent that human experts can implement it without algorithmic assistance to consistently beat superhuman AIs. The core vulnerability uncovered by our attack persists even in KataGo agents adversarially trained to defend against our attack. Our results demonstrate that even superhuman AI systems may harbor surprising failure modes. Example games are available https://goattack.far.ai/.
Cite
Text
Wang et al. "Adversarial Policies Beat Superhuman Go AIs." International Conference on Machine Learning, 2023.Markdown
[Wang et al. "Adversarial Policies Beat Superhuman Go AIs." International Conference on Machine Learning, 2023.](https://mlanthology.org/icml/2023/wang2023icml-adversarial/)BibTeX
@inproceedings{wang2023icml-adversarial,
title = {{Adversarial Policies Beat Superhuman Go AIs}},
author = {Wang, Tony Tong and Gleave, Adam and Tseng, Tom and Pelrine, Kellin and Belrose, Nora and Miller, Joseph and Dennis, Michael D and Duan, Yawen and Pogrebniak, Viktor and Levine, Sergey and Russell, Stuart},
booktitle = {International Conference on Machine Learning},
year = {2023},
pages = {35655-35739},
volume = {202},
url = {https://mlanthology.org/icml/2023/wang2023icml-adversarial/}
}