Exploring the Impact of Superhuman AI on Human Decision-Making: An Analysis of 5.8 Million Move Decisions in Go

Artificial intelligence that is superhuman can help improve human decision making

How will superhuman artificial Intelligence (AI) impact human decision-making processes? What will be the mechanisms that underlie this effect? These questions are addressed in a domain, where AI has already surpassed human performance. We analyze more than 5,8 million moves made by professional Go player over the last 71 years (1950-2021). In order to answer the first question we used a superhuman AI programme that generated 58 billion counterfactual patterns of game and compared the win rates between actual human decisions and those of counterfactual artificial decisions. After the introduction of superhuman artificial intelligence, humans started to make better decisions. After examining human players’ strategies over time, we find that novel moves (i.e. previously unknown moves) are more frequent and associated with better decision quality since the advent of superhuman artificial intelligence. Our findings suggest the advent of superhuman AI may have encouraged human players to abandon traditional strategies, and to experiment with novel moves. This may have led to improved decision-making.