Sociology: ‘Democratic AI’ makes popular decisions about how to distribute resources

Artificial intelligence algorithms may be able to find new mechanisms for distributing resources between people, according to a proof-of-concept study published in Nature Human Behaviour.

How proceeds should be distributed when humans collaborate has divided opinions among philosophers, economists and political scientists for years.

Christopher Summerfield and colleagues trained an artificial intelligence (AI) system, which they named ‘Democratic AI’, to design a new redistribution mechanism for public goods. The authors began by asking thousands of volunteers to play an investment game in groups of four. In the game, players receive different amounts of money, and have to decide whether to keep it for their own benefit, or share it for the good of the group, with shared funds distributed back to players with interest. The AI was trained to find a popular policy for redistributing funds back to players, as revealed by the votes that human players offered when asked which policy to play with again. The AI system was able to identify a policy that people were more likely to vote for than baseline policies, such as redistributing the funds equally, or returning funds to each player in proportion to their contribution. When other human volunteers were asked to play the role of redistributor, their strategies were less popular than those of the AI.

Although this study focused on one particular version of a four-player Public Goods Game, the authors suggest that future research could extend the idea of Democratic AI and examine whether it is effective in other scenarios with larger groups and more complex games.