https://www.nature.com/articles/s41562-022-01383-x.pdfi know ppl on this thread might have a poor opinion of social choice theory or ml, but deepmind came with a paper a few weeks about applying neural networks to mechanism design. it's lib bullshit from otherwise smart ppl but you be the judge
ABSTRACT:
Building artificial intelligence (AI) that aligns with human values is an unsolved problem. Here we developed a human-in-the-loop
research pipeline called Democratic AI, in which reinforcement learning is used to design a social mechanism that humans prefer by majority. A large group of humans played an online investment game that involved deciding whether to keep a monetary endowment or to share it with others for collective benefit. Shared revenue was returned to players under two different redistribution mechanisms, one designed by the AI and the other by humans. The AI discovered a mechanism that redressed initial wealth imbalance, sanctioned free riders and successfully won the majority vote. By optimizing for human preferences, Democratic AI offers a proof of concept for value-aligned policy innovation.
Result: Our AI system designed a mechanism for redistribution that was more popular than that implemented by human players. This is especially interesting because unlike our agent, human referees could integrate information over multiple timesteps to reward or sanction players on the basis of their past behaviour.