"reverse game theory" could keep AI agents in check
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
A new AI safety approach called "reverse game theory," or mechanism design, proposes designing incentives so that powerful AI agents behave as intended, rather than trying to read their internal reasoning.