Richard Ngo

I got this idea from the Moral Parliaments paper, which identifies that proportional chances voting incentivizes bargaining between voters. But they propose lying to voters and picking the most popular option instead. A classic utilitarian type of mistake! https://t.co/upFS8GZ2Zm

In the active inference paradigm, goals are *distributions* rather than utility functions. This is kinda weird—e.g. it means you don't always pick the best option. Anyone got a high-level intuition for why this makes sense?