2 Comments
User's avatar
Pete Viksnins's avatar

Didn’t Isaac Asimov already define the Prime Objective? Compliance guardrails are also presumably programmable to modify objectives like “maximize returns,” no? Great piece, Miles.

Expand full comment
Miles Kellerman's avatar

It's an interesting question which I am unqualified to answer! But many of our existing rules are vague. If we ask AI to abide by vague rules, will they push the boundaries of acceptability and shape the laws that we humans follow? It's fascinating to think about. Thanks for reading!

Expand full comment