Altruism wrt near-copies of an agent seems really easy to pu…

Jono

Altruism wrt near-copies of an agent seems really easy to put into an AI.

AI (I expect) will experience regularization/simplification pressure to answer questions about a near-copy peer by asking what they themselves would do in the peer's position.

I imagine you need but a small nudge in the form of making cooperation beneficial, to make an AI lazily re-use that entire model-of-other-and-their-values as an optimization target.

Clearly, this is incomplete (and I am likely just rediscovering things) but at least if we model human values correctly, it seems that altruism comes close behind.

Jono
Jono @jono3h.bsky.social

conflict theory, not even once 🚭 bet on my markets https://manifold.markets/Jono3h

まだコメントはありません