Home Artists Posts Import Register

Content

I recorded a talk I gave at the AI and Politics meetup at Newspeak House. There's not really anything new here if you watch the channel, but I thought people might be interested.

This was also a test to see if I can get decent talk videos using only equipment I can easily carry with me.

Files

Talk: AI and Politics VII

I gave a talk about AI Safety at the "AI and Politics" meetup!

Comments

Mr. Duck

AIs will be very alien to us. Given that we still don't have bullet-proof 100% correct way to raise kids, or deal with mentally ill individuals (equivalent of alien AI), I see no clear way how we are not screwed in the end. On other hand, if the AI safety problem will be solved, I think it will be generally applicable to ourselves on how to build a better, safer society.

Anonymous

What if your AI utility function forbids it from lowering anyone else's utility function? It can have the 20 or so normal things that might exist in everyones utility function, but its actual rule would be to determine the utility function of anyone it comes into contact with (as best it can) and not take any action which would reduce that. It is not obliged to increase anyones utility function only not to reduce it.