Re:It looks like you’re trying to take over the world
by FeepingCreature on Sunday March 27, 2022 @01:19PM
(#30821991)
No, the Clippy thought experiment asks the question: "might we accidentally build an AI that kills us all?"
It's not a very complicated thought experiment, you know. Not hard to interpret.
The actions you point out (environmental destruction, short-sighted carbon policies) optimize very clearly for an individualistic reward; it's in the best economical interest of a few to chop down forest / continue to burn coal / take your pick. And just as clippy can't seem to stop building paperclips while the world burns, we can't seem to stop optimizing for money/sex/domination while our future darkens.
But if you shift the reward function -- and it's not that huge of a shift -- from "win at life" to "long term survival of the species as a whole", these actions are very clearly unproductive.
So how do we change the reward score calculation?