4 ms·
It isn't hard to give an AI a goal but it is hard to do so safely. As a toy example we could design an AI who treated say, reducing carbon emissions as it's goa
by realbarack 7y ago
It isn't hard to give an AI a goal but it is hard to do so safely. As a toy example we could design an AI who treated say, reducing carbon emissions as it's goal just as you treat eating and sleeping as yours. The issue is that the sub goals to accomplish that top-level goal might contain things we didn't account for, say destroying carbon-emitting technology and/or the people that use it.
- etherealG 7y agoHuman's have many basic goals that are very dangerous when isolated in that way. It seems to me that nature didn't care (and of course, can't care) if it was dangerous at all when coming up with intelligence. Maybe we shouldn't either if we want to succeed with replicating it. Worrying about some apocalypse seems counterproductive to me.