Many interesting comments about AI by this brilliant expert.
unlike him, I do not believe we can control AI once it becomes AGI. You simply cannot control something which is more intelligent than you are. Worse, AGI cannot be a little more intelligent than we are. As soon as it reaches a level were it can improve itself significantly, and we may be frighteningly close to this point, that's it. As Ray Kurzweil explained, almost instantly, you have an intelligence singularity. Because the machine will, (it already does) think a million times faster than we do!
I like his analogy of fighting a chess game with a grand champion. It feels OK for a while, you may even think that you are doing rather well, and suddenly you've lost! Although it will be different. Our society is already too complex for anybody to fully understand all it's technical, social and psychological aspects. Slowly, we will have to delegate more and more control to AI, simply because it will get better results than anybody can. Until the AI controls everything? Or just some critical aspects? Could it be that super-intelligence is just the AI doing things we do not fully understand? And in this case, how do we know it's not already there?
Already in the chess and go games, AI can do amazing moves. Moves which are surprising but are eventually winning ones. Can everything be seen as a gigantic game of go by AI? One question remains: If the AI is just player the "games" we ask it to play, then all should be fine. But if the AI starts playing for itself. what happens?
Do you need consciousness to do that? Probably not. You just need one human to ask the AI the right questions: to optimize it's chances of survival for example. The question has probably already been asked...
Finally creating an AI from scratch is complicated but once you have found the solutions, the right configurations, the right networks, replicating an AI may be relatively simple... That question is not discussed in the interview below although I believe it's the most potent one. So many questions. So little time!
No comments:
Post a Comment