Reinforcement Learning the Key to AI?

[Arthur Hanson]

Reinforcement learning looks like it may be one of the keys to AI. The ability of AI to actually learn on its own when combined with the speed and endurance of the computational power now and in the future presents opportunities and dangers man has never faced before. Below is an excellent link to an explanation of reinforcement learning and some other methods. The art of learning and how to quantify it in a method that makes it useful is coming closer every day. The big question is if and when an artificial intelligence can become sentient. In my financial trading, I devised a method of competing with organizations that have access to high power computers and communications that seems to be working well. I don't think we will end up with a single singularity, but a number of them. If we did end up with a single singularity, that would definitely present some serious problems. Comments, thoughts and ideas on this solicited and welcome, the more input the better.

Addition, In my trading I made the assumption that the various programs would take on various traits, just like different people have different personalities. If all the AIs turn out identical and linked, we would definitely have a serious issue in how to interface with it or it might choose how it interfaces with us and we might like or not like it.

Reinforcement learning - Wikipedia