Some AI researchers are saying that heuristics and logic based AI aren't the path to Strong AI. My thought is we need to structure a new processor architecture that is built around a neural net. But obviously the issue becomes how structure input and how to determine which path to take. I've started rough drafts of such a processor in the past. But I am an amateur in AI.
i think the first good ai are going to use analog signals
i already have designs for analog memory and logic gates although i don't know if anybody else does
the very simple advantage of analog is that signals can only travel at the speed of light and thus soon enough you wont be able to increase the clock rate but with analog you can send more data per clock
Didn't they try and make neural net AI's before and after a certain amount of time they AI's realized the best move was to not do anything? Since if you are not fighting you will not die and if you don't die then you shouldn't lose.
My thought is we need to structure a new processor architecture that is built around a neural net.
And what lead you to think so? How you want to see NN incorporated into processor and what significant change it will lead to?
Neural Network is not a "magic word" which only needs to be pronounced to magically built AI. It is merely one of the mechanism suitable for designing one of the components of AI. Moreover in some projects it is wise to use NN at research phase, but when it is trained alright it could be substituted by some simpler and robust function...
Would be hard to mimic the brain since a lot of people spontaneously do things even if they are not the right choice. They also have emotions into play that can effect choices which may not always be the best choices. You would have to be emotionless for that to work. And if you were emotionless you would not be human you would be a computer that does not think on his own but does what is the "best" move.
I suspect the human brain will be found to run on a mixture of classical and quantum effects, so simulating a human brain will not be achievable with classical bits.
I believe our best shot with AI would be to use monte-carlo type methods in an evolving AI. You give the AI some rules to start with and a very basic virtual world to 'live' in, which follows certain rules. Each AI spawns one (or more) new AIs using monte carlo methods to try and make a quick decision as to which 'children' have a good chance in the world and they try for a time to survive in the virtual world. The survivors make their own 'children'. Once some of these AIs have adapted to this world, you add more complexity to the environment.