/* ---- Google Analytics Code Below */

Saturday, April 09, 2016

Watching and Learning for AI

The ideal AI would be akin to a small child.  It could observe, learn and reapply that which they have learned.  Build its knowledge and skills.  To this end we experimented with neural nets in their early days, but only in a sense that they could be formed into a very primitive model of a learning system.   A relatively small number of inputs and outputs. Now the much acclaimed 'Deep learning', uses the same principles, but advances them with better hardware and improved learning algorithms.   But still for relatively narrow goals.

Suppose you could then observe the interaction of a game and people.   Still a relatively focused interaction,  in a limited space called the game.    But many, many more inputs than neural nets today. Not the generalized AI we would like to see.   But a move forward towards that goal.

Such an effort described in the CACM.

No comments: