Saturday, 19 April 2008

Artificial Intelligence: Cybernetics and Learning

This is one way of linking research and teaching. I was working on my reading for my Diploma in Learning and Teaching and thinking about the concepts of deep and surface learners. I also had Ashby's cybernetics running around in my head, and also some of the ideas of control theory.

So how can we make a thinking robot? This would be real artificial intelligence. The problem we have with computers at the minute is that they do what we program. They are the perfect surface learners - they reproduce everything and so long as we can conceive an idea they can do it. They can learn to a degree by seeing patterns, many of them novel that we do not see, or by giving weights to repeated rules. But under it all they are like a control system that needs each possible state defining. They need far more rules and far more description so that they can find some experience in their programming to relate to the question that is posed. They cannot wonder and they cannot use intuition.

Fundamentally from a cybernetics point of view they need a control system that describes all possible inputs and their related outputs. In systems theory they are a very poor bow-tie model as the knot is as big as the bows.

No comments: