A.I.

Steve Grand, the creator of Creatures, and an independent AI researcher in his own right, has written an article in the IEEE Intelligent Systems magazine called Moving AI out of its Infancy: Changing our preconceptions. I find his ideas interesting, but not as groundbreaking or revolutionary as he imagines.

He dismisses traditional AI and connectionist methods as unsuccessful and based on false assumptions. A more or less fair assesment perhaps, but I think he has misunderstood the nature of what he calls "New AI", specifically claiming that it is influenced by the nervous systems of the simplest invertebrates. He is in search of what he calls the "periodic table" of AI, some radical new paradigm of intelligence that will transform AI from alchemy to chemistry, so to speak. I have serious issues with this. First, he bases his methods on being able to simulate the human brain, a very anthropocentric approach that may not be the best idea, given the tools of computation we have now and how little we understand (or even think we understand) the human brain.

Second, i think the view that intelligence must be based on one simple and easily described paradigm, that we just need to find and apply, is wrong. Researchers in Computational Brain Theory seem to be coming round to the view that the intelligence of the brain doesnt follow from one particular model of computation/cognition (say a hopfield network) but is an emergent behaviour of some pretty complex sub-systems which are functionally and operationally quite different.(More on this in a later post.) Intelligence must be engineered, using many different ideas, and we will most likely approach it asymptotically, like evolution did.

The rest of the article mentions some of the insights that he has, and I think a couple are worth mentioning, "Brains dont make decisions" and "Brains preform coordinate transforms". But I think these are just interesting alternate viewpoints on the problem itself and not necessarily constructive steps to a solution. I was disappointed that he seems to be focusing on questions that are exclusively in the realm of robotics. What about common sense, learning etc. ? Oh Well.