In the deepest sense, the AI champions see their critics as trying to reverse the triumph of the Enlightenment, with its promise that man’s mind can understand everything, and as retreating to an obscurantist, religious outlook on the world. They see humanity as having to choose, right now, between accepting the possibility, if not the actual existence, of thinking machines and sinking back into the Dark Ages. But these are not our only alternatives; there is a third way, the way of agnosticism, which means accepting the fact that we have not yet achieved artificial intelligence, and have no idea if we ever will.

Much more interesting is to ask whether there's an inherent flaw in the computing paradigm of thought, and whether another rational, scientific view of what thinking is could lead to success in the AI programme.

To say that thinking involves information processing but is not entirely about information may start us in the right direction. When you look at biological systems such as beetles or human beings, for instance, you find that the purpose of the brain is to move their bodies around. Everything that the brain does is anchored not by the requirements of some symbolic processing system, but by the need to move the body around. What anchors intelligence, in other words, is its manifestation in a specific physical system. From this point of view, we can define intelligence as a system's ability to couple into its surrounding environment in an efficient, flexible and responsive manner. Whether there are symbolic algorithms or reflexive systems lying behind its behaviour is beside the point. If we treat symbolic representation and internal calculation as being weakly-coupled (but still semi-physical) linkages to the external world, then questions of what representation is and what semantics means can be seen to be tied intimately to questions of what the physical world itself is. And only when we reach that point can we start to answer the question of what intelligence and thought are.

This territory is covered thoroughly by Brian Cantwell Smith in On the Origin of Objects. Briefly put, though, his assertion is that you can't talk about computing (or, by extension, computation in general or thinking or consciousness) without confronting metaphysical issues--specifically issues of what the world is made of. Failure to confront such issues leaves you with crippled computer science, and crippled AI studies, because you perpetuate a nonexistent divide between objects and reference to objects. --To put it another way, a divide between the physical and the mental. In contrast, a theory that explains how physically discrete objects can arise and interact in the first place might also explain represenation, semantics and, ultimately, intelligence itself.

It isn't a step backward to re-envision the project of AI from the point of view of a new metaphysics. Doing so may, in fact, be the only way forward.

<  |  >