View Single Post
hypnagogue is offline
May8-04, 03:31 AM
Sci Advisor
PF Gold
P: 2,264
If one claims that it is theoretically impossible for AI to emulate human intelligence, one is essentially making the strong and undesirable claim that the laws of physics cannot fully describe the functional workings of the brain. If we (quite plausibly) suppose that the laws of physics can, in fact, fully describe the functional workings of the brain, that implies that there is a completely formal description of the brain's computational processes, which implies that any properly constructed computer could implement this formal description and thus be intelligent in the same sense as a human.

Consciousness is a dodgier issue, but intelligence as such is a completely functional concept. While there are deep and apparently insoluable epistemological/ontological issues when it comes to attributing consciousness to another being, there are no such problems in attributing intelligence (at least to the extent that 'intelligence' is well defined, which in most cases it admittedly isn't). If I give an essay question to a computer and a human and they write equally good responses on the basis of some sort of reasoning out of the question (rather than having the computer spitting out a pre-formulated answer or somesuch), it automatically follows that their responses have been equally as intelligent. The intelligence is embodied in the process of 'thought' (meant here only in the purely computational sense) and ensuing action, and in principle doesn't need to refer to any epistemologically troublesome phenomena such as consciousness.

Not to say developing truly advanced and general application AI will be easy-- anything but! Human brains are easily the most complex sort of systems that are currently known to exist, so emulating their functional properties is a tall order. But there doesn't seem to be any in principle limits to such a task, only practical ones.

As a side note, expert systems typically perform at a high level in their narrow domains of application, but typically do not consistently out-perform the very best human experts. See for some further discussion.