Our Coming AI Dominion
Thomas Aquinas College published a ten-part YouTube series on AI, Mind and Machine by Michael Augros.
A quick thumbnail is that machines cannot be intelligent, if what we mean by intelligent is what humans do, because machines (and other human-made artifacts) can only originate transitive motions (parts working externally on each other), and not immanent motions or operations, and all truly cognitive acts are operations and intransitive. Only organisms, which are simple unities, can perform such immanent operations. In short, there is no true unity or "self" in an artifact (like there is in an intelligent organism) to operate immanently, and intelligence consists of immanent "inside" operations.
In the concluding episode, Augros discussed the difference in capability between human intelligence and machine quasi-intelligence. He admits that machines can perform almost all of tasks that humans can, and they can possibly do them even better and more efficiently. At this point, I was thinking that if machines can outperform us in everything, then what difference does it make that we are "truly" intelligent? We might as well be the Japanese emperor during the shogunate, a revered figurehead with no real power, like Edward IV, King and nominal ruler of England, but traded between the two sides of the War of the Roses like a pawn.
But it turns out what machines cannot do is originate goals, which have to be supplied by humans.
This is probably the appropriate place to mention the idea circulating that machines need to have an emobodied existence to be intelligent. This conception, not explicitly called out in the series, doesn't withstand Augros's argument because mechanical parts forming a body, even more than electromagnetic parts forming the computer "brains," are machines with parts external to each other and not organics unities.
Nevertheless, it is true that the bodies of organisms give them their drives and teleologies. But with machines, even the purposes of their bodies would have to be given by the humans that construct them. A machine, embodied or not, is not automatically going to, say, stay away from fire in an effort to preserve itself; that drive needs to be part of its programming.
Augros takes a somewhat optimistic tone in his conclusion. But far from a rosy future free of domination by the AI overlords portrayed in science fiction, the real concern is the purposes to which other humans set AI. That concern is not unique to AI, but rather is common to all machines. Purposes no less than the rest of their beings have to be given by their human creators. This peril is common to all technology. In reality our AI overlords will be human.
