A Unified Theory of Machine Life and Artificial Intelligence

It seems to me that artificial intelligence is a problem that ought to be set aside for a while. First we have to comprehend natural stupidity. If we are to be authentic (in Sartre’s sense), no matter how we characterize it, we all know that we have met people who are much smarter and much more stupid than we deem ourselves.

So we have to explain natural stupidity as part of the problem of explaining natural intelligence. Then we can begin to characterize AI software as AS (Artificial Stupidity) or AI (Artificial Intelligence) – and further, it’s not black and white, it’s a gradient from carrot to Einstein.

This is one thing that bothers me about the SETI project. Why not be satisfied with searching for stupid life? Wouldn’t that be enough to rock our religious and philosophical foundations? Admittedly, SETS would be more difficult to locate, but why do we raise the bar so high?

Back to AI software. We have to recognize a semantic issue here. Before the Wright brothers, there were numerous attempts to make men fly – modelled on bird’s wings, tree-seed-heliocopters, and so on. None worked. Then the Wright brothers made something work, and the definition of the verb “to fly” changed. Planes fly. I fly to Los Angeles.

The same redefinition is going to occur with the verb “to think” – which you alluded to. If it walks like a duck and quacks like a duck, then it’s a duck. But I think that to some extent, this misses the point. It doesn’t matter that Big Blue can beat Kasparov; it matters how quickly Big Blue can learn backgammon or bridge. We don’t measure intelligence on what we already know; that is a measure of knowledge. The measure of intelligence is, How quickly can we learn something new and unforeseen? That’s how we judge people around us: not by what they’ve read or what they’ve already learned, but by how quickly they adapt to new circumstances, see things we didn’t see before them, advance conjectures that never occurred to us, and have insights we didn’t even fathom. That’s intelligence.

I can teach you the rules of backgammon in 10 minutes. The statistics underlying the correct plays will take longer. And as my friend Vlad Dobrich phrased it so well, “The first year, you learn to move the pieces. The second year, you learn to move the doubling cube. The third year, you learn to move your opponent.”

The basic difference between AI and human intelligence is that humans don’t have to be pre-programmed. This is a huge chasm. I have spent my past 30 years writing software, some of it attempts at AI, some at Genetic Algorithms, etc. I am all too aware that the smartest AI program is dumber than a goldfish.