I.
If you were going to draw up a list of the smartest people to ever walk the planet Richard Feynman would probably qualify. He is probably best known for winning the Nobel Prize in physics in 1965 as well as an autobiographical series of books detailing his (mis)adventures. It was in the latter endeavor that Feynman’s prodigious talents as a polymath served him well. Throughout his life he indulged his interests in diverse fields ranging from lock picking to bongo playing to computer science.
So when asked about the potential for AI it shouldn’t be surprising that he was able to respond with insight and erudition. Will computers ever be able to think like human beings? No, Feynman hypothesized, that seemed unlikely. Humans attempting to replicate consciousness in silicon are constrained by the underlying materials and technology. In Feynman’s analogy, if the goal is to reproduce a mechanical cheetah1 it turns out that it is far easier to design a machine with wheels instead of legs. If the goal is to construct a mechanical bird it is far easier to use a propeller or a jet engine rather than trying to simulate the flapping of wings. Just as a car doesn’t move like a cheetah and a plane doesn’t fly like a bird a thinking machine is unlikely to reproduce human consciousness.
If Feynman is right that would seem to place a significant limit on the usefulness of artificial intelligence. One could argue that the basic metric by which AI should be judged is how useful it is to human beings. Being able to think like a human would seem to be a critical component.
But if you are breathing a sigh of relief at the idea that your job is secure, maybe think again. As Feynman points out cars may not maneuver like cheetahs but they can transport multiple people at a much faster speed for hours at a time. And just because a car functions differently from a cheetah (or the old standard of transportation—the horse) that doesn’t mean that it won’t successfully move you between points A and B. A calculator doesn’t “do” arithmetic like a person but it still gets the same answer for a math problem like log(7)/log(3).
Are there qualities to human cognition that AI can’t replicate or where it will remain markedly inferior for the foreseeable future? For example, human beings will probably be better at talking to other human beings for quite some time. Putting ourselves in Feynman mode we can ask if there are things that a horse can do better than a car. It could be argued that cars don’t handle rough terrain as well as a horse, or that horses do a better job at traversing narrow trails or fording streams. Plus they can refuel themselves given access to the right type of plant life, which is often widely available. In other words, horses evolved in the wilderness and are therefore better at moving around in nature in its untrammeled state.
But:
Human beings have done a sizable amount of work distorting their cities—and arguably their societies—to utilize the automobile. Consider the millions of miles of paved road laid down in the United States so that cars don’t have to traverse rough and broken ground. Consider also the enormous infrastructure required to drill petroleum out of the ground, refine it and then make it widely available to consumers through a massive network of refueling stations. And consider the number of wars that have been fought and the lives lost to preserve access to that fuel. The fact that the car isn’t a perfect fit for the natural world isn’t a fatal flaw when human beings will go to staggering lengths to adapt both their societies and their physical surroundings to accommodate something that is deemed useful—or desirable.
Feynman was born in Far Rockaway, Queens and kept his New York accent for the entirety of his life. Get over it.