Gigaom AI Minute – February 10

Transcript

It's no surprise that sports teams often take on a whole lot of superstitions. In the grand scheme of things, they just don't play that many games, so if player X stubs his toe and goes on to win, and then the next day stubs his toe again and wins, it is not unreasonable to assume that he should go around stubbing his toe before every game, and perhaps even encouraging his teammates to do the same.

The interesting things is that that humans are not the only creatures that behave superstitiously. Pigeons, for instance, do it as well. If a pigeon does some behavior and is rewarded with food, such as maybe some elaborate dance or something like that, it would become a superstition, and it would do that dance and expect a reward. And even when the superstition is proven to be unreliable, as maybe our stubbed toe example, the pigeon continues to do it to grasp onto this view of reality. This only happens, because, as I said, we experience life at such a slow pace.

Computers would also be, in theory, prone to superstitions, because their training algorithms are trained on limited sets of data in the real world. They can, in fact, have erroneous data sets. But if the data set is large enough and appropriately representative, all of the superstitions should be wiped out of the system.

I mean, think about it. If a baseball player stubbed his toe a hundred thousand times and won a hundred thousand games, and then didn't stub his toe a hundred thousand times and lost, you could reasonably conclude that perhaps stubbing his toe was, really in fact, a key constituent of winning the game.

I suspect that much of what we think we know about the world is in one form or another, superstition. We just deal in our minds, with such limited data sets, very few things about our world have really been subjected to the enormous data sets required to get rid of superstition entirely, thanks to the law of large numbers. In addition, we're right now probably training our machines to be equally superstitious, but as our datasets grow and as our computers get faster, that will no longer be the case in the future.