Gigaom brings you our unique analysis and commentary on the present and future of AI.
The kind of AI that people are excited about right now is machine learning, but there are other techniques. One is evolutionary AI. Now, machine learning requires a lot of data about the past on which to train the AI about the future. But some questions you want to pose to AI, don't have historic data to go with them. For instance, if the question put before the AI was, "What's an aerodynamic design for a car?" you can't just take every car that's ever been made and study them and figure out which one was the best. In fact, probably the best way to do it is to do a simple model of a wind tunnel and take a cube and then randomly change something about that cube and put it in the wind tunnel and see if it's any better than the cube. If it is better than you randomly alter it again and run it through the wind tunnel and keep doing that and you evolve a kind of better and more aerodynamic car.
This is a technique that is very much with us today and it's used in a number of different areas from figuring out routing and other areas where there are too many permutations and not enough historic data in order to train things. Evolutionary AI is probably going to be used more and more in the future, as a matter of fact, but it isn't the kind of thing that a normal enterprise is going to have a lot of need for now. Enterprises typically adopt questions that have historic data to inform, and evolutionary AI is for things that are more abstract in nature.