Defining an objective goal and defining a small set of actions to try at random, and then iterating with small changes to the most successful series, will obviously produce a series of instructions capable of achieving that goal, but to say it resembles intelligence is silly. It's not intelligence, it's brute force trial-and-error. It's like they are comp sci freshmen who wrote MyFirstGeneticAlgo.java, read a bit of Heidegger, and now they think they're geniuses.

This NES bot is a *way* cooler example of the same basic idea. It even finds its own goals: Instead of a defined goal of "entropy staying high", it searches for sets of lexicographically ordered values in RAM that get progressively higher based upon some example input sequences, and it "learns" to play the game based upon its self-defined goal and even invents new techniques to "win".