Cool. I’ve seen a few of these, including explanations of the literal pitfalls, where the computer deliberately jumps into pits because that allows it to continue moving to the right.
It had me wondering how different this was from just having a computer randomly press buttons until it’s mapped a path of least resistance through the level by pure trial-and-error, and I guess the difference is that instead of mapping buttonpresses in relation to the entire level, it is mapping buttonpresses in relation to the platforms and enemy blots, thus generating a short-range gameplay procedure instead of a long-range path. Fascinating stuff, and cool to watch. This is the first video I’ve seen that shows what the AI is actually seeing.