Actions Per Minute

3 minutes read —

I remember listening to some documentary about an AI developed for some strategy game called Starcraft. The developers had to put an arbitrary constraint on the AI ; they limited its APMs. APM is an acronym that stands for Actions Per Minute. They figured, even though a program could perform clicks at a very high speed, it wouldn’t make for a fair game against human competitors. You would think a strategy game is all about the quality of the decisions, not just the sheer quantity of them. And you would be right, probably. The AI, which was called AlphaStar, ended up beating the best starcraft competitors while playing with a handicap limiting its speed to well below that of the aforementioned human players. Therefore, you could say, good judgment makes for effortless wins, and thus, this parameter should be optimized for. It is all about finding the greatest moves, it would seem, rather than hastily putting things together in a half-assed manner. In this paradigm, it would make sense to slow down and evaluate and reevaluate every decision until one makes the perfect one ; minimizing the risks and maximizing the potential for rewards. This all makes perfect sense, except that when we confront this theory to real-life, it starts to crumble and fall apart. The best starcraft players also happen to have the highest APM scores by far. This is a strong predictor of their performance. They execute a lot of actions, really quickly, and the fastest players at this level usually end up getting the win. Although better and more efficient decisions do exist, as proven by machine learning, they are not available to our limited human brain powers, or at least not without a trade-off in speed so great that we end up losing by inaction. Our best bet for winning is limited to making the best actions we can think off as fast as we can, which is very different from making the best actions period. Doing so would require way more CPU than is humanly possible. JK Rolling said it another way : “it is impossible to live without failing at something, unless you live so cautiously that you might as well not have lived at all — in which case, you fail by default”. I guess what I’m writing about right now is the dangers of analysis paralysis and how that can prevent us from doing much at all. There is a limit to how careful our planning as to be before we can and should act on it. Decisions that carry no danger of being fatal should be made quickly. We can’t anticipate all possible effects, therefore sometimes we have to go with our guts and let the feedback come from real-life. It is about adaptability and knowing that whatever comes, we will be able to react with the same speed we used to make the decision in the first place. You know all about it, right ? Consistent and decisive action, even if imperfect, leads to better outcomes than waiting for the perfect solutions to emerge. So let’s keep moving forward !