Video games are great teaching tools. People have been learning and developing skills through the medium ever since they were invented. That being the case, there’s no reason artificial intelligence can’t learn from video games as well. In fact, it can be argued that video games are the ideal testing ground for artificial intelligence. The artificial intelligence program AlphaStar is one such AI using a game for training, and that game is the popular StarCraft; a strategy game in which a player must make a base, create an army, and destroy the base of their opponent. It’s an excellent platform to teach an AI how to innovate, because it requires a player to plan for numerous outcomes and attempt to predict enemy movements without solid information.
AlphaStar was pitted for the very first time against world class player Dario Wünsch, a 28 year old from Leipzig that had been playing StarCraft II for about a decade. He would be the first professional player to go against AlphaStar. Even the creators of the AI, research company DeepMind, didn’t think it would be able to defeat Wünsch. After all, many AI had been developed with the goal of besting human players in StarCraft before, and they had all failed.
At first, it looked like AlphaStar would fail as well, leaving its base totally undefended at the start of the first match. Wünsch swooped in and started picking off workers in the AI base, only to find, a minute later, that the tenacious machine had assembled a strike team and had intentionally left its defenses down. Wünsch, thinking he could take advantage of the opening, had accidentally left an opening of his own, resulting in a swift defeat. Of the five following rounds, AlphaStar employed various tactics and strategies, and ultimately defeated Wünsch all five times.
To some, this might not seem to be all that important. What’s the point of teaching artificial intelligence how to beat people at video games? It’s actually far more valid than you’d think; in the past, another AI program was created to play the game DOTA 2, in which five players need to work in conjunction to defeat the enemy. Like StarCraft, it is a very complex game, and the data the AI gathered in that virtual world was later used to help it control the five fingers of a robotic hand with dexterity that no human could match.
Point being, there is a great deal that artificial intelligence can learn from gaming, not for the sole purpose of gaming, but to help them better deal with complex real life situations and obstacles. There’s no denying the value of this type of learning, especially considering that there are no real repercussions if the AI fails in these virtual learning environments. Of course, on another note, gamers might have to be worried about cheaters in the future using AI to help them win, but that’s a small problem compared to the real things that AI will be able to achieve as it grows.
