AI Learning To Fight And Trounce Professional Players On Super Smash Bros.

It would seem that scientists are now developing a fascination for using video games to teach artificial intelligence. Some of the most alarming titles used include the crime simulator Grand Theft Auto V and the real-time strategy game, Starcraft II. Now, an AI is given the chance to learn fast combat maneuvers via a popular video game called Super Smash Bros.

For those who have never heard of it, Super Smash Bros. is Nintendo’s experiment that became a star. The idea is simple; throw various Nintendo characters together in one game and make them fight each other. It became such a huge hit that the Japanese gaming company has been remaking it regularly, with the fights and effects becoming more and more ludicrous. This makes the title an odd choice for teaching an AI, Futurism notes.

With games like Starcraft II, the algorithm is provided an environment that is a more advanced version of Chess or Go, which AIs have been excelling at for years. With Super Smash Bros., however, aside from using the right combinations of attacks, players also need to take their environment into account. By making use of ledges and tools, pros are able to turn the tide of battle, especially during a full rumble.

The AI is called Phillip and it was created by an MIT Ph.D. student, Quartz reports. Recently, it was pitted against a professional player named Dustin White. During their fight, it was clear that Phillip was the better player.

One of the biggest reasons why the AI is so much better at the game is sheer reaction time. Whereas humans typically react at about 200 milliseconds, Phillip has a reaction time of 33 seconds. This literally made the AI a player on a whole other level, with a fighting style akin to having its own plane of existence.