Artificial intelligence learns to play video games ... and cheats

A German university checks that the AI is able to find bugs in video games to complete the game optimally

Video games are the confirmation of how technology applied to leisure can result in hours of fun and amazing works of art. But they are also capable of removing things as embarrassing as the ability to cheat. Well, now a computer with artificial intelligence developed to play Qbert, has been so intelligent that has equaled the human being ... to cheat.

The experiment used a computer programmed with AI to play the classic Qbert de Atari. Games and video games are an area where AI is especially good, as Google's AlphaGo demonstrated by winning the Go world champion.
But sometimes the AI gets out of the script as it has happened to researchers at the University of Freiburg in Germany, where they used Atari games from 30 years ago. In one of the games -Quart- the system found a way to exploit a game bug between levels, which has allowed him to complete the entire game and get unlimited points.

The objective of the game is to go through all the squares of the level and make them change color by jumping on them. This is the trick that AI has found to overcome the game thanks to the bug. The difference compared to AlphaGo is that in video games there is a programming behind the game that maybe the AI is able to understand better than a human and find new ways to win that for a traditional player are invisible.

The fact is that AI researchers also like videogames a lot because they are capable of making the AI 'think' in a different way than any human had ever done before. And that, of course, is an important advance in this area ... even if it is a bit of cheating.