AI teaches itself video games, takes down human opponents
Artificial Intelligence continues to outperform all expectations
The emphasis on artificial intelligence (AI) has always been the intelligence part of the equation and the efforts to build machines that are capable of overpowering humans continue to advance. A new AI that was developed by researchers from DeepMind has been able to teach itself how to play games and how to beat human opponents in competition. This time, the game isn’t checkers or backgammon or chess – it’s Quake III Arena, a first-person shooter game.
DeepMind is probably the most famous AI currently being developed, with Deep Blue holding the title of the father of AI. Deep Blue, which was designed by IBM, beat Gary Kasparov at chess in 1997 and launched an entirely new era of AI construction.
More recently, AI has been shown to be more than successful playing games like poker and Go. Advanced AI players took on human counterparts in both games over the past couple of years – including the number one Go player in the world – and easily defeated them all each time.
Moving into first-person shooter games is completely different, though, and an AI that can teach itself the game is even more impressive. Quake III incorporates teams of players working toward the completion of a single mission and involves a great deal of interaction and complex interactions.
The AI was used to create “agents” in the game that were set free to learn the finer nuances of the action. Once they had a handle on the game, they took on professional gaming testers and lost. According to the researchers, “Even after 12 hours of practice, the human game testers were only able to win 25% of games against the agent team.”
It should be interesting to see what happens next.