DeepMind Technologies Limited is a British artificial intelligence company founded in September 2010. Acquired by Google in 2014, the company has created a neural network that learns how to play video games in a way similar to humans. On July 3rd, the experiment was concluded wherein multiple AI systems were trained to play a multiplayer first-person shooter game and capture the Flag on Quake III Arena.
The AI named ‘For the Win’ (FTW) trained here is a better player than any human player in the game regardless of the opponent. FTW had to play 450,000 games of Quake to reach up to human gamer level and to understand how to effectively work with other machines and humans.
“We train agents that learn and act as individuals, but which must be able to play on teams with and against any other agents, artificial or human,” posted a company member. As we know, from a multiplayer perspective, any game requires individual and teamwork to oppose the enemies without being affected by the players. According to a survey, humans have said that FTW is a better teammate than humans.
AlphaGo, created by DeepMind, is an AI system that won over the top GO platers in 2017. AlphaGo Zero is improved by making it play games with itself. DeepMind involved 30 agents totally, but 4 players play simultaneously against human or machines. In a tournament between machine and human teams, the machine team had 95% chances of winning only if it played against humans. Agents were more efficient than humans in tagging and achieving statistical results.
To train the AI agents better, the terrains layout was changed to a more flat terrain for the indoor environment and varying elevations for outdoor to know their basic general understanding for better outcomes. The agents developed their own rewards system depending on the speed and time to capture the max flag. They even learned the rules as the game proceeded. These games help the machine to learn machine strategy, memory, patterns interpretation, and many more things.