It’s official. Humans are now terrible at playing Ms Pac-Man, a 1980s game that involves a monster eating things being chased by ghosts and rated to be one of the hardest games for an AI to beat.
An AI from Microsoft‘s Maluuba team — a Canadian deep learning startup acquired earlier this year — has now scored the maximum possible number of points of 999,990 in the Atari game, beating the human record by four times.
This was done using a method of reinforcement learning called Hybrid Reward Architecture, which involves teaching 150 AI agents to work together in parallel to master the game. Agents were rewarded for eating pellets while others avoided ghosts. A top agent then got feedback from the agents and used a weighted average to make decisions.
The success of the Ai will help progress towards teaching AIs agents to perform better, and Hybrid Reward Architecture could help sales organisations predict when to target customers, or be used in understanding natural languages, said Microsoft researchers.
Special Reports: CNET’s in-depth features in one place.
Technically Literate: Original works of short fiction with unique perspectives on tech, exclusively on CNET.