acm-header
Sign In

Communications of the ACM

ACM TechNews

AI Smashes Videogame High Scores by Remembering Its Past Success


View as: Print Mobile App Share:
Montezumas Revenge is one of the most challenging Atari games.

An artificial intelligence that can remember its previous successes and use them to create new strategies has achieved record high scores on some of the hardest video games on classic Atari consoles.

Credit: ArcadeImages/Alamy

Artificial intelligence (AI) developed by researchers at AI research centers Uber AI Labs and OpenAI achieved record high scores on videogames played on classic Atari consoles, by recalling previous successes.

The algorithm remembers all the various strategies it has attempted, and continually returns to moments in which it had a high score as a jumping-off point for new tactics.

The software retains screen grabs from a game as it plays to remember past strategies, clustering similar images to identify points it should revisit as a starting point.

The AI aims to maximize its score, and updates its record of a starting point when used to reach a new high score with a new screen grab from that part of the game.

In playing 55 Atari games that have become a standard benchmark for reinforcement learning algorithms, the AI bested state-of-the-art algorithms 85.5% of the time.

From New Scientist (02/24/21) Matthew Sparkes
View Full Article


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account