Building a slide deck, pitch, or presentation? Here are the big takeaways:
- Researchers used evolutionary strategy algorithms to create an AI system that could gain a high score on the game Q*bert by exploiting an old bug.
- AI systems have beat human players of games including chess, Go, and Pong.
An artificial intelligence (AI)-powered bot exploited a bug in popular 1980s arcade game Q*bert to gain an all-time high score, according to a paper from researchers at Germany's University of Freiburg.
The AI bot was programmed using evolutionary strategy (ES) algorithms, which include machine learning and allow the AI to learn, adapt, and change tactics depending on the situation and other players, as noted by our sister site ZDNet. ES algorithms also serve as an alternative to the more common reinforcement learning (RL) methods that have been used to train the systems that beat humans at other games.
In Q*bert, players must jump from cube to cube to change colors, while avoiding obstacles and enemies to make it to the next round. However, the AI system was able to jump quickly from cube to cube in no particular order, causing the colors to change rapidly. A bug caused the platforms to keep blinking and the AI to continue to bounce around and gain a score of nearly one million points, as reported by The Register.
SEE: IT leader's guide to the future of artificial intelligence (Tech Pro Research)
The AI bot was trained on some 1.5 million parameters to create the ES system, The Register reported. Exploiting the bug in the game was the easiest way to gain the goal of a high score.
This is far from the first time that an AI system beat a human player at a game—though it may be one of the first times that cheating was involved. In March 2016, Google DeepMind's AlphaGo machine learning platform defeated world champion Lee Sedol in Go, a game more complex than chess. In October 2017, Google revealed AlphaGo Zero, which was quickly able to defeat the previous version and become a Go world master. Google's DeepMind system has also beat human scores at a number of other games, including Pong.
In their paper, the researchers found that ES can beat RL in a number of cases, which could have implications for the future of AI training.
"This success of a basic ES algorithm suggests that the state-of-the-art can be advanced further by integrating the many advances made in the field of ES in the last decades," according to the paper.
- Special report: How to implement AI and machine learning (free PDF) (TechRepublic)
- Researchers reveal how poker playing AI beat the best human players (ZDNet)
- Machine learning: The smart person's guide (TechRepublic)
- AlphaGo returns upgraded, wins 60 straight games online (ZDNet)
- AI: How big a deal is Google's latest AlphaGo breakthrough? (TechRepublic)
Alison DeNisco Rayome is a Staff Writer for TechRepublic. She covers CXO, cybersecurity, and the convergence of tech and the workplace.