Google’s artificial intelligence company DeepMind has developed a breakthrough AI system that learns in a similar way to a human in order to prevent “catastrophic forgetting.”
The technique allows the system to learn based on past experiences, meaning it can learn how to play any computer game without any prior knowledge of it.
DeepMind has previously developed AI capable of mastering Atari games like Breakout. However, each system was only capable of playing one game. The new research is an important step toward developing artificial general intelligence—or human-level intelligence.
“We took inspiration from neuroscience-based theories about the consolidation of previously acquired skills and memories in mammalian and human brains,” DeepMind researchers wrote in a blogpost describing the research.
“Our algorithm specifically takes inspiration from [mammalian mechanisms] to address the problem of catastrophic forgetting.”
Google has invested heavily in assembling some of the brightest minds in AI research, employing them to develop everything from self-driving cars to improved search algorithms. In 2014, Google acquired the London-based startup DeepMind for $500 million, which has since gone on to be Google’s AI flag bearer, making headlines for its creation of the first computer capable of beating a human champion at the boardgame Go.
The broader challenge has been to create AI that can compete with humans at more than just one task. Leading AI academic Nick Bostrom said in 2016 that he believed Google is leading the global race to develop human-level artificial intelligence.
“There are different bets on what approach [to developing human-level AI] is most promising, and since we don’t know what approach will ultimately work, there is some uncertainty there,” Bostrom said last October.
“At this point in time I think that DeepMind is very strong… It is probably the largest group specifically trying to solve general intelligence. But if this happens three decades from now, there might be some entirely new thing that doesn’t exist yet, just as three decades ago a lot of the current players wouldn’t be on the table. A lot could change many times over in the remaining time.”