Reports this week suggest that Google’s DeepMind, the AI programme which learns independently has developed some interesting strategies under stress.
The team responsible for DeepMind have published a blog outlining how it responded to stressful circumstances during an apple gathering computer game with another AI agent. It appears that the agents didn’t want to share their spoils, employing aggressive tactics to ensure they won. So what’s going on with DeepMind?
The AI agents went head to head to see who could gather the most apples. When the apples in the game were abundant, the agents “operated smoothly” leading to a situation where “the agents learn to peacefully coexist and collect the apples as they can.” according to the team.
Scarcity, stress and aggression
It was a different story when the apples were in short supply. The team went on to describe how “as the number of apples is reduced, the agents learn that it may be better for them to tag the other agent to give themselves time on their own to collect the scarce apples.” The team reported that DeepMind resorts to “highly aggressive strategies” to make sure it wins. Rather than playing nicely, both agents resorted to using laser beams to knock each other out of the game.
So was Stephen Hawking right when he warned that AI might be the ‘worst thing’ to happen to humanity? Interestingly enough, the less intelligent, smaller versions of DeepMind opted to share the apples. In a subsequent game where it was in the DeepMind agents interests to work together, they learned that cooperation was more likely to lead to success than aggression and selfishness.
The jury’s still out on the potential for AI and DeepMind but in the meantime we’ll pass on playing scrabble with either of these two laser touting tech brats.
