Can robots live and learn by the laws of evolution?
That's what one research team in Switzerland wanted to answer. So they created a scenario in which robots with simple "neurons" were overseen by a computer playing the role of nature by randomly picking traits or behaviors that would survive or die.
What happened next wasn't surprising, said biologist and researcher Laurent Keller, because it's exactly the way it should work in theory.
But to see it was nonetheless stunning.
In one of the most interesting scenarios, the scientists put a colony of robots in a kind of hockey rink. Some of the robots were programmed with similar genes or traits, while others were given completely different sets of genes. The scientists dropped tokens throughout the rink and decided that if the robots moved them to a certain part of the rink, they'd receive a positive point.
This is what happened:
Media no longer available
I was floored by the fact that the robots went from helpless to being able to coherently move the tokens without human interference. That and the fact that this "evolution" took, as Keller told me, only a few days thanks to the amazing processing speed of their computers.
"What was important was kin selection," said Keller. What you see in the video at generation 149 is robots with the same — or very similar — genetic make up worked together to move tokens across the rink. The other robots — the ones with different genes — worked alone.
The kin robots, said Keller, developed altruistic behaviors. In other words, they decided to work for the good of the group instead of the good of the individual.
And none of this behavior was programmed into the robots, said Keller. The natural selector computer was responsible for randomly selecting the fittest traits.
(If you're immensely curious, you can read the nitty gritty details of the scientific setup by reading Keller and Dario Floreano's paper at the biology journal PLoS Biology.)
Keller is a biologist, so he's focused on what this experiment says about evolution, about how they've been able to reproduce what many intelligent design proponents say is too complex to happen naturally.
But the whole time, I was thinking: Whoa! Robots that learn from a computer! Robots that work together toward a greater goal! This has got to be a harbinger of global robot domination!
What do you say to that? I asked Keller.
"Well," he said. "It's not like we had robots building each other."
"Is that even possible?" I asked, concerned.
Without hesitation, he shot back, "Not yet."