The study, published in the journal Swarm Intelligence, could enable machines to analyse and predict both artificial and natural systems, including human behaviour. Known as Turing Learning, the approach requires no prior machine knowledge, and simply involves rewarding behaviour that is analogous to that desired, and which can fool the ‘interrogator’ in a Turing Test.
“Our study uses the Turing test to reveal how a given system – not necessarily a human – works,” said Sheffield University’s Dr Roderich Gross. “In our case, we put a swarm of robots under surveillance and wanted to find out which rules caused their movements. To do so, we put a second swarm – made of learning robots – under surveillance too. The movements of all the robots were recorded, and the motion data shown to interrogators.”
The interrogators were in fact self-learning computer programs whose task was to distinguish between the swarms. The programs were rewarded when they correctly identified the counterfeit data, and the learning swarm was rewarded when it fooled the interrogating program. According to Gross, the advantage is that humans no longer need to show machines what to look for.
“Imagine you want a robot to paint like Picasso,” he said. “Conventional machine learning algorithms would rate the robot’s paintings for how closely they resembled a Picasso. But someone would have to tell the algorithms what is considered similar to a Picasso to begin with.”
Turing Learning, however, would simply reward painting that fooled the interrogators, simultaneously learning how to interrogate and how to paint. According to the researchers, the approach could be used to detect abnormalities in behaviour, and has applications ranging from livestock monitoring to preventative maintenance.