Machines can learn how systems work just by watching them, say Sheffield academics

It's one of the defining characteristics that separates mankind from machines.

Undated Film Still Handout from The Imitation Game, Pictured: Benedict Cumberbatch as Alan Turing, PA Photo/Handout/StudioCanal

But according to University of Sheffield researchers, it is now possible for artificial beings to learn how systems work simply by observing them and without being told what to look for.

Academics say the discovery made in an experiment with swarm robots could mean advances in the world of technology with machines able to predict, among other things, human behaviour.

Sign up to our daily newsletter

The i newsletter cut through the noise

It takes inspiration from the work of pioneering computer scientist Alan Turing, who proposed a test, which a machine could pass if it behaved indistinguishably from a human.

The Sheffield study uses a similar test to reveal how any given system works. In it, two swarms of robots, one of which has learning ability, have their movements recorded.

The motion data is shown to ‘interrogators’, not humans as in the Turing test but computer programs that learn by themselves.

The learning robots that succeed in fooling an interrogator, by making it believe their motion data were genuine, receive a reward.

According to Dr Roderich Gross from the Department of Automatic Control and Systems Engineering at the University of Sheffield, the new approach mean humans no longer need to tell machines what to look for.

The discovery could be used to create algorithms that detect abnormalities in behaviour. This could prove useful for the health monitoring of livestock and maintaining machines, cars and airplanes.

Dr Gross said the advantage of the approach, called ‘Turing Learning’, is that humans no longer need to tell machines what to look for.

“Imagine you want a robot to paint like Picasso. 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 does not require such prior knowledge. It would simply reward the robot if it painted something that was considered genuine by the interrogators. Turing Learning would simultaneously learn how to interrogate and how to paint.”