8th June 2016

In order to prevent the human race being overrun by robots in the future, scientists are working on a ‘kill switch’ which Artificial Intelligence (AI) will be unable to overrun.

The scientists, from Oxford University and DeepMind, the AI division at Google, outlined in an academic paper how they could code AI to stop them from being able to learn how to override the input of humans.

Google DeepMind’s Laurent Orseau, and Oxford’s Stuart Armstrong, have been developing a framework which will look to ensure the AI will always remain under the control of humans.

They acknowledged that it is unlikely AIs in the future will “behave optimally all the time,” hence the need for research to be done to ensure that human overseers will remain in control.

“Now and then it may be necessary for a human operator to press the big red button to prevent the agent from continuing a harmful sequence of actions,” they wrote.

Their research focuses on using a method that prevents AIs from learning how they can manipulate, or even avoid, human interventions as they perform their tasks. AIs learn through reinforcement, and humans are able to interrupt them safely, and repeatedly, but if the AIs found a way to avoid this, then it could cause significant problems.

AI is increasingly being introduced into our lives, but there are very real concerns – even among technology experts – over whether we can remain in control and stop the AIs from learning how to override the ‘kill switch’.

“It is sane to be concerned,” Dr Orseau said. “But, currently, the state of our knowledge doesn’t require us to be worried. It is important to start working on AI safety before any problem arises. AI safety is about making sure learning algorithms work the way we want them to work.”

However, he did add that “no system is ever going to be foolproof. It is a matter of making it as good as possible, and this is one of the first steps.”