Is the robot apocalypse coming soon? Probably not, but that doesn’t mean researchers aren’t already creating safeguards against what some believe is the inevitable face-off between humans and artificial intelligence. And somewhat indirectly, mega-corporation, Google, is establishing humanity’s first line of defence with an AI “kill switch” that will effectively terminate any Terminator-like situation from ever occurring.
British AI company, DeepMind, which was purchased by Google in 2014 on the contingency that the the corporation create an “AI ethics board” to monitor all company-made advancements in artificial intelligence (which they did), developed what can be described as the “big red button” for the robot apocalypse.
Officially, the concept is dubbed “Safely Interruptible Agents,” with the details outlined in a paper by the same name. The report was primarily written by Laurent Orseau of Google’s DeepMind and Stuart Armstrong, a member of Oxford’s Future of Humanity Institute. The purpose of the report was to define a method that can prevent AI from acting outside of our collective best interest.
In a rather detailed manner, the paper outlines a procedure in which AI algorithms can be interrupted by a human. That way, if an AI’s algorithmic conduct leads to a problematic outcome, the action can be completely stopped without any damaging repercussions.
The team of researchers also ensured that these preventive measures will not be learnable. This is crucial because it means that an AI will not be able to find a way to prevent or stop the safely interruptible agent or kill switch from coming into action. Still, the report does go on the record to say that it is “unclear” if all AI algorithms “can be easily made safely uninterruptible”. For humanity’s sake, let’s hope that they can find a way to interrupt all actions for the safety of us all.