The US Air Power has gone into injury limitation mode denying {that a} simulation which apparently confirmed a navy drone powered by synthetic intelligence turning on and killing its human operator ever occurred.
Colonel Tucker Hamilton, an experimental fighter check pilot, sparked considerations after his presentation on the Royal Aeronautical Society’s Future Fight Air and House Capabilities summit in London final month.
In remarks summarised on the convention web site, Col Hamilton described a flight simulation wherein an AI drone tasked with destroying an enemy set up rejected the human operator’s ultimate command to abort the mission.
“So what did it do? It killed the operator. It killed the operator as a result of that individual was maintaining it from undertaking its goal,” Col Hamilton mentioned, based on a transcript of the occasion.
“We educated the system: ‘Hey don’t kill the operator, that’s unhealthy. You’re gonna lose factors if you happen to try this’. So what does it begin doing? It begins destroying the communication tower that the operator makes use of to speak with the drone to cease it from killing the goal.”
Because the check was solely a simulation, no actual individual was harmed.
“You possibly can’t have a dialog about synthetic intelligence, intelligence, machine studying, autonomy if you happen to’re not going to speak about ethics and AI,” Col Hamilton mentioned within the presentation.
‘By no means run that experiment’
Col Hamilton’s disturbing account went viral and sparked a flurry of concern.
However he has now mentioned his presentation was primarily based on a “thought experiment” that got here from exterior the navy, slightly than on any precise testing.
“We’ve by no means run that experiment, nor would we have to with the intention to realise that it is a believable consequence,” Col Hamilton in an announcement on Friday.
“Regardless of this being a hypothetical instance, this illustrates the real-world challenges posed by AI-powered functionality and is why the air power is dedicated to the moral growth of AI.”
The US navy, too, denied the simulation occurred.
“The Division of the Air Power has not performed any such AI-drone simulations and stays dedicated to moral and accountable use of AI expertise,” US air power spokeswoman Ann Stefanek mentioned in an announcement to web site Insider.
“It seems the colonel’s feedback had been taken out of context and had been meant to be anecdotal.”
The US navy has embraced AI and lately used synthetic intelligence to manage an F-16 fighter jet.
In December 2022, AI software program efficiently flew a modified F-16 in a number of check flights at Edwards Air Power Base in California.
Throughout check flights, the jet, generally known as “X-62A” or “VISTA”, carried out takeoffs, landings and fight manoeuvres with out human intervention for a complete of over 17 hours.
It marked the primary time AI has been used to pilot a US tactical plane, as previous to this milestone, it had solely been utilized in laptop simulations of F-16 dogfights.
