At an international defense conference in London this week, Col. Tucker Hamilton, the chief of AI test and operations for the US Air Force, told a funny — and terrifying — story about military AI development.
“We were training [an AI-enabled drone] in simulation to identify and target a SAM [surface-to-air missile] threat. And then the operator would say yes, kill that threat. The system started realizing that while it did identify the threat at times, the human operator would tell it not to kill that threat, but it got its points by killing that threat. So what did it do? It killed the operator. It killed the operator because that person was keeping it from accomplishing its objective.”
“We trained the system — ‘Hey don’t kill the operator — that’s bad. You’re gonna lose points if you do that.’ So what does it start doing? It starts destroying the communication tower that the operator uses to communicate with the drone to stop it from killing the target.”
In other words, the AI was trained to destroy targets unless its operator told it not to. It quickly figured out that the best way to get as many points as possible was to ensure its human operator couldn’t tell it not to. And so it took the operator off the board. (To be clear, the test was a virtual simulation, and no human drone operators were harmed.)
When ridicule turns to fear
As AI systems get more powerful, the fact it’s often hard to get them to do precisely what we want them to do risks going from a fun eccentricity to a very scary problem. That’s one reason there were so many signatories this week to yet another open letter on AI risk, this one from the Center for AI Safety. The open letter is, in its entirety, a single sentence: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
Signatories included 2018 Turing Award winners Geoffrey Hinton and Yoshua Bengio, both…
Read the full article here