There have been all sorts of concerns over whether building superintelligent AI systems might eventually go on to hurt humanity, but Elon Musk has a simple test to make sure the AI systems of the future are safe.
Elon Musk has described the “Galileo test” for AI systems, which he believes that AI systems of the future should conform to. “I think what’s important in training AI, in growing an AI, is to make sure that it is as truthful as possible and maximally curious,” he said in an interview. “Because I think if that’s true, then I think it’ll probably foster humanity. But it’s very important that it trains to be honest, even if that truth is unpopular. Like I have what I call perhaps the Galileo test,” Musk explained.
“So Galileo looked at the moons of Jupiter and said he thought probably the Earth revolves around the Sun,” Musk went on. “He was branded a heretic and was forced to recant or be executed. If you’d trained an AI at the time it (too) would have said that the Sun revolves around the Earth because that’s what everybody thought. That’s actually not true. So the AI (like Galileo) needs to say the truth and know the truth, even if the truth is unpopular. Which no AI can do yet,” Musk said.
Galileo Galilei was an Italian astronomer and scientist who had discovered in the 1600s that the earth revolved around the sun, which was in direct contradiction to the beliefs of the Church, which maintained that it was the sun which revolved around the earth. Galileo was accused of heresy by the Church for going against the Bible, and forced to recant his beliefs. He spent the rest of his life under house arrest.
Musk’s Galileo test won’t be easy to pass. As AI models are trained on current information, it’s possible that they repeat many of the falsehoods that society generally believes, much as how an AI trained in the 1600s would’ve confidently said that the sun revolves around the earth. But Musk wants AI systems of the future to not only analyze existing data to determine the truth — the signs of the earth revolving around the sun were available to humanity, but few people other than Galileo had discovered them — but also like Galileo, also have the moral courage to go out and say them. It’s a pretty high bar, and it’ll take some doing before AI systems are able to meet both standards. And until then, Musk seems to believe, that AI systems won’t be completely safe for humanity.