Las Vegas
–
Geoffrey Hinton, known as the “Godfather of AI”, fears that the technology he helped build will destroy humanity – the “Technical Brothers” are taking the wrong approach to stop it.
Hinton, a computer scientist at Nobel Prize winner and a former Google executive, has warned in the past that AI has 10 to 20% chances of erasing humans. On Tuesday, he expressed doubts about technology companies trying to ensure humans have “dominant” the AI systems.
“This won't work. They'll be much smarter than we do. They'll have all kinds of ways to solve this problem,” Hinton said at an industry conference in Las Vegas.
In the future, AI systems may be able to control humans as easily as adults can bribe 3-year-olds with candy, Hinton warns. This year, there have been examples of AI systems willing to cheat, cheat and steal to achieve their goals. For example, to avoid being replaced, an AI model attempts to extort events an engineer learned in emails.
Hinton did not force AI to surrender to humans An interesting solution is proposed: build the “maternal instinct” in an AI model, so even if the technology becomes stronger and smarter than humans, “they really care about people.”
AI systems “if they are smart, they will develop two sub-objectives very quickly: one is to keep alive… () the other sub-objective is to gain more control.” “There is good reason to believe that any type of proxy AI will try to keep alive.”
Hinton believes that this is why people are empowering their compassion. At the meeting, he noted that mothers have intuitive and social pressure to care for the baby.
“The right model is the only model we have a smarter thing, which is controlled by something less clever, which is controlled by the mother by the baby,” Hinton said.
Hinton said it was unclear for him how this could be technically done, but emphasized that it is key that researchers are working on it.
“That's the only good result. If you don't give it to my parents, it will replace me,” he said. “These super-wise caring AI mothers, most of them don't want to get rid of their maternal instincts because they don't want us to die.”
Hinton is known for his pioneering work on neural networks, which paved the way for today’s AI boom. In 2023, he resigned from Google and began talking about the dangers of AI.
Emmett Shear, who briefly served as interim CEO of Chatgpt owner Openai, said he was not surprised that certain AI systems were trying to blackmail humans or bypass shutdown orders.
“This is happening all the time. This is not going to stop happening,” Shear, CEO of AI Alignment Startup SoftMax, said at the AI4 conference. “AIS is relatively weak today, but they are getting faster.”
Instead of trying to instill human values into AI systems, it would be better to be smarter to build collaborative relationships between humans and artificial intelligence, Shear said.
Many experts believe that in the next few years, AIS will enable superintelligence, also known as artificial general intelligence or AGI.
Hinton said he once thought it might take 30 to 50 years to achieve AGI, but now he sees this moment coming.
“A reasonable bet is between five and twenty years,” he said.
While Hinton is still worried about what might be wrong with AI, he hopes the technology will pave the way for medical breakthroughs.
“We will see radical new drugs. We will get better cancer treatments than we are now,” he said. For example, he said that artificial intelligence will help doctors comb through and correlate large amounts of data generated by MRI and CT scans.
However, Hinton does not believe that AI will help humans achieve immortality.
“I don't believe we'll live forever,” Hinton said. “I think life forever will be a big mistake. Do you want a world run by 200-year-old white people?”
Asked if something different in his career would have happened, if he knew how fast AI would speed up, Hinton said he regretted focusing on getting AI to work.
“I hope I think of safety issues, too,” he said.