The man who is known as the “godfather of AI” still talks about the dangers of AI weeks after he quit his job at Google. In a recent interview, Geoffrey Hinton said that there was a “serious danger” that “things will soon be smarter than us, and these things may have bad intentions and take over.” He said that politicians and business leaders need to start thinking about what to do right away.
Hinton warned that technological progress is no longer a science fiction story, but a serious problem that is probably going to happen very soon. For example, he told the outlet that artificial general intelligence, which can understand or learn any intellectual task that a human can, might not be too far away. “And I thought for a long time that wasn’t going to happen for another 30 to 50 years,” he said. “I think we’re much closer now. Maybe only five years until that happens.”
Some people have compared chatbots like OpenAI’s ChatGPT to autocomplete, but Hinton said the AI was taught to understand, and it does. “Well, I’m not saying that it has consciousness. “But I’m also not saying it’s not conscious,” he told.
“They can certainly think and understand things,” he said. “Some people use the word ‘sentient’ to mean ‘does it have its own experiences?’ I think that bringing up the issue of subjective experience just makes the whole thing harder to understand, and you end up talking about a lot of things that are kind of religious about how people are. Let’s not do that.”
He said that he was “unnerved” by how smart Google’s PaLM model had become, pointing out that it could understand jokes and why they were funny. Since then, Google has released PaLM 2, the next-generation large language model with “improved multilingual, reasoning, and coding capabilities.” When this kind of AI comes out, people worry that it will take over their jobs, cause political fights, and spread false information.
Some leaders, like Elon Musk, who has his own stake in AI, had signed an open letter to “immediately pause for at least six months the training of AI systems more powerful than GPT-4,” but Hinton doesn’t think it’s possible to stop the research.
“The research will be done in China if it doesn’t happen here,” he said. He pointed out that AI would have many benefits and said that leaders need to put a lot of time and money into figuring out how to “keep control even when they’re smarter than us.” “All I want to do is sound the alarm about the existential threat,” he said, adding that others had been written off as “a little bit crazy.”
