Superhuman artificial intelligence that is smarter than anyone on Earth could exist next year, Elon Musk said, unless the industry’s demand for power and computing becomes unsustainable before then.
The prediction is a sharp extension of an earlier claim by the multi-billionaire that super-intelligent AI would exist by 2029. While “superhuman” is generally defined as smarter than any individual human at a specific task, superintelligent is instead often defined as smarter than the combined ability of each human at any task.
“I suspect that by the end of next year we will probably have AI that is smarter than any human,” Musk said in a livestream interview on his social network to the most powerful AI training chips could limit their capabilities in the short term.
“Last year there was chip restriction,” he said. “People couldn’t get enough Nvidia chips. This year we will switch to a power supply with voltage transformers. In a year or two it will only be electricity supply.”
In 2023, when he predicted a five- to six-year runway for superintelligence, Musk was vocally concerned about the consequences. Speaking at the launch of his AI startup, xAI, that year, he said: “If I could press pause on AI or really advanced AI digital superintelligence, I would. It doesn’t seem realistic, so xAI is essentially going to build an AI. In a good way, hopefully.
“It’s actually important that we worry about a Terminator future to avoid a Terminator future,” Musk added, referring to the film in which a sentient computer system wages war on humanity.
A year later, xAI is firmly trying to lead the development of super intelligence. In a recent interview, Musk said the latest version of his chatbot Grok AI is similar to GPT-4, OpenAI’s leading model. GPT-4 is over a year old and competitors have already reached or exceeded its capabilities, with Anthropic’s Claude 2 Sonnet widely seen as the new market leader.
The entrepreneur’s predictions are notoriously made with complete freedom. In 2016, he incorrectly predicted that within two years it would be possible for a Tesla to drive autonomously from New York to Los Angeles. That same year, he said his SpaceX rocket company would fly to Mars in 2018 — that still hasn’t happened. And in 2017, Musk suggested that the first product from his Neuralink brain chip startup would be on the market “in about four years.” The first human received an implant from the company seven years later.
Musk has his hands full. Last weekend, he declared war on a Brazilian Supreme Court judge, calling for him to resign or be impeached over court orders issued against X requiring the accounts of some Brazilian users to be deleted. On Monday, news broke out testimony he gave in court about his posts on the site, in which he admitted that he “may have done more to harm the company financially than to help it.”
He added: “I guide my messaging not by what is financially beneficial, but by what I think is interesting, important or entertaining to the audience.”