ELON MUSK, not content with helming recent purchase Twitter alongside SpaceX and his other long-standing firms, has announced an artificial intelligence start-up called xAI. People have speculated that it might be an attempt to challenge OpenAI’s ChatGPT, an AI-powered chatbot that has grown to 100 million monthly users in the blink of an eye. But a veil of mystery hangs over the venture, whose goal is “to understand the true nature of the universe”. Musk isn’t averse to grandiose statements and marketing bluff – a SpaceX mission to Mars is seemingly always just on the horizon – but a well-funded firm tackling the big questions could be just what AI needs.
Academics in the field complain their ability to do experiments and work with the smartest minds is hampered by the private sector, with investors throwing billions of dollars at promising start-ups. Universities can’t compete with the salaries or computational firepower of companies like Microsoft, so big-money moves out of public institutions are commonplace. This has arguably become worse since the release of ChatGPT. The move signalle d to companies like Google that, instead of doing fundamental research behind closed doors, the priority should instead be commercialisation, which led to the release of the Bard chatbot.
Advertisement
Although academia has its own idiosyncrasies and research preferences, it is still less attached to direct commercial interest than big tech and more open to science. But if resources are pooled in firms, then basic questions may never get looked at, let alone solved.
So when I read that Musk was starting an AI company that might take on the big questions, I felt a glimmer of hope. Putting aside his sometimes bizarre online antics, consistent overpromising and flirtations with fringe right-wing politics on Twitter, Musk has a successful track record of running companies, namely SpaceX and Tesla. Maybe xAI will succeed.
Information is scant on what, exactly, the 12 researchers Musk has hired will be working on, so I decided to look at their past research to get a better sense of the company’s possible future.
Last year, I interviewed one of the team, Yuhuai Wu, for his work on using AI to translate maths problems from plain English into a machine-readable format. Wu, working at Google at the time, was supervised by Christian Szegedy, now also at xAI, who oversaw a team looking at how to solve maths problems with AI more generally. Szegedy has predicted that a computer will match the capabilities of the best human mathematicians by 2026.
Other xAI employees, like ex-Microsoft engineer Greg Yang, are experts in the underlying mathematics of neural networks, the architecture that powers AIs like ChatGPT, or have worked on creating more general-purpose AIs, like Jimmy Ba, whose self-described research goal is to build “general problem-solving machines with human-like efficiency and adaptability”.
Not all of Musk’s new employees deal with the theoretical. Some, like ex-DeepMind employee Guodong Zhang, were involved with more commercially applicable ventures, such as the Gemini chatbot – DeepMind’s cross between GPT-4 and problem-solving AIs like its chess and Go-playing AlphaZero. The xAI team is also advised by Dan Hendrycks, the director of a non-profit organisation that focuses on AI safety.
Together, the team has the technical nous to probe questions such as whether maths can be conquered by machine. Musk’s interest in moonshot projects – like human-machine interface start-up Neuralink – is a bonus, too. xAI will work closely with Twitter and Tesla, whose massive computational resources are likely to be far beyond what any university might be able to offer.
But it might all go horribly wrong. Musk often expounds confidently on matters of reality in which he is no expert. Last week, as well as announcing he would rename Twitter as X, Musk responded to a proposal that the universe is twice as old as thought, writing on Twitter that “dark matter is what seems most sketch to me”. Astrophysicists quickly responded that dark matter is one of the most well-supported, though unexplained, phenomena in all of physics.
Understanding the “true nature of the universe” might just be cover to build a chatbot that is free to be politically incorrect. At a launch event for xAI, Musk said that AIs, like ChatGPT, could be “incredibly dangerous” because false outputs might lead to real-world harm. The only way to counter this harm, he says, would be to build a “maximally truth-seeking” system. If this is what Musk means by understanding the universe, then the future will be a lot less interesting.
Alex’s week
What I’m reading
The Door by Magda Szabo, a brilliant, mysterious novel about the relationship between a writer and her housekeeper in Hungary.
What I’m watching
I have been glued to the summer’s sporting events: The Ashes, Wimbledon and the Tour de France.
What I’m working on
I have been following up on research leads from the Goldschmidt geochemistry conference in Lyon, France.
