Elon Musk wants to master the universe with xAI

But set aside the usual Musk eccentricities and it looks like he could make life difficult for Google in the battle to develop cutting-edge AI.

Musk’s description of the AI challenge might sound familiar to the team at DeepMind, the artificial intelligence lab that Google bought in 2015. DeepMind’s founder Demis Hassabis was the first AI entrepreneur to openly state that he wanted to build artificial general intelligence (AGI) back when he co-founded DeepMind in London in 2010, and he has consistently described his motives in abstract, even philosophical terms.  His Twitter bio today still reads, “Working on AGI. Trying to understand the fundamental nature of reality.”

AGI was an odd thing to aim for 12 years ago. But Hassabis has stayed true to the goal, saying from the beginning that DeepMind wanted to “solve intelligence, and use that to solve everything else.” That doesn’t mean focusing only on practical problems in areas like drug discovery but addressing more abstract ones too.

In fact, the desire to tackle both practical and theoretical questions seems to fit with certain fringe philosophies coming into vogue in Silicon Valley in the midst of the latest AI craze. The transhumanist movement, for instance, poses that technology can greatly enhance humans’ physical and mental limitations to help them live longer. But such augmentations can also help answer our deepest questions. Some transhumanists who work in AGI don’t just want to solve human problems but find a path to living longer, too.

Figuring out how the universe works means some scientists can “transcend the human condition,” says Emile P. Torres, a philosopher and historian studying the nature of human extinction. “The idea is that once you have super intelligence, then all our problems are solved. You could solve the problem of aging, for instance.”

With its more scientific approach to building AGI, DeepMind has marked out a different approach than OpenAI. The Microsoft-backed ChatGPT creator has a much higher proportion of engineers on staff and has addressed the goal of building AGI in more practical terms. OpenAI products, which also include image generator DALL-E, have been released to the public with an eye toward generating significant revenue.

Musk has given few details about xAI, but it sounds like he wants to follow the more theoretical path taken by DeepMind. He has convened a group of scientists that have some AI street cred — for instance, building important techniques like the Adam optimiser, which improves the speed and performance of AI models — but the team itself is small, and most aren’t well known in the field. “[It’s a] very junior team,” said one AI startup founder.

One thing is more certain: xAI represents a growing talent problem for Google as it tries to keep up in the AI race.

Half the 12 AI researchers that Musk announced for xAI had previously worked at Google. Like the eight researchers who created the important Transformer technique that made ChatGPT possible, who have all since left Google to start their own companies, these scientists apparently didn’t see a long-term future at Google. The company has developed a reputation for being slow to execute on innovative ideas, thanks to many layers of management that require signoff.

Many employees from Google Brain, the company’s other AI lab that recently merged with DeepMind, have left to join startups like Adept, Cohere and Inflection, and more AI scientists have been leaving DeepMind, too, in the past year, former staff of both companies have told me. If Google’s bloated bureaucracy was causing them grief, the explosion of interest in generative AI will have pushed many of them over the edge.

Given how Musk treated the Twitter workforce, cutting 80 percent of that company’s staff, setting erratic targets and trash-talking employees via tweet, it’s hard to see why the best AI scientists would want to work under the Tesla billionaire. But Musk has picked the right time to poach talent from Google, with many eager to see their work put into action. He also has plenty of data — thanks to Twitter — and powerful chips from Tesla on which to train his new AI models. If he also borrowed the “universe” line from Google’s DeepMind, that will sting all the more.

Parmy Olson is a Bloomberg Opinion columnist covering technology. Views are personal and do not represent the stand of this publication.

Credit: Bloomberg