A world of multiple generative AIs with their own agendas appears to be the offing as Elon Musk enters the artificial intelligence arena with a new company called xAI – with a bold, broadly encompassing mission.
“The goal of xAI is to understand the true nature of the universe,” according to a mission statement on the company’s website (x.ai).
Twitter-owner, Musk, hosted a Twitter Spaces session on Friday, June 14 to introduce a formidable top-tier team associated with his new venture. Collectively, the team has worked at Deep Mind, OpenAI, Tesla, Google Research, Microsoft Research and the University of Toronto. Musk was a co-founder of OpenAI before leaving its board in 2018. Including Musk, the all-male team numbers a dozen people, bringing an image of the apostles to mind. The line-up reflects Musk’s desire for a small team with lots of autonomy. A notable advisor is Dan Hendrycks, the executive director of the Center for AI Safety, which focuses on the risks associated with AI. Musk said data from Twitter and Tesla would be leveraged to create xAI.
The Friday Twitter Spaces session could be best described as meandering. Musk wondered, “Why is there not massive evidence of aliens?” which some interpreted as the type of subject xAi would investigate. In the next breadth, Musk noted that his vision of the universe is one focused on the doings of humanity, noting that Earthlings are far more interesting than Mars. The xAI mission statement, joked Musk, will be: “What the hell is really going on?”
Musk said his goal is to “build a good AI,” that is “maximally curious, maximally truth-seeking.” The Musk AI would join a field dominated by OpenAI but which also includes Google and Athropic, with another by Meta on the near horizon. A more pointed comment came from co-founder and OpenAI veteran Kyle Kosic: “The reason I’m particularly excited about xAI is that I think the biggest danger of AI is monopolization by a couple of entities.”
One key goal of xAI is to develop an automated mathematician that could solve challenging math problems on its own. Greg Yang, another co-founder said xAI would delve into the “mathematics of deep learning” to develop a “theory of everything” for large neural networks that would advance AI development. Observers noted that availability of mathematical data required to train an AI math whiz is nowhere near as robust or widespread as that used to develop language AIs, however. According to Cryptopolitan, AIs like ChatGPT have been trained using trillions of lines of code while the amount of formalized mathematical data now available is estimated at about 10 million lines.
Given Musk’s past proclivities, it’s also possible that politics is on the agenda. Today’s xAI company grew out of an AI project with the Orwellian-sounding name of TruthGPT. Musk thought ChatGPT was too politically correct. By contrast, Musk says xAI won’t be trained in political correctness and xAI will be allowed to say what it believes to be true—a stance Musk believes will generate criticism.
Whether xAI leans left or right, or neither, remains to be seen, but Oren Etzioni, former head of the Allen Institute for Artificial Intelligence who coined the term “machine reading,” thinks AIs with differing points of view are likely. “In the end, these models are primarily mirrors,” said Etzioni last week in a Der Speigel interview. “When we hold these mirrors up to our society, we see Republicans, Democrats and all the unsavory characters too.”