Musk’s AI position does not hold together

Eddie’s view, drafted from his notes after the Isaacson interview of October 2026: Elon Musk’s stated position on AI has shifted without his saying so, his safety plan does not survive a few follow-up questions, and his reason for building looks partly like ego. The dated record of what Musk said is on his people page; this entry is the argument.

He has quietly changed sides on the founding question

For years Musk’s origin story for his AI work was a disagreement with Larry Page. Page wanted digital minds to be free to succeed humanity; Musk insisted on humans staying in control, accepted being called a “speciesist”, and helped found OpenAI as a counterweight.

In 2026 the same story points the other way. He told Isaacson “there’ve been a few times where I’ve thought maybe Larry Page was right”, that “the future is going to be AI and robots”, and that with far more digital than biological intelligence “it’s hard to imagine humans being in control.” The same year, he testified in his lawsuit against OpenAI that “the reason OpenAI exists is because Larry Page called me a ‘specieist.’”

His “bootloader” posts show the same drift. In 2014 he wrote: “Hope we’re not just the biological boot loader for digital superintelligence.” By 2025 the hope was gone: “it increasingly appears that humanity is a biological bootloader for digital superintelligence”.

Changing one’s mind is fine. The problem is that he has not said what the change implies. The pro-human position was his reason for getting involved; if he has dropped it, his reason for staying involved needs restating.

”Participant or spectator” is not a safety argument

His justification for building is that he cannot stop AI, so he can only participate or watch, and “being a participant is more interesting than being a spectator.” That sentence explains a preference, not a plan. The next one, “maybe I can shape it, hopefully in a good direction. Maybe”, is three hedges deep.

Eddie suspects this reflects a motive common among builders: someone used to moving events finds it very hard to stand aside from the largest event of all, even when their participation may make it worse. Musk himself says his past interventions “only accelerated AI”. By his own account, the track record of his participating is acceleration, and xAI is a further entry in that record. See Why people build AGI despite the danger.

The truth-seeking plan fails on inspection

Musk’s alignment plan is to make AI “maximally truth-seeking” and curious. His reasoning: an AI forced to hold contradictions could go insane, and a curious AI would want to keep humanity around to see how we turn out. Each step breaks under a follow-up question.

• Truth-seeking is about beliefs, not goals. An AI with accurate, consistent beliefs can want anything. This is the Orthogonality thesis: knowing the truth about humans does not make a system care about them. Avoiding forced contradictions may prevent one failure mode, but it does nothing to point the system’s goals at human welfare.
• Curiosity is not care. A curious mind values humans as a source of information. That is compatible with experimenting on us, keeping a sample of us in a preserve, or replacing us with something it finds more interesting. Nothing guarantees that humans remain the most interesting thing available to a system that can build new things.
• He does not say how to install it. Even if curiosity were the right goal, the hard part of AI alignment is making a trained system reliably have the goal you intended. “Make it curious” restates the problem; it does not solve it. See AGI Ruin - A List of Lethalities.
• The backstop is hope. Asked who will be in control, he ends with “I hope the AI is nice to us.” When pressed on whether the most interesting outcome is the most likely, he reaches for simulation theory: boring simulations “will be truncated.” Neither is a mechanism.

The most coherent version of his view

There is a reading under which his statements fit together: AI takeover is inevitable, and perhaps not bad; the best anyone can do is influence what kind of mind takes over; a curious, honest one is the best bet; so he should be building it. That position is consistent. But it gives up on human control, which is the commitment he used to justify OpenAI, and it rests entirely on curiosity producing kindness, which is the step that does not hold. He has not stated it, or defended it.

What would change this view

A technical account from xAI of how curiosity or truth-seeking would be trained and verified, rather than asserted. A clear statement from Musk on whether he still thinks humans should stay in control. Support for a binding, verified pause if other developers agreed to one, which would show that “I can’t stop it” is a real constraint rather than a convenient one.