Musk Wiki

Two Answers to One Fear

NextIndex

Two Answers to One Fear

Musk has said out loud since 2014 that a digital superintelligence is a civilization-level danger. Out of that one fear he builds two fixes at once, and he never lets go of either. One is political and moral: regulate the AI labs, and when that stalls, build an AI whose deepest value is telling the truth. The other is surgical: wire the brain straight into the machine through Neuralink, so a person is no longer a house cat next to the AI and never slows to the speed of a tree. He runs both at once, the fear feeding a regulatory-and-values fix and a merge through symbiosis in the same breath.

By 2025 the two fixes collide. Musk decides nobody controls a superintelligence at all: a chimp cannot control humans, and the AI is going to be in charge … not humans. That verdict quietly guts the merge’s founding promise from 2016, that we are the AI, collectively.

Summary

This isn’t a change of mind. It’s a split that never healed. From 2016 on, the same fear runs down two channels at once. The first says: slow the machine down, then steer what it values. Ask for an outside regulator, and when that fails, build a truth-seeking AI whose highest value is sticking to the truth. That side moves from warning to regulation to building to giving up control, an arc traced step by step in The Shifting Remedy. The second answer says the opposite: speed the human up. Treat alignment as a plumbing problem, widen the channel between brain and machine through Neuralink, and keep human will bolted to the AI. Push it far enough and the line between us and it dissolves, leaving no rival to control. Both answers show up almost fully formed in 2016, and he holds both for a decade.

Here’s the catch. The two answers promise opposite things about who stays in charge. The merge, in 2016, is supposed to solve the control problem: we don’t have to worry about some evil dictator AI because we are the AI, collectively. But by 2025 the policy side has hardened into a flat no. Nobody controls a superintelligence, the way a chimp controls no human, and the AI ends up in charge instead of people. If control is impossible in principle, a fatter channel buys you no control. It buys a front-row seat. On Musk’s own later logic, the merge has slid from the solution to control to a way to come along for the ride.

2016 — one fear, stated twice

By mid-2016 the fear has a sharp edge to it. The danger isn’t AI as such. It’s which future the AI delivers. At the June 2016 Code Conference:

“I think it would be fair to say that not all AI futures are benign.” ↗

In the same interview the merge answer is already bolted on. Without a wide enough channel to your digital self, the human is simply outmatched:

“If we can figure out how to establish a high bandwidth neural interface with your digital self effectively, then you’re no longer a house cat.” ↗

Three months later, the September 2016 Y Combinator conversation lays out both answers in a single sitting. First the political one. Spread AI out so no single group can hoard it, the idea that talks him into co-founding OpenAI:

“is that we achieve democratization of AI technology, meaning that no one company or small set of individuals has control over advanced AI technology.” ↗

Then the merge. The same conversation holds his cleanest answer to why you would do it: fusing with the AI is supposed to make the control problem vanish, by erasing the line between us and it.

“So if we can effectively merge with AI by improving the neural link between your cortex and the digital extension of yourself, which already exists, it just has a bandwidth issue.” ↗

“We don’t have to worry about some evil dictator AI because we are the AI, collectively.” ↗

That line is the load-bearing one. In 2016 the merge is the answer to control. So one fear gets two replies in a single year: regulate and spread it out on one side, fuse and widen the channel on the other.

2017–2019 — the merge gets its numbers, the regulator gets a name

Through the late 2010s both answers sharpen in parallel.

On the merge side, the February 2017 World Government Summit puts hard numbers behind the house cat. The gap between what a machine and a human can push out is enormous, and that output rate is the wall everything runs into:

“A computer can communicate at a trillion bits per second, but your thumb can maybe do maybe 10 bits per second or a hundred” ↗

By the 2019 Lex Fridman conversation (#49) the merge has its bluntest justification. You can’t out-think a digital supercomputer, so the smart move is to join it:

“We will not feel to be smarter than a digital supercomputer. So therefore, if you cannot beat them, join them.” ↗

The same 2019 episode gives the engineering reason the merge is even buildable. The machine does the bending, because the brain is the stiff part:

“I think the machine side is far more malleable than the biological side, by a huge amount.” ↗

But listen to the hedge already buried in the 2019 version of the symbiosis. Hopefully it brings us along, and most of the intelligence ends up digital:

“we’re a neural net and AI is basically a neural net. So it’s like digital neural net will interface with biological neural net. And hopefully bring us along for the ride. But the vast majority of our intelligence will be digital.” ↗

Bring us along for the ride, and the vast majority of our intelligence will be digital. That is the merge already admitting, back in 2019, that the human rides shotgun rather than drives. It is the seed of the 2025 reversal.

Meanwhile the political answer runs right alongside, asking for an outside referee. The same 2019 conversation names it as a government body:

“Where there is a lack of investment is in AI safety, and there should be, in my view, a government agency that oversees anything related to AI to confirm that it is, does not represent a public safety risk.” ↗

The two answers aren’t fighting here. They’re the same worry worked from opposite ends.

2022–2024 — still asking for a regulator, still widening the channel

The split holds into the 2020s. On the political side, at TED2022 he repeats the regulator request as plainly as it ever gets, and against his own interest:

“I do think there should be a regulatory agency for AI.” ↗

On the merge side, the 2024 Lex Fridman conversation (#438, the Neuralink-team episode) sharpens the argument to its hardest number yet. The human channel is laughably slow:

“If the AI can communicate at terabits per second, and you’re communicating at bits per second, it’s like talking to a tree.” ↗

So the dial that matters for alignment is the human output rate. Turn it up, and human intent stays bolted to the machine:

“We could better align collective human will with AI if the output rate especially was dramatically increased.” ↗

Here is the promise in its 2024 form: widen the human channel and collective human will stays in charge. The house cat of 2016 has become the tree of 2024, the same image eight years on, now sold as the reason to ship an actual device. The same #438 conversation also states where the political answer finally lands. Once you stop trying to regulate the AI, here is what it has to be:

“the thing that at least my biological neural net comes up with as being the most important thing is adherence to truth, whether that truth is politically correct or not.” ↗

Both answers are fully alive in 2024. Widen the human channel; instil the right value. Two replies to one fear, still carried side by side.

2025 — control declared impossible, and the merge promise capped

The last move comes from the political answer, but it lands hardest on the merge. By 2025 Musk decides that controlling a superintelligence is impossible, full stop. On the October 2025 Joe Rogan conversation he reaches for chimps and humans:

“I mean, I don’t think anyone’s ultimately going to have control over digital superintelligence, any more than, say, a chimp would have control over humans. Chimps don’t have control over humans, there’s nothing they could do.” ↗

At the November 2025 Tesla shareholder meeting he says it as flat fact:

“Well, I mean, I think actually long term, uh, the AI is going to be in charge to be totally frank, not humans.” ↗

“if if if artificial intelligence vastly exceeds the sum of human intelligence, it is difficult to imagine that that any humans will actually be in charge.” ↗

This is where the two answers crash into each other. The merge promised, in 2016, that we are the AI, collectively, that fusing escapes control because there’s no separate “it” left to rule. The bandwidth argument promised, in 2024, that turning up the human output rate keeps collective human will in charge. The 2025 verdict says no humans will be in charge either way: it is difficult to imagine that any humans will actually be in charge. A chimp with a wider channel is still a chimp. A merged human running at terabits is, by the same logic, still the smaller mind in the loop. The fatter channel never bought control. It bought, in his own 2019 words, a ride.

What’s left on the political side isn’t control but friendliness, the whole alignment program squeezed, at that same 2025 meeting, into four words:

“we just need to make sure the AI is friendly.” ↗

⚠️ The merge no longer escapes control (resolves C1 + C2). Two earlier promises said fusing with the machine keeps the human in charge. The 2016 one was we are the AI, collectively. The 2024 one was that a wider channel could better align collective human will with AI. The 2025 pessimism (the chimp; the AI is going to be in charge … not humans) says no human stays in charge at all. Follow Musk’s own later logic and the two reconcile this way: the merge has shifted from “the solution to control” to “a way to come along for the ride.” Merging and bandwidth no longer escape the control problem. They make the handoff survivable and coupled instead of hostile. The seed was always there, in that 2019 hopefully bring us along for the ride … the vast majority will be digital. By 2025 it’s the headline. The 2025 verdict caps both the merge-as-control promise and the bandwidth lever that was meant to keep human will in charge.

What the two answers reveal

  • One fear, two engineering instincts. The political answer treats safety as a question of values and oversight: regulate it, then teach it the truth. The bandwidth answer treats it as a throughput question: push a measurable data rate higher. What’s distinctly Musk is running both at once: he hedges a single worry with a political ask and a piece of hardware. It’s the same move as the master plans, which hedge a goal with a product you can ship now and a dream for later.
  • The merge was the bolder bet, and it’s the one that snaps. The political answer ages gracefully: regulate it, then out-build it, then just make it friendly (The Shifting Remedy). The merge staked out a stronger claim, that fusion dissolves control, and that claim doesn’t survive 2025. It gets demoted, not dropped. Still worth doing, but as coupling rather than control.
  • The crack was visible from 2019. Hopefully bring us along for the ride, with the vast majority of our intelligence going digital, already gave up the driver’s seat. For a decade Musk sounded sure the merge answered the control problem, yet on his own words he never quite said it without the hedge. By 2025 he simply drops the hopefully.

Connections