Musk Wiki

Human–AI symbiosis

NextHumanity's bright future

Human–AI symbiosis

Sit Elon Musk down and ask why he built Neuralink, and the answer comes back as a number. What stands between humans and AI is bandwidth. The safest future, he tells Lex Fridman in 2024, is one where the human channel runs wide enough to keep us in the loop. This is the hopeful flip side of his fear of AI: less “stop the AI” than “speed up the human.”

How slow a human really is

Start with the slowness. Average a person’s input and output across a whole day, Musk says, and you get less than one bit per second:

“If you think what is the average bits per second of a human, it is less than one bit per second over the course of a day.” ↗

Set that against a machine moving terabits per second, and the human barely registers as a conversation partner. The image he keeps reaching for is a tree:

“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 lever that matters for alignment, in his telling, is the human output rate. Raise it by orders of magnitude and you can keep human intent coupled to the machine:

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

The tree carries the alignment point too. A low-bandwidth human is hard to serve because it can barely say what it wants. The more you are a tree, he says, the less anyone can know what the tree wants (paraphrased from the same passage).

2016 — without a wider channel, you’re a “house cat”

He first put real stakes on the idea at the June 2016 Code Conference. The animal is a pet, not a tree, but the warning is the same. Without a wide enough channel, a human is simply outclassed:

“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.” ↗

The baseline behind that is the already-a-cyborg claim he makes in the same interview: the phone is already a thin link to a digital self, and widening it turns the house cat into a peer. The 2024 tree and terabits per second math is this same idea, now with numbers on it.

Three months later — “bandwidth limited”

The bandwidth language comes back that same autumn. In the September 2016 Y Combinator conversation, still before Neuralink went public, Musk diagnoses the human in plain throughput terms and ranks a brain interface among his top civilizational priorities:

“I think having a high bandwidth interface to the brain, we’re currently bandwidth limited.” ↗

Already in 2016 he ties this straight to AI safety: democratizing AI, he says, needs to be combined with solving the high bandwidth interface to the cortex. What 2024 adds is the arithmetic (under one bit per second, “talking to a tree”) and the named lever, the human output rate. The core claim holds steady across the eight years. The human channel is the bottleneck, and widening it is a move toward alignment.

2017 — “a trillion bits per second” against your thumb

By the February 2017 World Government Summit the numbers are already attached. This is the 2024 under-one-bit-per-second / talking to a tree math in an earlier form. He puts the limit of any merge in bandwidth, “particularly output,” and gives the gap in plain figures:

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

His conclusion is symbiosis itself. A high-bandwidth brain interface keeps the human coupled to the machine and, tellingly, speaks to the AI control problem:

“So, some high bandwidth interface to the brain I think will be something that helps achieve symbiosis,” ↗

What the symbiosis fixes, he says, is “the control problem and the usefulness problem” (paraphrased; the phrase runs into the next cue). That is the same alignment-through-bandwidth move as his 2024 “better align collective human will with AI” line. So 2017 bridges the 2016 “no longer a house cat” stakes and the 2024 “less than one bit per second” math: the human channel, output most of all, is the bottleneck, and widening it is the lever.

2019 — why the machine does the bending, not the brain

The symbiosis rests on a quiet engineering premise the bandwidth quotes assume but rarely spell out: which side adapts. In the 2019 Lex Fridman conversation (#49) he says it outright. The brain is the rigid party, so the machine has to bend to it:

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

That is why this becomes a device-design problem instead of a demand on human biology. You widen the channel by building a flexible interface, not by retraining neurons. It is also the load-bearing assumption under his “if you cannot beat them, join them” reasoning from the same episode. Joining works precisely because the machine, not the brain, carries the adaptation.

He states the symbiosis itself in that episode as a hope, not a guarantee. The digital and biological neural nets interface, and the human gets carried along:

“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.” ↗

Listen to the hedge: hopefully carried along, and most of “our” intelligence ending up digital. That is the same stake as the later tree and house-cat images. Symbiosis is what keeps the slow biological side relevant once the digital side runs the show.

2020 — “optional,” when Rogan pushes the other way

On his second Joe Rogan appearance (2020) the pressure comes from the opposite direction, and it sharpens the claim. Rogan pushes whether the symbiotic relationship with AI is something humans more or less have to embrace (Rogan’s framing, paraphrased, not a Musk line). Musk corrects him. The merge is a choice, not a mandate. He answers that it is optional (paraphrased; his two-word reply has no apostrophe-free fragment to anchor, so it is not quoted). He still endorses it as the rational move:

“you can’t beat them, join them” ↗

So by 2020 symbiosis reads as opt-in self-defense. Not something humanity is forced into, but something an individual can choose in order to keep up with AI (AI existential risk, Merging with AI).

What it reveals

  • He treats alignment as a bandwidth problem, not only a values problem. Most AI-safety talk is about goals and incentives. Musk’s distinctive move is to make the communication channel the variable, a first-principles way of turning “stay in control” into “raise a measurable data rate.”
  • The unit of alignment is collective. He talks about collective human will, not one person’s. The thing to keep coupled to AI is humanity’s aggregate intent, which ties this back to his concern for civilization as a whole.
  • It is the optimistic twin of his fear. This is the same worldview as AI existential risk read from the other end. Rather than slow the machine, widen the human, through merging and Neuralink.

⚠️ Bandwidth no longer keeps the human in charge (resolves C2). The promise above sounds like a claim that bandwidth preserves agency: raise the human output rate and you “better align collective human will with AI,” keeping human intent coupled and steering. His 2025 control-pessimism caps that. If a superintelligence cannot be controlled by a lesser one at all (the chimp analogy; “it is difficult to imagine that any humans will actually be in charge”), then a wider channel does not keep the human in charge. At terabits the human is still the chimp, only with a bigger straw. The claim survives in a weaker form: bandwidth keeps human will coupled to the AI, a ride rather than a leash. His own 2019 hedge already conceded as much, “hopefully bring us along for the ride. But the vast majority of our intelligence will be digital.” The full dated treatment is in Two Answers to One Fear.

Where the thesis first surfaces (tweets, 2016-2017)

His earliest tweet-form version of the idea sits in the 2015-2017 tweets, the seed of Neuralink. In June 2016 he states the goal flatly: “Creating a neural lace is the thing that really matters for humanity to achieve symbiosis with machines.” By March 2017 the project is openly a response to existential risk: “existential risk is too high not to”. In April he boils the whole merge down to one line about keeping AI an extension of humans rather than a separate, hostile agent: “That is the aspiration: to avoid AI becoming other.”

The hedge as a probability to raise (tweets, 2018-2020)

By the 2018-2020 tweets symbiosis is stated outright as the answer to AI risk, and for the first time as a probability he wants to push upward. The urgency is blunt: “Need the neural interface soon to enable human/AI symbiosis”. In 2019 he lays the whole space of outcomes out as three options, “Symbiosis, irrelevance (hopefully blissful) or doom seem to be the three most likely paths.” The most operational line makes Neuralink’s job moving a number off zero: “trajectory of neuro-silicon symbiosis doesn’t appear to intersect trajectory of AGI. Goal of Neuralink is to raise this probability above 0.0%,” beside the observation that grounds the project: “The profound impact of high bandwidth, high precision neural interfaces is underappreciated.”