The Neuralink Illusion: Descartes, Mind-Body Dualism, and Transhuman Economics
Updated: Aug 27
The Wire Inside the Skull: Neuralink’s First Human Trial
In January 2024 a surgical robot threaded sixty-four polymer filaments, each thinner than a human hair, into the motor cortex of a twenty-nine-year-old quadriplegic named Noland Arbaugh.
More than a thousand electrodes, a device sealed inside the skull and almost invisible from outside. Within weeks Arbaugh was controlling a cursor through neural patterns associated with intended movement — not exposing sentences or private thoughts. He played chess. He returned to Civilization VI. The achievement was neither magic nor trivial: an investigational implant had given him a new route into the digital world. Nobody unmoved by paralysis could begrudge the joy.
But a wire had entered the skull, and gratitude did not end the questions that followed it.
Thread Retraction: What the First Human Trial Could Not Yet Explain
In the weeks after surgery, Neuralink reported that a number of threads retracted, reducing the number of effective electrodes and degrading cursor-control performance. The company changed the recording algorithm, decoder, and user interface; performance improved and later surpassed Arbaugh’s initial level. The public record did not establish that his brain had ‘rejected’ the implant.
Technically, a success.
The image invites a story: flesh expelling an upgrade. But that story outruns the evidence. Neuralink publicly reported thread retraction; it did not establish that inflammation, scar tissue, or Arbaugh’s immune system caused it. Later procedures sought to reduce brain motion during surgery and narrow the gap between the implant and the brain’s surface.
The foreign-body response is nevertheless a real engineering problem for intracortical electrodes. Research describes inflammation and glial tissue forming around implanted materials, processes that can affect recording stability. That general risk does not diagnose what happened to one participant.
The cleanest conclusion is less absolute. Intracortical electrodes must remain functional inside living, moving tissue; mechanics, inflammation, materials, and software can all matter. Neuralink changed its surgical technique for later participants to reduce brain motion and the gap between implant and brain surface. Consent does not make those uncertainties vanish. Tissue is not neutral — but neither is every setback an immune rejection.
In Mamoru Oshii’s Ghost in the Shell, Major Kusanagi — fully prosthetic, combat-optimized — dives alone into water too deep for her shell because the risk of drowning is the last sensation that registers as hers.
The upgrades succeed so completely that she cannot locate the edge where she ends and the machine begins. Arbaugh’s threads retract; Kusanagi dives. Placed together, they form an analogy, not a diagnosis: both force the question of where agency begins and technology ends. Arbaugh’s case does not prove that his body said ‘this is not me.’ It proves that a first human implant met biology the engineering still had to learn from.
The Cartesian Bet Against the Body: Why Descartes Was Wrong About the Room
The Cartesian inheritance is simple enough to state and ruinous enough to build a company on: the mind and the body are separate substances. The body is mechanism. The mind is something else — immaterial, indivisible, the sovereign tenant in a mechanical house.
For four centuries this division has governed how the West speaks about consciousness, and the translation into Silicon Valley’s vernacular is immediate: the brain is hardware, the soul is software, upgrade the hardware and the software improves.
Read philosophically, Neuralink can look like a Cartesian bet dressed in silicon: treat one class of neural activity as a signal, decode it, and reroute intention through another port. The current PRIME Study has a narrower purpose — evaluating safety and functionality while enabling people with severe paralysis to control external devices. The company’s public language places a broader ambition beside it: restore autonomy today and unlock human potential tomorrow. That tension is real. It does not prove that the medical trial is a pretext.
Flesh is not a peripheral. It is the processor’s native language.
Antonio Damasio’s work offers a narrower but still powerful challenge to disembodied rationality. Studies of people with ventromedial prefrontal damage found that conventional intellectual abilities could remain while real-world judgment and affect-guided decision-making were impaired.
Some could describe their options yet repeatedly made damaging choices in ordinary life. The somatic-marker hypothesis argues that body-linked emotional signals help mark possible outcomes with value. It remains a scientific account shaped by multiple studies, not proof that every decision is a secret vote from the body. The philosophical point survives: reason is not necessarily purified when affect is erased.
Neuralink’s current system makes a deliberately narrow engineering move: neural activity associated with intended movement is decoded into cursor or device control. That abstraction is useful; it is not the same as claiming that the machine has read a wish whole, and the public trial materials do not establish that the interface changes the phenomenology of intention.
Damasio’s work cautions against treating a measurable signal as the whole mind. But the current PRIME system records and decodes motor-related activity; it does not feed Grok’s language or other interpretations back into the cortex. Future bidirectional interfaces may raise that question. Arbaugh’s present implant is not evidence that the answer has already arrived.
The wire changes the practical conditions under which intention becomes action: a movement once blocked can become cursor motion through a company-built decoder. That is enough to raise questions of dependency and agency. It is not evidence that the implant has altered Arbaugh’s mind itself.
The Assisted Sentence: When the Interface Meets Prediction
By September 2025, Neuralink reported that twelve people worldwide had received implants and accumulated more than fifteen thousand hours of use. One participant, Bradford Smith, lives with ALS and cannot speak. His implant decodes intended movement so that he can control a computer cursor; it does not decode words directly. He used that control to type, post replies, and edit a video.
Smith also used AI tools at the application layer. An AI clone of his earlier voice could speak written text, and Grok helped draft some replies from material he supplied. Smith said that he remained responsible for the content while AI assisted with the wording.
That distinction matters. Neuralink helped Smith move a cursor; Grok proposed language; Smith selected or accepted the output. There is no evidence that the chatbot completed thoughts before he formed them, and no reason to erase his authorship merely because tools mediated it. The better question is narrower: when software supplies the exact phrasing, how clearly can its contribution be inspected, rejected, and disclosed?
Brain-computer-interface decoders estimate intended control from noisy neural signals. Predictive keyboards and chatbots may separately help turn selections into language. Those layers can interact, but they are not identical, and the current Neuralink interface is not a semantic thought-reader.
The line between assistance and authorship is therefore not a single gradient inside the skull. It crosses several layers: neural decoder, cursor, keyboard, predictive or generative software, and human choice. The output alone may conceal their relative contribution, but that ambiguity is a design and transparency problem, not proof that the user has disappeared.
When an AI system offers a complete sentence, it participates in expression; so do autocorrect, translation, and editing, though not to the same degree. The ethical task is to preserve the user’s ability to know what was generated, reject it, and claim only what he endorses. Smith’s own distinction — responsible for the content, assisted in drafting — is exactly the boundary an honest interface should keep visible.
There is no public evidence that Neuralink records Grok completions as decoder-training examples, retrains its brain-computer interface on generated prose, or allows the system to learn recursively from its own sentences. The defensible concern is prospective: a commercial interface should disclose what data trains each component, keep generated language distinguishable from neural control, and give the user meaningful control over both.
The Regulatory Door: From Therapeutic Trial to Possible Augmentation
The N1 is wireless and rechargeable. Its performance depends on the implant, decoder, application, calibration, maintenance, and continuing study support. Those dependencies create a durable relationship between participant and sponsor — a surface where neural activity becomes digital control. The dependencies compound.
Participants enter an investigational study through informed consent and a clinical and regulatory process, not an arrangement nobody signed. Yet consent to a long experiment does not settle what dependency would mean in a commercial product: support can end, hardware can fail, a decoder can change, and a company can close. Architecture matters, but it begins with documented consent rather than its absence.
The Dive: Kusanagi, Arbaugh, and the Last Negotiation Between Person and Machine
To speak of bodily sovereignty once meant something close to integrity — the body as a perimeter the self controlled, a territory in which consent had meaning because the territory could be defended. The wire changes the geometry. Part of digital agency now depends on an implant and vendor-designed software. Public sources do not justify claiming that every signal travels through company infrastructure. The sharper question is what control a user retains over the device, updates, data, and continuity of support if the system becomes an ordinary product.
What remains of sovereignty under those conditions has been a philosophical question since Spinoza.
The Neuralink era turns it into a technical one as well.
Kusanagi dives because the body has succeeded so thoroughly that she needs the weight of the deep to remember she is, underneath the engineering, still someone. The dive is the last negotiation between a person and the machine that has replaced every part of her except the part that knows the difference.
Arbaugh has described long daily use and life-changing autonomy. The device is rechargeable, and Neuralink continues to update the system. The company has also publicly named further medical programs — robotic-arm control, speech restoration, and a visual prosthesis — so it is false to imply that the next generation has no declared purpose or that patents reveal a concealed one. The legitimate question is how those therapeutic programs may eventually meet the broader ambition to unlock human potential.
Neuralink publishes general and Patient Registry privacy notices, while the public pages reviewed for this essay do not expose the complete study-specific data terms. An outside reader should not turn that gap into a claim that Arbaugh has no consent or data agreement. What can fairly be asked is whether future users will receive clear answers about collection, retention, model training, sharing, and device support.
Neural data does not sit outside all law. The FDA has specific guidance for investigational implanted brain-computer interfaces, and PRIME is registered as an early-feasibility study. Colorado and California now expressly include neural data within privacy statutes. Coverage remains fragmented and fact-dependent: medical-research rules, consumer privacy laws, and contracts can apply differently depending on the actor, jurisdiction, and use.
Public materials establish that the implant and its software process neural activity into device-control commands. They do not establish that Arbaugh’s raw signals are transmitted to company servers, sold, or routed toward undisclosed destinations. The future risk is still serious: neural data could become a new commercial category unless rules specify purpose, access, retention, security, portability, and deletion before the product scales.
A therapeutic trial can prove a technical possibility whose later uses exceed its first indication. That possibility deserves governance, not a claim that today’s patient is merely a wedge.
PRIME is not compassionate use. It is an FDA-authorized early-feasibility study designed to assess device and procedure safety and functionality in people with severe paralysis. That distinction matters: an investigational study does not establish general safety, medical approval, or elective-augmentation use. Neuralink’s broader future language justifies scrutiny, but it does not make the current participants camouflage for a hidden project.
Medical technologies sometimes expand into new indications, but expansion is neither universal nor automatic. Each new use can change the expected benefits, risks, and evidence required.
A device tolerable enough to continue studying in a small paralysis cohort has not thereby been proven safe for healthy people seeking enhancement. A different population, endpoint, and tolerance for risk would require separate evidence and regulatory decisions.
The next generation may remain therapeutic, move toward elective augmentation, or divide into both; the public record cannot settle that future. The wire entered Noland Arbaugh’s skull within a study intended to restore digital autonomy. The question that follows it is not whether his benefit was a trick. It is who comes next, for what indication, under whose evidence, and with what right to say no before the wire is already inside.
Comments