BCI and the Consciousness question

An earlier version of this essay began with the James Webb Space Telescope and ended, in a state of high excitement, with the claim that humanity was building its first telescope pointed inward. The image was right and the tone was wrong. Enthusiasm is cheap; a report is harder. So here is one. The inward telescope now exists in the sense that matters: threads of metal finer than hairs have been placed inside living human brains and made to carry useful signals out of them. The question a serious person should ask is not whether that is astonishing — it is — but what the signals actually contain, what they might one day contain, and what the marketing already pretends they contain. These are three different things, and most writing about brain-computer interfaces blends them into one glowing blur.

What the threads actually read

The earlier essay dismissed robotic arms and thought-typing as the "kindergarten" of this technology. That had the difficulty backwards. Making a machine infer spoken words from neural activity is not kindergarten; it is among the hardest things anyone has attempted with the brain. In August 2023, two teams reported striking but distinct single-participant results. At Stanford, Willett and colleagues used intracortical microelectrode arrays in a man with amyotrophic lateral sclerosis to decode attempted speech at 62 words per minute — 3.4 times faster than the previous BCI record in their comparison. At UC San Francisco, Metzger and colleagues used high-density surface recordings from the speech cortex of a woman paralysed by a brainstem stroke. Their text decoder reached a median 78 words per minute with a 25 per cent median word error rate; the system also synthesised speech and controlled a facial avatar. The two figures describe different participants and systems, not a single standard speed for BCI (Willett et al., Nature, 2023; Metzger et al., Nature, 2023).

The denominator matters: each 2023 study involved one participant. Their data demands were substantial but different. The Stanford study collected tens of minutes of training data on each evaluation day, combined it with prior sessions and retrained the decoder daily. The UCSF team reported high performance with less than two weeks of training. Both results depended on neurosurgically placed electrodes; neither demonstrates a ready-made, general system that can be transferred unchanged to another person. Neuralink's PRIME study, which the company announced with its first human implant in January 2024 and reports on through its updates page, is likewise an early-feasibility trial rather than mass-market evidence (Neuralink updates).

The arithmetic of slowness

One comparison is useful if it stays attached to its assumptions. Willett and colleagues use about 160 words per minute as a natural-conversation benchmark. The Stanford participant's 62 words per minute is about 39 per cent of that rate; the UCSF participant's median 78 words per minute is about 49 per cent. These are results from two single-participant studies, not an average across users or a prediction of everyday communication. The Stanford paper reports a 23.8 per cent word error rate for its large-vocabulary result; the UCSF paper reports a 25 per cent median error rate for text decoding. Speed without accuracy is not fluent conversation, and each study's reported error rate belongs to its own system and test design.

The mirror, examined

The middle of the earlier essay contained its one genuinely important idea: a brain-computer interface regarded not as a controller but as a mirror — a device that shows you the state of your own mind the way a bathroom mirror shows you your face. Is there any evidence such a mirror can be built?

A little. In 2018, Sani and colleagues reported in Nature Biotechnology a framework for decoding mood from multi-site intracranial recordings in seven people with epilepsy, who had electrodes implanted for clinical seizure monitoring and who repeatedly rated their own mood while the electrodes listened. The model tracked those self-ratings from the neural activity. Seven people; mood defined by each person's own reports; electrodes placed where the seizures were, not where serenity is thought to live. A mirror, but a small and clouded one (Sani et al., Nature Biotechnology, 2018).

The next step toward the mirror is stranger. In 2021, Scangos and colleagues described in Nature Medicine a single individual with treatment-resistant depression fitted with a device that sensed a neural biomarker of the depressive state and stimulated a targeted circuit only when that biomarker appeared — a closed loop between reading and acting. The authors report rapid and sustained improvement over the observation period. One person. The paper does not hide this; the celebratory coverage usually does (Scangos et al., Nature Medicine, 2021).

Now look soberly at what a "neural signature of peace" — the earlier essay's phrase — would actually require. It would require a definition of peace that holds across people rather than one meditator's diary. It would require neural signatures stable enough to generalise from one skull to another. And it would require sensing that does not involve opening the skull, because scalp electrodes read the smeared electrical weather of billions of neurons through bone and tissue, and carry vastly less information than the implanted arrays behind the 62-words-per-minute result. None of these problems is solved. The earlier essay also asked readers to hear "the sound of your own compassion" in "the coherent electromagnetic field generated by your heart." The heart does produce a faint field, measurable at short range with sensitive instruments, and it is not a readout of compassion. That sentence was poetry wearing a lab coat, and it should be retired.

The commercial fog

Between the laboratory and the temple stands a market. Consumer headbands now sell meditation scores and focus scores derived from scalp electrical signals. No device on the open market reads thoughts; the physics of the skull forbids anything like the resolution the implanted devices achieve. What the consumer devices can do is infer coarse arousal and attention from noisy signals and paint the inference in calm colours. Whether that helps anyone meditate is a fair question; whether it shows "the entire, multi-layered state of your being," as the earlier essay hoped, is not supported by anything published. The claim to beware is not that these devices do nothing, but that they borrow the vocabulary of the implanted-electrode papers they cannot replicate.

Who owns the mirror

If a mirror of the mind is ever built, the first question will not be philosophical but proprietary. The world's governments have begun, belatedly, to notice. In November 2025 the general conference of UNESCO, meeting at its 43rd session, adopted the Recommendation on the Ethics of Neurotechnology — the first global standard-setting instrument on the subject, covering the technology's whole life cycle and naming mental privacy, informed consent and human agency among its principles. It is a recommendation, not a treaty; it binds nobody and enforces nothing. It is a beginning, which is what recommendations are (UNESCO, 2025).

Kenya's position is worth stating precisely. The Data Protection Act of 2019 established the Office of the Data Protection Commissioner, and the Commissioner's Guidance Note on the Processing of Health Data, issued in December 2023, treats health information as sensitive personal data carrying heightened obligations of consent and security. Whether raw neural recordings — the chatter of electrodes rather than a doctor's diagnosis — fall under "health data" is a live question that the guidance, drafted before consumer neurotechnology spread, does not settle. Until it is settled, the safest reading is that neural data deserves at least the protection owed to health data, because it can be made to reveal health (ODPC Guidance Note).

Evolution by intention

The earlier essay closed by naming the destination: Homo Coherens, the coherent human, whose heart and mind play in tune. Kept as a belief, that is respectable; mistaking it for a forecast is not. So, sorted honestly:

By evidence, we know that neural systems have decoded attempted speech at 62 and 78 words per minute in two single-participant studies; that one research model tracked self-reported mood in seven epilepsy patients with clinically placed electrodes; and that a sensing-and-stimulation system was tested in one person with treatment-resistant depression.

By scenario, a society that regulates its nervous systems the way it now regulates blood sugar is imaginable, but it depends on assumptions — about safety, cost, implant longevity and who pays — that nobody can yet test.

And by belief, stated as belief, the earlier essay's central conviction stands: that the mystics who said the kingdom is within were giving a location, not a metaphor, and that the inward journey is the one worth building instruments for. A belief is not weakened by being called a belief. It is weakened by being dressed as a finding.

The lamp in the workshop

The James Webb telescope shows us the birth of galaxies more than thirteen billion years ago. The inward telescope, at present, resolves syllables from a single skull, slowly, with errors. Before dismissing the comparison, remember that the first optical instruments were crude, and that what made them matter was not their power but their direction. The great work was always an inside job. The workshop now has a lamp in it. The lamp flickers, it lights one corner of one bench, and everyone who has tried to build by its light says the same thing: it is enough to work by, and not nearly enough to see by.

An EEG-based brain-computer interface experiment photographed on 16 July 2013. This is a non-invasive cursor-control setup, not an implanted device or a direct picture of consciousness. Credit: Laurens R. Krol. Image source · CC BY 4.0 International