Brain–computer interface: revision 4 to 5

Revision 4

Revision 5

categories: ["cybernetics", "foundations"]

tags: ["bci", "neurotechnology", "paralysis", "neural decoding", "implants", "assistive technology"]

summary: "A system that measures nervous-system activity, decodes an intention or state from it, and converts that decoded signal into a command for an external device."

summary: "A system that measures nervous-system activity, decodes an intention or state from it, and converts that decoded signal into a command for an external device."

updated: "2026-10-28"

updated: "2026-10-29"

humanEvidence: "Implanted arrays have given research participants with paralysis cursor control, robotic-arm use and conversational-rate speech decoding; no implanted BCI has been shown to improve any capability in a healthy person."

humanEvidence: "Implanted arrays have given research participants with paralysis cursor control, robotic-arm use and conversational-rate speech decoding; no implanted BCI has been shown to improve any capability in a healthy person."

access: "No implanted BCI is approved by any regulator; the only route is enrolment in an early feasibility trial. Consumer EEG headsets are sold freely as wellness devices."

access: "No implanted BCI is approved by any regulator; the only route is enrolment in an early feasibility trial. Consumer EEG headsets are sold freely as wellness devices."

reversibility: "difficult"

}

```

**Brain–computer interface** (BCI) is the general term for a system that measures activity in the

**Brain-computer interface** (BCI) is the general term for a system that measures activity in the

nervous system, extracts a variable of interest from that activity, and turns it into a command for

a computer, a prosthesis, or the person's own muscles. The defining feature, in the standard

formulation, is that the command bypasses the ordinary output path of peripheral nerve and

muscle.[^wolpaw2002] A BCI is therefore useful mainly to people

muscle.[^wolpaw2002] A BCI is therefore useful mainly to people whose ordinary output path is

whose ordinary output path is broken, and as of 2026 every serious clinical result comes from that

broken, and as of 2026 every serious clinical result comes from that population: participants with

population: participants with tetraplegia, brainstem stroke, or amyotrophic lateral sclerosis.

tetraplegia, brainstem stroke, or amyotrophic lateral sclerosis.

The field has produced genuinely striking demonstrations — a person with no usable movement

The field has produced striking demonstrations in a very small number of research participants: a

controlling a robotic arm, another producing fluent synthesized speech from motor-cortex activity

person with no usable movement controlling a robotic arm, another producing fluent synthesized

alone — in a very small number of research participants, using hardware that is tethered, requires

speech from motor-cortex activity alone. The hardware behind them is tethered, requires frequent

frequent recalibration, and has not been approved by any regulator for permanent use. The gap

recalibration, and has not been approved by any regulator for permanent use. The gap between those

between those demonstrations and a product is largely a gap in reliability, longevity, and

demonstrations and a product lies largely in reliability, longevity, and manufacturing rather than

manufacturing, not in principle.

in principle.

```figure

{"key": "braingate-array", "caption": "The implanted array against a human hand. The entire recording surface is a few millimetres square."}

{"key": "braingate-array", "caption": "The implanted array against a human hand. The entire recording surface is a few millimetres square."}

## The signal chain

Every BCI is the same four stages: acquisition, feature extraction, decoding, and output, usually

Every BCI has the same four stages: acquisition, feature extraction, decoding, and output, usually

with a feedback loop closing back to the user.

*Acquisition* determines everything downstream. Signals differ by how close the sensor sits to the

neurons producing them. Penetrating microelectrodes recording action potentials from individual

cells carry the most information per channel; scalp electroencephalography, separated from cortex

cells carry the most information per channel; scalp electroencephalography, separated from cortex by

by dura, skull, and scalp, carries the least, because the skull acts as a spatial low-pass filter

dura, skull, and scalp, carries the least, because the skull acts as a spatial low-pass filter that

that smears contributions from many square centimetres of cortex into one measurement.

smears contributions from many square centimetres of cortex into one measurement.

*Feature extraction* reduces raw voltage into something a decoder can use: firing rates binned over

tens of milliseconds, power in a frequency band, the amplitude of an evoked potential. In cortical

recordings the workhorse features are threshold crossings and high-gamma power (roughly 70–200 Hz),

recordings the workhorse features are threshold crossings and high-gamma power (roughly 70-200 Hz),

the latter being a reasonable proxy for local population spiking that can be measured without

penetrating the cortex.

*Decoding* maps features to intent. Early systems used linear filters and the population-vector

approach inherited from primate motor physiology; current systems use Kalman filters, recurrent

neural networks, and increasingly transformer-style sequence models paired with [language models](https://aiwiki.ai/wiki/large_language_model).

neural networks, and increasingly transformer-style sequence models paired with [language

[[neural-decoding]] is where most of the recent performance gains have come from — the electrodes

models](https://aiwiki.ai/wiki/large_language_model). Most of the recent performance gains have come

in the best 2025 speech systems are not fundamentally better than those used in 2012.

from [[neural-decoding]]; the electrodes in the best 2025 speech systems are not fundamentally

better than those used in 2012.

*Output* is a cursor, a robotic arm, a synthesized voice, a wheelchair, or stimulation delivered to

the user's own spinal cord or to a powered [[exoskeleton]]. Bidirectional systems also write

a character grid and detects the evoked response that follows an attended flash; steady-state

visually evoked potential systems tag each option with a distinct flicker frequency and read off

which frequency dominates visual cortex; sensorimotor-rhythm systems ask the user to imagine

movement of a limb and detect the resulting suppression of mu and beta rhythms. All three work.

movement of a limb and detect the resulting suppression of mu and beta rhythms. All three work, but

None is fast: information transfer rates are conventionally reported in the range of tens of bits

none is fast: information transfer rates are conventionally reported in the range of tens of bits

per minute, on the order of a few characters, against several hundred bits per minute for competent

typing.

Functional near-infrared spectroscopy and functional MRI measure haemodynamics rather than

electrical activity, and are therefore limited by the seconds-long delay of the blood-oxygen

response. fMRI remains valuable for decoding research precisely because it covers the whole brain,

response. fMRI remains valuable for decoding research because it covers the whole brain, but a

but a scanner is not a wearable device. Time-domain fNIRS, which times returning photons to gain depth sensitivity, has been built into a whole-head wearable helmet by [[kernel]], whose first-generation Kernel Flow weighed about 2 kg and was benchmarked mainly on optical phantoms and a two-person finger-tapping task.[^ban2022] Kernel sells it for research and brain-health measurement rather than for device control. Consumer EEG headsets sit at the weakest end

scanner is not a wearable device. Time-domain fNIRS, which times returning photons to gain depth

of the spectrum, typically with a handful of dry electrodes and heavy artefact contamination from

sensitivity, has been built into a whole-head wearable helmet by [[kernel]], whose first-generation

eye movement and jaw muscle.

Kernel Flow weighed about 2 kg and was benchmarked mainly on optical phantoms and a two-person

finger-tapping task.[^ban2022] Kernel sells it for research and brain-health measurement rather than

for device control. Consumer EEG headsets sit at the weakest end of the spectrum, typically with a

handful of dry electrodes and heavy artefact contamination from eye movement and jaw muscle.

```compare

{

"rows": [

    { "label": "Surgery", "values": ["Craniotomy, cortical penetration", "Craniotomy or slot", "Catheter via jugular vein", "None"] },

    { "label": "Surgery", "values": ["Craniotomy, cortical penetration", "Craniotomy or slot", "Catheter via jugular vein", "None"] },

    { "label": "Resolution", "values": ["Single neurons", "Millimetre populations", "Coarse populations", "Centimetre populations"] },

    { "label": "Resolution", "values": ["Single neurons", "Millimetre populations", "Coarse populations", "Centimetre populations"] },

{ "label": "Typical channels", "values": ["96–1,024", "64–1,024", "16", "8–256"] },

{ "label": "Typical channels", "values": ["96-1,024", "64-1,024", "16", "8-256"] },

    { "label": "Long-term stability", "values": ["Degrades over months to years", "Good over years", "Good over years", "Good, but user-dependent"] },

    { "label": "Long-term stability", "values": ["Degrades over months to years", "Good over years", "Good over years", "Good, but user-dependent"] },

    { "label": "Demonstrated ceiling", "values": ["Multi-degree-of-freedom limb control, fluent speech", "Fluent speech, walking via spinal stimulation", "Discrete clicks", "Spellers, simple switches"] }

    { "label": "Demonstrated ceiling", "values": ["Multi-degree-of-freedom limb control, fluent speech", "Fluent speech, walking via spinal stimulation", "Discrete clicks", "Spellers, simple switches"] }

]

[

  { "year": "1924", "title": "Human EEG recorded", "text": "Hans Berger records electrical rhythms from the human scalp, establishing that cortical activity can be measured non-invasively." },

  { "year": "1924", "title": "Human EEG recorded", "text": "Hans Berger records electrical rhythms from the human scalp, establishing that cortical activity can be measured non-invasively." },

  { "year": "1973", "title": "The term is coined", "text": "Jacques Vidal at UCLA publishes 'Toward direct brain-computer communication', proposing that computers could interpret EEG in real time as control signals." },

  { "year": "1973", "title": "The term is coined", "text": "Jacques Vidal at UCLA publishes 'Toward direct brain-computer communication', proposing that computers could interpret EEG in real time as control signals." },

  { "year": "1999–2002", "title": "Motor decoding in animals", "text": "Groups led by John Chapin, Miguel Nicolelis, and Andrew Schwartz show that rats and monkeys can control levers and robotic arms using decoded cortical ensemble activity." },

  { "year": "1999-2002", "title": "Motor decoding in animals", "text": "Groups led by John Chapin, Miguel Nicolelis, and Andrew Schwartz show that rats and monkeys can control levers and robotic arms using decoded cortical ensemble activity." },

  { "year": "2004–2006", "title": "First long-term human implant", "text": "Matthew Nagle, paralysed by a spinal cord injury, receives a Utah array in the BrainGate pilot and controls a cursor, a television, and a prosthetic hand." },

  { "year": "2004-2006", "title": "First long-term human implant", "text": "Matthew Nagle, paralysed by a spinal cord injury, receives a Utah array in the BrainGate pilot and controls a cursor, a television, and a prosthetic hand." },

  { "year": "2012–2013", "title": "Robotic arm control", "text": "BrainGate and University of Pittsburgh teams report multi-degree-of-freedom robotic arm control by participants with tetraplegia, including self-feeding." },

  { "year": "2012-2013", "title": "Robotic arm control", "text": "BrainGate and University of Pittsburgh teams report multi-degree-of-freedom robotic arm control by participants with tetraplegia, including self-feeding." },

  { "year": "2016", "title": "Fully implanted communication BCI", "text": "A Utrecht group implants a wireless electrocorticography system in a person with late-stage ALS, who uses it at home for years to select letters." },

  { "year": "2016", "title": "Fully implanted communication BCI", "text": "A Utrecht group implants a wireless electrocorticography system in a person with late-stage ALS, who uses it at home for years to select letters." },

  { "year": "2021–2024", "title": "Speech decoding becomes fluent", "text": "UCSF, Stanford, and UC Davis teams decode attempted speech at conversational rates from motor cortex, with large vocabularies and single-digit word error rates in the best case." },

  { "year": "2021-2024", "title": "Speech decoding becomes fluent", "text": "UCSF, Stanford, and UC Davis teams decode attempted speech at conversational rates from motor cortex, with large vocabularies and single-digit word error rates in the best case." },

  { "year": "2023", "title": "Brain–spine interface", "text": "A Lausanne team restores volitional walking in a man with spinal cord injury by linking a cortical implant to an epidural spinal stimulator." },

  { "year": "2023", "title": "Brain-spine interface", "text": "A Lausanne team restores volitional walking in a man with spinal cord injury by linking a cortical implant to an epidural spinal stimulator." },

  { "year": "2024–2026", "title": "Commercial implants enter trials", "text": "Neuralink, Synchron, Precision Neuroscience, Paradromics, and several Chinese groups run early feasibility studies; none has regulatory approval for routine use." }

  { "year": "2024-2026", "title": "Commercial implants enter trials", "text": "Neuralink, Synchron, Precision Neuroscience, Paradromics, and several Chinese groups run early feasibility studies; none has regulatory approval for routine use." }

]

```

**Cursor and device control.** People with tetraplegia can move a computer pointer, click, type on a

virtual keyboard, and drive a robotic arm through several degrees of freedom.[^hochberg2006] The

2012 and 2013 robotic-arm reports, in which participants reached, grasped, and drank

2012 and 2013 robotic-arm reports, in which participants reached, grasped, and drank unassisted,

unassisted, remain the reference demonstrations.[^hochberg2012][^collinger2013] Performance in the

remain the reference demonstrations.[^hochberg2012][^collinger2013] Performance in the best

best participants approaches a slow but usable pointing device. The [[utah-array]] remains the most

participants approaches a slow but usable pointing device. The [[utah-array]] remains the most

common sensor for this work, and the multi-site BrainGate consortium has produced most of the

long-run data. A fully implanted surface-electrode system used at home by a person with late-stage

ALS demonstrated that the approach can survive outside a laboratory.[^vansteensel2016]

([[neuralink]], Paradromics, Blackrock Neurotech), thin-film surface arrays ([[ecog-interfaces]],

Precision Neuroscience), and endovascular electrodes threaded through a vein ([[stentrode]],

developed by [[synchron]]). Each trades bandwidth against surgical risk. Several Chinese academic

and state-backed groups have also reported implants in participants since 2024, though public

and state-backed groups have also reported implants in participants since 2024, though public detail

detail is limited and independent verification scarce.

is limited and independent verification scarce.

What has not been demonstrated is equally worth stating. No BCI has restored a rich sense of touch;

No BCI has restored a rich sense of touch; the best feedback so far is a small number of

the best feedback so far is a small number of discriminable pressure percepts delivered by

discriminable pressure percepts delivered by intracortical or peripheral stimulation, well short of

intracortical or peripheral stimulation, well short of what a hand needs. No system has restored

what a hand needs. No system has restored memory in a clinically meaningful way, despite the

memory in a clinically meaningful way, despite the encoding-model work described under

encoding-model work described under [[memory-prosthesis]]. Nor has any implanted BCI been shown to

[[memory-prosthesis]]. And no implanted BCI has been shown to improve any capability in a healthy

improve any capability in a healthy person, which is the premise on which most enhancement-oriented

person, which is the premise on which most enhancement-oriented interest rests.

interest rests.

> [!caution] Sample sizes are tiny

> The headline results in this field typically come from one to four participants, often the same

## The bandwidth problem

Popular framing treats BCI progress as an electrode-count race, on the assumption that information

transfer scales with channels. It does not scale cleanly. Motor cortical activity during reaching

transfer scales with channels. It does not scale cleanly. Motor cortical activity during reaching is

is well described by a low-dimensional dynamical structure — a neural manifold — in which a few

well described by a low-dimensional dynamical structure (a neural manifold) in which a few dozen

dozen latent dimensions capture most of the variance.[^gallego2017] Adding electrodes samples that same structure

latent dimensions capture most of the variance.[^gallego2017] Adding electrodes samples that same

more densely rather than revealing new independent signals, so returns diminish. This is why a

structure more densely rather than revealing new independent signals, so returns diminish. This is

96-channel array from the 1990s can support performance comparable to systems with ten times the

why a 96-channel array from the 1990s can support performance comparable to systems with ten times

channels, and why decoder architecture has mattered more than sensor count.

the channels, and why decoder architecture has mattered more than sensor count.

Where more channels do help is coverage: sampling several cortical areas, or sampling a functional

map finely enough to separate articulators in speech cortex. The genuinely hard constraints are

map finely enough to separate articulators in speech cortex. The hard constraints lie elsewhere: the

elsewhere — the number of independent control dimensions a person can learn to modulate, the

number of independent control dimensions a person can learn to modulate, the non-stationarity of the

non-stationarity of the recorded population from day to day, and the near-total absence of a

recorded population from day to day, and the near-total absence of a high-resolution write channel.

high-resolution write channel. Reading intention out is a solved-enough problem for many

Reading intention out is a solved-enough problem for many applications; writing rich sensory

applications; writing rich sensory information in is not, which is why touch feedback in

information in is not, which is why touch feedback in [[neuroprosthetics]] remains crude compared

[[neuroprosthetics]] remains crude compared with motor decoding. One laboratory response is to make

with motor decoding. One laboratory response is to make the stimulating hardware itself spike like a

the stimulating hardware itself spike like a neuron, at biological voltages and timescales; such

neuron, at biological voltages and timescales; such

[[artificial-neurons|bio-interfacing artificial neurons]] have so far driven living neurons only in

acute experiments, such as slices of mouse cerebellum.[^hadke2026]

## Limitations

Implanted arrays lose channels. The brain mounts a foreign-body response — microglial activation,

Implanted arrays lose channels. The brain mounts a foreign-body response (microglial activation,

astrocytic encapsulation — that displaces neurons from the recording tip, while insulation and

astrocytic encapsulation) that displaces neurons from the recording tip, while insulation and metal

metal traces degrade in a warm saline environment. Signal yield typically declines over months to

traces degrade in a warm saline environment. Signal yield typically declines over months to years,

years, though some implants have supported usable control beyond a thousand days and a few for

though some implants have supported usable control beyond a thousand days and a few for several

several years; the failure statistics are set out under [[utah-array]].

years; the failure statistics are set out under [[utah-array]].

Percutaneous connectors, still standard in academic systems, carry infection risk and tether the

user to a lab. Fully implanted wireless systems solve this but constrain power and data rate; the

most aggressive proposal for removing the tether, untethered ultrasonic motes, is covered under

[[neural-dust]] and has no human results.

[[neural-dust]] and has no human results. Decoders drift, requiring recalibration that ranges from

Decoders drift, requiring recalibration that ranges from minutes daily to occasional, depending on

minutes daily to occasional, depending on the system. Almost all reported performance comes from

the system. Almost all reported performance comes from controlled settings with a technician

controlled settings with a technician present.

present.

There is also a structural risk specific to commercial neurotechnology: a company can fail while

There is also a structural risk specific to commercial neurotechnology: a company can fail while its

its devices remain in people's heads. The discontinuation of the Argus II retinal system left users

devices remain in people's heads. The discontinuation of the Argus II retinal system left users with

with unsupported implants, a precedent that shadows every BCI trial. See [[retinal-implant]].

unsupported implants, a precedent that shadows every BCI trial. See [[retinal-implant]].

## Ethics and governance

Implanted BCIs raise questions that older neurotechnology such as [[deep-brain-stimulation]] and

Implanted BCIs raise questions that older neurotechnology such as [[deep-brain-stimulation]] and the

the [[cochlear-implant]] introduced in milder form: who owns the neural data, what happens when a

[[cochlear-implant]] introduced in milder form: who owns the neural data, what happens when a

decoder misreads intent, and whether a device that shapes behaviour compromises agency. The

decoder misreads intent, and whether a device that shapes behaviour compromises agency. The specific

specific novelty is that recorded activity can support inferences the user did not intend to

novelty is that recorded activity can support inferences the user did not intend to disclose, an

disclose, an issue taken up under [[mental-privacy]]. Legislatures have begun responding —

issue taken up under [[mental-privacy]]. Legislatures have begun responding under the banner of

constitutional and state-level neural-data provisions exist in several jurisdictions — under the

[[neurorights]], and constitutional and state-level neural-data provisions exist in several

banner of [[neurorights]], though enforcement mechanisms remain thin.

jurisdictions, though enforcement mechanisms remain thin.

Consumer neurotechnology is a separate regulatory category and a much weaker technical one.

Consumer neurotechnology is a separate regulatory category and a much weaker technical one. Wearable

Wearable EEG headsets and [[non-invasive-neuromodulation]] devices are sold for attention training,

EEG headsets and [[non-invasive-neuromodulation]] devices are sold for attention training, sleep,

sleep, and gaming with evidence that ranges from thin to absent, and the "brain-reading" claims

and gaming with evidence that ranges from thin to absent, and the "brain-reading" claims attached to

attached to them substantially outrun what scalp electrodes can resolve. Regulators have generally

them substantially outrun what scalp electrodes can resolve. Regulators have generally treated these

treated these as wellness products rather than medical devices, the same line that governs

as wellness products rather than medical devices, the same line that governs

[[wearable-health-sensors]], which is why the emerging neural-data statutes target data handling

rather than efficacy claims.

> [!debate] How invasive is worth it

> Synchron's position is that a device delivered by catheter, with sixteen electrodes, will reach

> far more patients than one requiring a craniotomy, and that discrete reliable control is enough

> for most assistive tasks. The counter-position, taken by the penetrating-array developers, is

> for most assistive tasks. The counter-position, taken by the penetrating-array developers, is that

> that low-bandwidth control caps the addressable applications at switch-like interaction and rules

> low-bandwidth control caps the addressable applications at switch-like interaction and rules out

> out speech and dexterous limb control. Both claims are testable and neither has been settled.

> speech and dexterous limb control. Both claims are testable and neither has been settled.

## Outlook

The near-term agenda is legible: more participants, wireless and fully implanted systems, decoders

The near-term agenda is clear: more participants, wireless and fully implanted systems, decoders

intended to survive a week without recalibration, and eventually a pivotal trial in a well-defined

indication such as loss of speech in ALS. If that sequence holds, it could produce an approved

assistive device for a small clinical population, though no timeline for it is well supported and

none of the current studies is large enough to predict one.

The longer-term claims — that BCIs will serve as a general-purpose channel for

Longer-term claims rest on assumptions the clinical work does not test: that BCIs will serve as a

[[human-ai-merger]], or provide the read-out fidelity that [[whole-brain-emulation]] would require

general-purpose channel for [[human-ai-merger]], or provide the read-out fidelity that

— rest on assumptions the clinical work does not test. A speech decoder recovers a motor plan, not

[[whole-brain-emulation]] would require. A speech decoder recovers a motor plan, not a thought; a

a thought; a 1,024-channel array samples a millionth of a cortical hemisphere. Whether elective

1,024-channel array samples a millionth of a cortical hemisphere. Whether elective implantation in

implantation in healthy people ever clears the risk-benefit bar is a question about surgical

healthy people ever clears the risk-benefit bar is a question about surgical complication rates and

complication rates and device lifetimes, and neither is currently known well enough to answer.

device lifetimes, and neither is currently known well enough to answer.

## See also