# Part II. What the body already broadcasts | The Intent Layer Source: https://setu-labs.co/manifesto-2.html > Intent leaves physical traces all the way from cortex to air. The question is which trace you can catch, and where you have to stand to catch it. [Download PDF ](https://setu-labs.co/assets/setu-the-intent-layer.pdf) ## What the body already broadcasts Intent leaves physical traces all the way from cortex to air. The question is which trace you can catch, and where you have to stand to catch it. [Image: A historical ink drawing of nerve cells arranged in the layers of the cerebral cortex.] *The cerebral cortex, drawn cell by cell from a stained section. Every method in this part listens, from a different distance, to tissue like this. Image: [Wellcome Collection, CC BY 4.0](https://commons.wikimedia.org/wiki/File:Cerebral_cortex_in_a_child._Wellcome_L0002048.jpg).* When you decide to say a word, neurons in the cortex change their firing. Currents flow. Seconds later, blood supply shifts to feed the active tissue. Commands descend to the muscles of the tongue, jaw and larynx. Tissue moves. Air vibrates. Each stage is a place where an instrument can listen. A BCI measures brain activity and translates it into commands for a device, giving the brain an output that does not depend on muscle.[[4](https://setu-labs.co/manifesto-sources.html#s4)] The term was coined in 1973.[[5](https://setu-labs.co/manifesto-sources.html#s5)] This is an old science with new tools. Three rules make any result in this field easier to read. Reading is not writing: recording the brain and putting information into it are different problems, and this essay is about reading. Movement is easier than meaning: the strongest results decode attempted handwriting and speech. And the verb matters: seeing, hearing, attempting and imagining are different brain states, so always ask what the participant was doing. ### Nine ways to listen A scalp electrode stands to a neuron roughly as a weather satellite stands to a raindrop. It cannot see the single event. It sees the organised behaviour of millions, blurred by the skull. An implanted electrode is a rain gauge: exact, local, and somebody has to go through bone to install it. Satellites changed the world and rain gauges did not, for one reason. Satellites cover everyone. EEG, the recording of electrical activity at the scalp, is the satellite of this field. It is a century old, safe, light and cheap enough to build anywhere. That is why it is the instrument this essay keeps returning to. #### Figure 3 (interactive): Where each instrument listens Seven methods record the brain from different depths; two record the periphery and never touch neural tissue. Select a number to see what each does and does not measure. - Brain signal, no surgery - Brain signal, implanted - Peripheral signal, not a brain recording ##### 1 · Scalp EEG Brain · no surgery - **Measures** Voltage at the scalp from large populations of cortical neurons firing in step. Millisecond timing. - **Does not** Resolve small or deep sources. Skull and tissue smear the signal; muscle and eye movement easily swamp it. - **Practical** Cheap, portable, decades of clinical use. Dry electrodes trade convenience for signal quality. Hair, sweat and motion matter. ##### 2 · fNIRS Brain · no surgery - **Measures** Changes in blood oxygenation in the outer cortex, using near-infrared light shone through the scalp. - **Does not** Track neural events directly or quickly: the blood response lags by seconds. Only shallow cortex is reachable. - **Practical** Wearable, and less confining than a scanner, though motion artifacts remain a real problem. Low information rate; hair and skin pigmentation affect light coupling and must be designed for. ##### 3 · MEG Brain · no surgery - **Measures** The tiny magnetic fields produced by neural currents. Millisecond timing, and less distorted by the skull than EEG. - **Does not** Work in an ordinary room. The fields are far weaker than everyday magnetic noise, so recording needs heavy shielding. - **Practical** Classic systems are cryogenic and fixed. Newer optically pumped sensors can be worn on the head, but still inside a shielded space.[[30](https://setu-labs.co/manifesto-sources.html#s30)] ##### 4 · fMRI Brain · no surgery - **Measures** Blood-oxygen changes across the whole brain at millimetre scale, including deep structures. - **Does not** Capture fast dynamics. One image of a slow blood response every second or two. - **Practical** A multi-tonne magnet in a dedicated room; the participant lies still. A research and clinical instrument, not a wearable. ##### 5 · ECoG Brain · implanted - **Measures** Electrical potentials from electrode grids laid on the brain’s surface, under the skull. Clean, stable, broad coverage. - **Does not** Record individual neurons. - **Practical** Requires open neurosurgery. Used in the leading speech-neuroprosthesis trials.[[7](https://setu-labs.co/manifesto-sources.html#s7)] ##### 6 · Intracortical arrays Brain · implanted - **Measures** Spiking of individual neurons and small groups, from fine electrodes that enter the cortex. The richest signal available. - **Does not** Cover much territory: a few square millimetres per array. Signals can change as tissue responds over years. - **Practical** Open neurosurgery; long-term durability and upkeep are active clinical questions. ##### 7 · Endovascular electrodes Brain · implanted, no open surgery - **Measures** Cortical potentials from a stent-like electrode delivered through a vein to sit beside motor cortex. - **Does not** Match the channel count or resolution of open-surgery implants. So far used for a small set of commands. - **Practical** Catheter procedure rather than craniotomy; a first-in-human safety study followed four patients for twelve months.[[24](https://setu-labs.co/manifesto-sources.html#s24)] ##### 8 · Surface EMG Peripheral · not brain - **Measures** Electrical activity of muscles under the skin: jaw, lips, throat or forearm, as a movement is made or nearly made. - **Does not** Measure the brain or anything you have not tried to articulate. It needs working motor nerves and muscle. - **Practical** Small, cheap, wearable. Sensitive to electrode placement and skin contact. The basis of most silent-speech research.[[18](https://setu-labs.co/manifesto-sources.html#s18)] ##### 9 · Radar and motion sensing Peripheral · not brain - **Measures** Sub-millimetre movement of skin and tissue around the mouth and throat, by radio reflection, ultrasound or strain sensors. - **Does not** Record any neural or muscular electrical signal. It senses that something moved, not why. - **Practical** Can work without skin contact. Early-stage for speech; a sensor that can see a face can also see a bystander’s. ***Schematic, not anatomical scale.** Sensor positions are illustrative. The dotted outline around the head stands for an fMRI scanner bore; the dashed arc for an MEG sensor helmet. Descriptions summarise standard measurement physics; study citations are in Figure 4.* [Image: A person wearing a fabric cap studded with EEG electrodes.] EEG · scalp electrodes [Image: A person wearing a head cap fitted with optical fNIRS sensors and cables.] fNIRS · light through the scalp [Image: A magnetoencephalography scanner: a seat beneath a large white helmet-shaped sensor housing.] MEG · inside a shielded room [Image: A hospital MRI scanner with its patient table extended.] MRI · the scanner fMRI uses [Image: Medical illustration of an electrode grid placed on the exposed surface of the brain.] ECoG · grid on the cortex (illustration) [Image: Patent drawing of a square array of one hundred needle-like electrodes.] Intracortical array · patent drawing *The instruments themselves. The first four involve no surgery; the last two are implanted. Images via Wikimedia Commons: [Chris Hope, CC BY 2.0](https://commons.wikimedia.org/wiki/File:EEG_Recording_Cap.jpg); [Alexarje, CC BY 4.0](https://commons.wikimedia.org/wiki/File:Alexander_Refsum_Jensenius_with_fNIRS.jpg); [National Institute of Mental Health, public domain](https://commons.wikimedia.org/wiki/File:NIMH_MEG.jpg); [Jan Ainali, CC BY 3.0](https://commons.wikimedia.org/wiki/File:MRI-Philips.JPG); [Blausen Medical (BruceBlaus), CC BY 3.0](https://commons.wikimedia.org/wiki/File:Intracranial_electrode_grid_for_electrocorticography.png); [Richard A. Normann, patent drawing, public domain](https://commons.wikimedia.org/wiki/File:Utah_array_pat5215088.jpg).* ### The evidence, verb by verb **Attempting to write.** In 2021 a man paralysed from the neck down imagined writing letters by hand while implanted arrays recorded his motor cortex. The system produced 90 characters a minute at 94.1% accuracy.[[6](https://setu-labs.co/manifesto-sources.html#s6)] **Attempting to speak.** In 2023, two implant studies decoded attempted speech at 78 and 62 words a minute, with about one word in four wrong.[[7](https://setu-labs.co/manifesto-sources.html#s7)][[8](https://setu-labs.co/manifesto-sources.html#s8)] A year later a third held 97.5% accuracy across eight months of daily conversation.[[9](https://setu-labs.co/manifesto-sources.html#s9)] **Attempting to move.** People with paralysis have steered robotic arms,[[20](https://setu-labs.co/manifesto-sources.html#s20)] felt touch written back into the brain,[[21](https://setu-labs.co/manifesto-sources.html#s21)] and walked again through a bridge between cortex and spinal cord.[[23](https://setu-labs.co/manifesto-sources.html#s23)] EEG alone has driven a robotic arm in continuous tracking.[[22](https://setu-labs.co/manifesto-sources.html#s22)] **Perceiving.** From inside a scanner, models have recovered the gist of a story a person was hearing,[[11](https://setu-labs.co/manifesto-sources.html#s11)] rebuilt pictures they were looking at,[[12](https://setu-labs.co/manifesto-sources.html#s12)] and adapted to a new person with one hour of data.[[13](https://setu-labs.co/manifesto-sources.html#s13)] With EEG and MEG, models can pick out which stretch of speech a listener was hearing.[[17](https://setu-labs.co/manifesto-sources.html#s17)] These pair a recording with a generative model. The signal supplies the meaning; the model paints the detail. **Typing.** With no surgery, sentences have been decoded from magnetic recordings of people typing, at 39% word error.[[16](https://setu-labs.co/manifesto-sources.html#s16)] The same task on EEG is harder today: two characters in three were wrong.[[15](https://setu-labs.co/manifesto-sources.html#s15)] We read that as the starting line. Every curve in Part IV says EEG has the most to gain from data. **Mouthing.** Away from the brain, silently mouthed words have been turned into speech from facial muscle signals,[[18](https://setu-labs.co/manifesto-sources.html#s18)] and a wrist sensor trained on thousands of people now works on strangers with no calibration.[[19](https://setu-labs.co/manifesto-sources.html#s19)] **Thinking in words.** In 2025, imagined sentences were decoded from four implanted participants. The same study showed how to stop a device decoding inner speech the user had not offered.[[25](https://setu-labs.co/manifesto-sources.html#s25)] #### Figure 4 (evidence): The evidence wall Sixteen studies, each stated the same way: what was measured, what the person was doing, what came out, and what the result does not show. 2012 · Nature Peer-reviewed ##### A robotic arm, steered from motor cortex Signal: intracortical array · Verb: **attempting to reach** 3-D reach and grasp Two people with long-standing tetraplegia; one drank from a bottle unaided. **Did not show:** speed or accuracy near able-bodied movement; use outside supervised sessions. Details and source - **Tested** 2 participants, 96-channel array - **Source** Hochberg et al.[[20](https://setu-labs.co/manifesto-sources.html#s20)] 2021 · Nature Peer-reviewed ##### Handwriting without a hand Signal: intracortical arrays · Verb: **attempting to write** 90 characters / min 94.1% raw accuracy online; >99% offline with autocorrect. **Did not show:** 90 *words* per minute; more than one participant; multi-year stability. Details and source - **Tested** 1 participant with spinal-cord injury - **Source** Willett et al.[[6](https://setu-labs.co/manifesto-sources.html#s6)] 2021 · Science Peer-reviewed ##### Touch, written back to the brain Signal: intracortical recording + stimulation · Verb: **attempting to grasp** 20.9 s → 10.2 s Median task time with a robotic arm, without and with evoked tactile sensation. **Did not show:** rich or naturalistic sensation; writing of language or concepts. Details and source - **Tested** 1 participant with tetraplegia - **Source** Flesher et al.[[21](https://setu-labs.co/manifesto-sources.html#s21)] 2023 · Nature Peer-reviewed ##### Speech, voice and a face from the cortical surface Signal: ECoG grid · Verb: **attempting to speak** Median 78 words / min Median 25% word error rate; text, synthetic voice and avatar. **Did not show:** error rates low enough for unassisted daily use; decoding of unattempted thought. Details and source - **Tested** 1 clinical-trial participant with severe paralysis - **Source** Metzger et al.[[7](https://setu-labs.co/manifesto-sources.html#s7)] 2023 · Nature Peer-reviewed ##### Large-vocabulary speech from single neurons Signal: intracortical arrays · Verb: **attempting to speak** 62 words / min 23.8% word error on a 125,000-word vocabulary; 9.1% on 50 words. **Did not show:** conversational accuracy (see 2024); results beyond one participant. Details and source - **Tested** 1 participant with ALS - **Source** Willett et al.[[8](https://setu-labs.co/manifesto-sources.html#s8)] 2024 · NEJM Peer-reviewed ##### A speech prosthesis used for months Signal: intracortical arrays · Verb: **attempting to speak** 97.5% word accuracy Sustained over 8.4 months; ~32 words / min; 248+ hours of self-paced use; 30-minute first calibration. **Did not show:** generalisation across patients; performance without ongoing recalibration. Details and source - **Tested** 1 participant with ALS, 256 electrodes - **Source** Card et al.[[9](https://setu-labs.co/manifesto-sources.html#s9)] 2025 · Cell Peer-reviewed ##### Inner speech, and a lock for it Signal: intracortical arrays · Verb: **imagining speech** Imagined sentences, real time Inner speech is represented in motor cortex; some unprompted inner speech was decodable; safeguards blocked it. **Did not show:** decoding of arbitrary private thought; anything outside implanted motor cortex. Details and source - **Tested** 4 implanted participants - **Source** Kunz et al.[[25](https://setu-labs.co/manifesto-sources.html#s25)] 2023 · Nat. Neurosci. Peer-reviewed ##### The gist of a story, from blood flow Signal: fMRI · Verb: **listening** (also imagining, watching) Meaning, not words Generated word sequences that recover the sense of heard speech. **Did not show:** word-for-word transcripts; decoding without the person’s cooperation, which was required to train and to apply the decoder. Details and source - **Tested** 3 participants, ~16 hours of scanning each - **Source** Tang et al.[[11](https://setu-labs.co/manifesto-sources.html#s11)] 2023 · NeurIPS Peer-reviewed ##### Rebuilding a viewed image (MindEye) Signal: 7-tesla fMRI · Verb: **looking at pictures** Retrieval and reconstruction Identifies the viewed image among similar candidates; generates a resembling image with a diffusion model. **Did not show:** imagined or remembered images; exact readout. The generator supplies detail the recording lacks. Details and source - **Tested** 4 participants from a public dataset,[[29](https://setu-labs.co/manifesto-sources.html#s29)] tens of hours each - **Source** Scotti et al.[[12](https://setu-labs.co/manifesto-sources.html#s12)] 2024 · ICML Peer-reviewed ##### A new brain with one hour of data (MindEye2) Signal: 7-tesla fMRI · Verb: **looking at pictures** ~1 hour to adapt After pretraining across 7 other people in the same dataset. **Did not show:** zero-calibration transfer; transfer across scanners, tasks or to portable sensors. Details and source - **Tested** Held-out participants within one dataset - **Source** Scotti et al.[[13](https://setu-labs.co/manifesto-sources.html#s13)] 2024 · NeurIPS Peer-reviewed ##### Short video from a scanner (NeuroClips) Signal: fMRI · Verb: **watching video** Clips up to 6 s at 8 fps Resembling the watched footage, via a pretrained text-to-video model. **Did not show:** a live feed, imagined scenes, or frame-accurate recovery. Details and source - **Tested** A public fMRI–video dataset - **Source** Gong et al.[[14](https://setu-labs.co/manifesto-sources.html#s14)] 2025 · arXiv Preprint ##### Typed sentences from outside the skull (Brain2Qwerty) Signal: MEG and EEG · Verb: **typing memorised sentences** 32% vs 67% character error Average for MEG vs EEG. Best MEG participants: 19%. **Did not show:** free-form thought to text. Participants were physically typing; EEG got two characters in three wrong. Details and source - **Tested** 35 healthy volunteers - **Source** Lévy et al.[[15](https://setu-labs.co/manifesto-sources.html#s15)] 2026 · arXiv Preprint ##### More data, fewer errors (Brain2Qwerty v2) Signal: MEG, real time · Verb: **typing natural sentences** 39% word error rate Average; improves log-linearly with data. Best participant: half of sentences with at most one wrong word. **Did not show:** anything wearable (MEG needs shielding); peer review is pending; ~10 hours of recording per person. Details and source - **Tested** 9 participants, 22,000 sentences - **Source** Zhang et al.[[16](https://setu-labs.co/manifesto-sources.html#s16)] 2025 · arXiv Preprint ##### The same task on four machines Signal: EEG, MEG, 3T and 7T fMRI · Verb: **looking at pictures** 498 hours, 84 people The most precise devices decode best. No plateau with more data. Hours per person matter more than more people. **Did not show:** where any device’s ceiling lies; anything about language or intent. Details and source - **Tested** 8 public datasets, 2.3 million brain responses - **Source** Banville et al.[[34](https://setu-labs.co/manifesto-sources.html#s34)] 2020 · EMNLP Peer-reviewed ##### A voice for silently mouthed words Signal: facial surface EMG · Verb: **mouthing words silently** 64% → 4% word error Closed-vocabulary condition. Open vocabulary: 88% → 68%. **Did not show:** brain decoding of any kind; usable open-vocabulary accuracy; multi-speaker generalisation. Details and source - **Tested** Single-speaker dataset, wet electrodes - **Source** Gaddy & Klein[[18](https://setu-labs.co/manifesto-sources.html#s18)] 2025 · Nature Peer-reviewed ##### A wristband that works on strangers Signal: wrist surface EMG · Verb: **gesturing and handwriting** 20.9 words / min Handwriting, with no per-user calibration; +16% with personalisation. **Did not show:** speech or brain decoding. Requires functioning hand muscles. Details and source - **Tested** Models trained on thousands of consenting participants - **Source** CTRL-labs at Reality Labs[[19](https://setu-labs.co/manifesto-sources.html#s19)] ***Reading rules.** Numbers are quoted from each paper’s abstract in its own metric; they are not comparable across cards. Preprints have not completed peer review. No stimulus or reconstruction images are reproduced here because image rights rest with the original authors and datasets; follow the source links to see them, including failure cases.* > The laboratory has proved that the information exists. It has not yet proved that it travels.