Skip to content
ContentLora

    Tip: press / anywhere to search.

    Explainer

    How brain-computer interfaces work

    A brain-computer interface measures brain signals, decodes the intent behind them and uses that to control a device such as a cursor, a robotic arm or a voice synthesizer.[1][2] Machine-learning decoders have driven recent gains: research systems now turn attempted speech into text with about 97 percent accuracy in a single participant, though none has full FDA marketing approval as a long-term implant.[3][4][5]

    Editor reviewedStrict sourcingUpdated Neuroscience and brain-computer interfacesLife sciencesHealth and medicine

    Three steps: measure, decode, control

    Every brain-computer interface (BCI) does three things. It measures brain signals, it decodes what the person is trying to do, and it uses that to control a device.[1] The device might be a computer cursor, a robotic arm or a speaker that talks for someone who can no longer speak.[2]

    Most medical BCIs work by listening to the parts of the brain that plan movement. A person with paralysis tries to move their hand, or tries to say a word. The brain still produces the signal even though the body cannot act on it, and the BCI picks it up.[6][7]

    Medical BCIs mostly decode motor intent: attempted hand movements for cursor control, or attempted articulation for speech. The 2006 BrainGate result showed that intended hand motion still modulated motor-cortex spiking three years after spinal cord injury.[6] Speech decoders read from ventral precentral gyrus, where a 2023 study found a detailed articulatory code that persists years after paralysis.[7][8] FDA’s 2021 guidance frames implanted BCIs as neuroprostheses that restore lost motor or sensory function.[9]

    Decoding: where AI comes in

    Brain signals differ from person to person, so each BCI has to learn its user. That calibration used to take a long time. AI shortens it, and language models help by predicting which words the person is likely trying to say.[4]

    In a 2024 study, a man with ALS was using his speech BCI within 30 minutes of first calibration, at 99.6 percent accuracy on a small 50-word vocabulary.[10][3]

    Modern decoders map neural features to phonemes or kinematics with recurrent or transducer networks, then use a language model to resolve word sequences.[4][11] Recalibration is a practical burden because recorded signals drift. Neuralink reported in October 2026 that encoders pretrained on 50,000 hours of unlabelled brain data kept decoders working for weeks without recalibration.[12] Synchron has also announced a brain “foundation model”, called Chiral.[13]

    Outputs: cursor, text, voice, movement

    • Cursor and computer control. Neuralink measures cursor performance in bits per second and says able-bodied people average 8 to 10 bits per second on its task; it reported an 11.32 bits-per-second record in 2026.[14][12] Apple added a protocol in 2025 so its operating systems can accept BCI input through Switch Control.[15]
    • Text from attempted speech. A 2023 Stanford system reached 62 words per minute; a 2024 UC Davis system held 97.5 percent accuracy over 8.4 months.[7][3] See speech-neuroprostheses.
    • Synthesized voice. A 2025 system produced a man’s voice in real time and let him change intonation and sing short melodies.[16]
    • Movement. A 2023 brain-spine interface linked cortical signals to spinal cord stimulation so a man with tetraplegia could stand and walk.[17]

    Measuring progress

    Researchers use speed and accuracy. For talking, they count words per minute and the share of words decoded wrongly. Normal conversation runs at about 160 words per minute, so the best BCIs are still slower than natural speech.[18][7]

    Headline numbers come from very small samples, often one participant, and from different tasks and vocabularies, so they are not directly comparable.[8][7] Word error rate depends heavily on vocabulary size: the Stanford system’s error rose from 9.1 percent on 50 words to 23.8 percent on 125,000 words.[7] Company figures such as Neuralink’s bits-per-second records or Paradromics’ preclinical rate of more than 200 bits per second are not peer-reviewed in the same way.[12][19]

    Limits and open questions

    Implant trials are open only to people with severe disabilities because of surgical risk.[20] Some trial participants have had devices removed when post-trial funding or support ran out.[21] And as decoders reach into inner speech, mental privacy has become a design question as well as an ethical one.[22][23]

    Questions readers ask

    Does a BCI read thoughts?

    Current medical BCIs decode intended movements or attempted speech from motor areas. A 2025 study showed inner speech can also be decoded from motor cortex, and demonstrated ways to stop a device from decoding private inner speech unintentionally.[24][22]

    How long does it take to set up a BCI?

    Decoders must be calibrated to each user. In one 2024 speech study, 30 minutes of recordings gave 99.6 percent accuracy on a 50-word vocabulary on the first day.[4][3][10]

    How is BCI performance measured?

    Common measures are words per minute and word error rate for communication, and bits per second for cursor control. Neuralink reported an 11.32 bits-per-second record in October 2026.[7][14][12]

    Sources

    Each numbered claim is a statement we checked against the sources listed with it. Status shows how well established it is.

    1. [1]

      GAO describes a BCI as working in three steps: it measures the user's brain signals, decodes the intent of those signals, and uses that to control a device. confirmedas of 2024-12-17

    2. [2]

      Brain-computer interfaces are electronic systems, either implanted in the brain or worn on the head, that let people control computers, robots or other devices using brain signals. confirmedas of 2024-12-17

    3. [3]

      The UC Davis neuroprosthesis reached 99.6 percent accuracy on a 50-word vocabulary on its first day of use and sustained 97.5 percent accuracy over 8.4 months, with the participant conversing at about 32 words per minute for more than 248 hours. confirmedas of 2024-08-15

    4. [4]

      GAO reports that BCIs must be individually calibrated to each user's brain signals, and that AI, including language models that predict what a user intends to say, can cut calibration time and make devices faster and more accurate. confirmedas of 2024-12-17

    5. [5]

      As of December 2024, FDA had not granted marketing authorization for any implantable device for BCI use. confirmedas of 2024-12-17

    6. [6]

      In a 2006 Nature study, a man with tetraplegia used a 96-microelectrode array implanted in his primary motor cortex to move a computer cursor, open simulated e-mail and operate a prosthetic hand. confirmedas of 2006-07-13

    7. [7]

      A 2023 Stanford study decoded the attempted speech of a participant with ALS from intracortical arrays at 62 words per minute, with word error rates of 9.1 percent on a 50-word vocabulary and 23.8 percent on a 125,000-word vocabulary, and found a detailed articulatory representation of phonemes persisting years after paralysis. confirmedas of 2023-08-23

    8. [8]

      In the UC Davis speech neuroprosthesis study, a 45-year-old man with ALS received four microelectrode arrays with 256 electrodes in his left ventral precentral gyrus. confirmedas of 2024-08-15

    9. [9]

      FDA defines implanted BCI devices as neuroprostheses that interface with the central or peripheral nervous system to restore lost motor or sensory capabilities in patients with paralysis or amputation. confirmedas of 2021-05-20

    10. [10]

      Calibrating the UC Davis neuroprosthesis took 30 minutes of recordings on the first day, and after 1.4 more hours of training it reached 90.2 percent accuracy on a 125,000-word vocabulary. confirmedas of 2024-08-15

    11. [11]

      A 2025 UCSF study used high-density surface recordings to drive a streaming speech synthesizer in the participant's pre-injury voice, decoding in 80-millisecond increments. confirmedas of 2025-03-31

    12. [13]

      In March 2025 Synchron unveiled Chiral, which it described as a cognitive AI brain foundation model, and in January 2025 announced a partnership with NVIDIA's Holoscan platform. confirmedas of 2025-03-19

      • News · Synchron · March 19, 2025 press release listing (retrieved 2026-10-10)
    13. [15]

      In May 2025 Apple announced that iOS, iPadOS and visionOS would add a protocol supporting Switch Control for brain-computer interfaces. confirmedas of 2025-05-13

    14. [16]

      A 2025 UC Davis study synthesized a man's voice in real time from 256 implanted microelectrodes, and let him change intonation and sing short melodies. confirmedas of 2025-06-12

    15. [17]

      A 2023 study reported a fully implanted brain-spine interface that linked cortical signals to spinal cord stimulation and let a man with chronic tetraplegia stand and walk in community settings. confirmedas of 2023-05-24

    16. [18]

      Natural conversation runs at about 160 words per minute, according to the 2023 Stanford speech BCI paper. confirmedas of 2023-08-23

    17. [19]

      Paradromics says Connexus exceeded 200 bits per second of information transfer in preclinical testing. confirmedas of 2026-10-10

    18. [20]

      The risks of brain surgery currently limit participation in implantable BCI clinical trials to people with severe disabilities. confirmedas of 2024-12-17

    19. [21]

      Experts told GAO that some clinical trial participants have had a BCI removed because there were no funds or medical support after the trial. confirmedas of 2024-12-17

    20. [22]

      The same study found some aspects of free-form inner speech could be decoded during recall and counting tasks, and demonstrated strategies that prevent speech BCIs from unintentionally decoding private inner speech. confirmedas of 2025-08-21

    21. [23]

      The UNESCO Recommendation states that privacy, including mental privacy, is fundamental for personal identity and agency, and calls for prior, free and informed consent to collect neural data except in life-threatening emergencies. confirmedas of 2025-11-11

    22. [24]

      A 2025 study in four participants found that inner speech is robustly represented in motor cortex and that imagined sentences can be decoded in real time. confirmedas of 2025-08-21

    Revision history (1)
    1. Page created.

    Created Oct 10, 2026. Last reviewed by an editor on Oct 10, 2026. Next scheduled review: Jan 10, 2027.

    Cite this page

    "How brain-computer interfaces work." ContentLora, updated Oct 10, 2026. https://contentlora.com/explain/how-brain-computer-interfaces-work

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    Keep exploring

    ☰ Contents