Signal Atlas

Research library

Brain–computer interfaces

Reading limited signals from the brain to restore movement or speech, and the step from therapy to enhancement. What 6G adds: indirect. No official 6G document targets brain interfaces; low delay and edge AI matter only if decoding moves off the implant.

Of 39 findings in this theme, 27 are real today, 6 designed, 5 forecast and 1 not supported.

Real today 27

  • A BCI is a chain of translations. A brain-computer interface measures some brain activity, cleans it up, picks out a pattern, turns it into a command and shows the result back to the user. Both the person and the software learn over time.
  • BCIs decode narrow tasks, not whole thoughts. Most interfaces tell apart a small set of trained options, such as left versus right movement or a chosen letter, after calibration to one person. Implanted speech interfaces now go further, turning attempted speech into text from a very large vocabulary, but still only for trained users doing a cooperative task.
  • Bionic limbs read muscles and send back sensation. Prostheses can pick up electrical signals from muscles to drive movement and return electrical stimulation that the brain reads as touch. The user and device then adapt to each other over time.
  • Bionic limbs share control with their user. Advanced prosthetic limbs combine muscle-signal sensing, onboard sensors, local decision-making and electrical feedback the user feels as touch. The limb adjusts grip faster than conscious control.
  • Body signals partly predicted when drivers used autopilot. In a simulated convoy, brain activity, heart rate, breathing and context predicted when drivers switched an autonomous assistant off (about 80% accuracy) or on (about 65%). The same research found an averaged brain pattern matched none of the 44 people tested.
  • Brain implants restore speech near conversation speed. In 2023 two teams (Stanford and UC San Francisco) used brain implants to let paralysed people produce text at 62 and 78 words per minute, and one system also spoke through a digital voice and face.
  • Brain scans can be turned into rough sentences. In 2023 a University of Texas team used an AI model to turn fMRI brain scans into running text that captured the gist of stories a person heard or imagined. No implant was needed.
  • Brain stimulation treats symptoms, not yet augmentation. Optogenetics is in research and not expected to augment humans within ten years; deep brain stimulation helps some Parkinson's patients; magnetic stimulation is in trials for depression and similar conditions. Research earphones that read brain activity to amplify the conversation a listener attends to show early human-device symbiosis.
  • Brain-computer interface research dates to the 1970s. A 1973 Annual Review paper by J. J. Vidal, 'Toward direct brain-computer communication', shows researchers were already working on brain-to-computer links five decades ago. DARPA's N3 program came much later.
  • Brain-computer interfaces remain small and mostly medical. In 2019 the invasive BCI market was about $125 million, mostly treating brain and nerve disorders. Non-invasive BCIs were too crude for gaming or equipment control, and a seamless mind-machine link was described as bordering on science fiction.
  • Brain-to-brain links exist, but at a trickle. Lab experiments have linked two or three people by reading one brain with EEG and nudging another with magnetic stimulation, letting them jointly solve simple tasks.
  • Brain-to-brain messages were shown in a 2014 lab study. A 2014 PLoS ONE paper sent coded words between two people's brains over the internet, combining an EEG interface with magnetic stimulation. The authors called it a proof of principle.
  • DARPA tested EEG-assisted threat spotting in 2012. DARPA's CT2WS paired a wide-angle camera and software with an EEG cap on a human operator, whose brain response helped flag threats. Reported field tests found about 91 percent of targets with far fewer false alarms than software alone.
  • Empathy shows up in measurable brain networks. Brain-scan studies consistently link empathy to the anterior insula and cingulate cortex, and researchers can predict how much empathic concern someone feels from their brain activity. Empathy is described as mirroring others' feelings plus working out their perspective.
  • FDA cleared a thin brain-surface electrode for 30 days. In March 2025 the FDA cleared Precision Neuroscience's thin-film electrode array for temporary use on the brain's surface for under 30 days, for recording, mapping and stimulation.
  • Head sensors can tell which speech you heard. A Meta AI team trained a model on magnetic and electrical brain recordings so that three seconds of brain activity could pick out the matching speech clip from over a thousand options, far above chance.
  • Implanted versus external BCIs trade quality for risk. Electrodes inside the brain give finer signals but bring surgery, infection, tissue reaction and drifting signal quality. Caps and headsets are safer and easier but capture less detail.
  • Lab work lets people steer small drones by thought. A 2013 study had volunteers fly a small drone indoors using scalp-EEG signals from imagined movement, and DARPA's N3 program listed drone control as an envisioned use. DARPA's SoSITE, by contrast, is about networking aircraft, not brain control.
  • Memory implant boosted recall by about a third. In a DARPA-funded pilot, an implant recorded a person's own memory-forming brain patterns and replayed them as stimulation, improving memory test scores by roughly 35-37 percent.
  • Optogenetics and CRISPR are real, but need delivery. Optogenetics has restored partial sight in one patient, and CRISPR has an approved therapy, but both involved medical delivery steps. DARPA's N3 non-surgical interface programme did not use either to change DNA remotely.
  • Read-and-write implants restore walking and lift depression. Bidirectional implants both read the nervous system and stimulate it back: a brain-spine 'digital bridge' let a paralysed man walk again (2023), and a personalised brain stimulator that fires only when it detects a depression pattern gave lasting relief (2021). Epilepsy devices have worked this way since 2013.
  • Some people can raise goosebumps at will. A 2018 study of 32 people found they could produce goosebumps voluntarily, often linked to emotion such as awe or music. It follows three earlier single-person case reports.
  • Speech-restoring brain implant entered US trial in 2026. Paradromics received FDA approval for a speech-restoration trial in November 2025 and implanted its first participant, a woman with motor neuron disease, on 17 June 2026. Signals pass wirelessly through the skin to an outside receiver.
  • Stimulating the brain is different from reading it. Sending energy into nerve tissue to change activity and inferring information from measured activity are two separate abilities. A device can stimulate without knowing your thoughts, or decode a task without being able to force it.
  • Toronto did the first Neuralink implants outside the US. Toronto Western Hospital implanted Neuralink brain-computer interfaces in two people with spinal cord injuries on 27 August and 3 September 2025, in the CAN-PRIME study, so they can control devices by thought.
  • Vein-delivered brain implant met one-year safety goal. Synchron reported in September 2024 that its US COMMAND study met its one-year primary safety goal: none of the six people had a device-related serious adverse event leading to death or permanent increased disability. Its implant is placed through a blood vessel rather than open brain surgery.
  • Wristband reads nerve signals to control devices. Meta's neural wristband picks up tiny electrical nerve signals at the wrist and turns them into clicks, gestures and air-handwriting (about 21 words per minute), working for new users without long calibration. It is aimed at controlling AR glasses.

Designed, not deployed 6

  • Augmentation runs from restoring to transforming. Human augmentation covers restoring lost function, assisting a task, optimising existing ability, enhancing beyond normal, and transforming identity or embodiment. A 2017 IEEE Future Directions post groups it into bionics, BCIs, neurotechnology, drugs and genetics.
  • Closed-loop neural devices are sensitive to delay. Some interfaces both read the brain and send something back, such as feedback or stimulation, and then adjust. If the response arrives too slowly, the user's sense of control breaks down and the loop can become unstable.
  • DARPA funded nerve-stimulation therapy research in 2015. DARPA's ElectRx programme funded teams to treat pain, inflammation and post-traumatic stress by stimulating peripheral nerves such as the vagus nerve, using closed-loop devices that sense and respond.
  • DARPA's 2024 biology office sought human-performance ideas. In September 2024 DARPA's Biological Technologies Office opened a broad call for ideas covering human performance, neurological health, cognitive performance, sensors and biodefence.
  • Early study links AI-assisted essays to weaker connectivity. A preprint on essay writing found lower EEG connectivity and weaker ownership of the text among people who used an AI assistant. It reports cognitive costs in a small sample, not a measured fall in intelligence.
  • Militaries are planning for human augmentation. DARPA's 2019 N3 program sought non-surgical headsets that read and write brain activity, the UK Ministry of Defence called human augmentation a new paradigm, and NATO's 2020-2040 trends report expects human-machine teams.

Forecast 5

  • Augmented humans first, 'Human 2.0' not before 2040s. The IEEE papers describe a ladder: devices that restore lost function, then augment a few people, then 'Human 2.0' where many have seamless brain-web links, then transhumanism. They place Human 2.0 and superhuman intelligence in the 2040s-2050s.
  • BCIs move from reading, to read-write, to networked. The research's roadmap puts brain interfaces in three stages: read-only devices, closed-loop read-and-write therapies, and speculative cloud-connected 'neural telepathy' after 2030. Narrow closed-loop therapies already exist: an epilepsy stimulator approved in 2013 and adaptive deep brain stimulation for Parkinson's approved in 2025.
  • Devices could turn invisible signals into felt sensations. Hearing aids, cochlear implants and night vision are established. A 2017 IEEE Future Directions post imagines devices that turn electrical, magnetic or radio fields into touch; a wearable belt that buzzes toward magnetic north has already been shown to give users a learned sense of direction.
  • Human 2.0 and Homo Intelligentes are concepts, not species. An IEEE white paper labels gradual human-digital symbiosis as Human 2.0; a 2024 article groups augmented, bionic and cyborg-like humans as Homo Intelligentes within an Industrial Revolution 5.0 story.
  • Nanobots linking brains to the cloud: a forecast. A 2019 Frontiers paper and futurist Ray Kurzweil predict that tiny robots in the bloodstream could one day connect neurons to cloud computers for instant knowledge access and full-immersion virtual reality, within roughly 20-30 years.

Not supported by the evidence 1

  • Biomarkers cannot yet build super-resilient soldiers (claim). The military volume found no consistent biological marker of resilience and no reliable way to enhance one. A newer systematic review names promising candidate markers (cortisol, DHEA, neuropeptide Y, IGF-1), but none is validated as a test, and none is a lever for building super-resilient soldiers.

Other themes