Brain-computer interface · Grey Matter

A brain-computer interface (BCI) records activity from the brain and turns it, in real time, into an action outside the body, such as moving a cursor, typing, speaking through a synthesiser or moving a robotic arm, and today its main use is giving back communication and control to people who are paralysed.

Brain-computer interface. A brain-computer interface (BCI) records activity from the brain and turns it, in real time, into an action outside the body, such as moving a cursor, typing, speaking through a synthesiser or moving a robotic arm, and today its main use is giving back communication and control to people who are paralysed.

Every BCI has the same four stages: a sensor (electrodes on the scalp, on the cortex, inside it or in a vessel next to it), amplification and digitisation, a decoder that maps features of the signal onto an intended output, and feedback, since the user sees the result and adjusts. Most implanted systems record over the motor cortex, where the intention to move a hand or to articulate a word is represented even when the body can no longer carry it out. In 2023 two speech systems in people who could not speak decoded attempted speech at 62 and 78 words per minute, with a language model as the last stage turning noisy guesses into sentences.

Invasiveness sets the ceiling. Scalp EEG supports slow selections and spellers; signals from the cortical surface or from penetrating arrays support fluent cursor control and speech, at the cost of surgery.

Decoders learn the user and the user learns the decoder. Calibration takes minutes to hours, and recalibration is needed as signals drift with the tissue response and with time.

It reads intention and can read more. Speech systems in motor cortex also picked up some words people only said to themselves during tasks such as counting, and the same study showed safeguards, including a decoder that stays locked until the user imagines a chosen keyword.

It is a medical device for now. Every high-performance system is in clinical trials, with the companies on this half of the section racing toward the first approval for a permanent implant.

The decoding is far ahead of the access.

Modern decoders turn a few hundred channels into fluent text; the hard part is placing those channels, keeping them working and doing it without a neurosurgeon.

Questions: Could a speech implant pick up words its user never meant to say aloud? A study of four people with speech implants in motor cortex found that imagined sentences could be decoded in real time, and that some free-form inner speech could also be decoded while participants recalled sequences or counted silently. The same study demonstrated ways to prevent it, among them keeping decoding locked until the user imagines a chosen keyword. It is direct evidence that a device built to restore communication can capture more than the user means to share, which turns mental privacy into a design requirement for speech implants. How can someone give informed consent to a brain-computer interface when they can barely communicate? The people a brain-computer interface could help most, those with ALS, brainstem stroke or locked-in syndrome, may communicate only through eye movements, letter boards or a carer, slowly and with effort. Their capacity to decide is usually intact, so the task is to give them enough time, accessible formats and independent support to understand the risks and to ask questions, and to check understanding as the study goes on. Reviews of neural device research add that strong hopes for benefit, and dependence on the research team for care, make it especially important to state plainly that early studies test safety first and to agree what happens to the device afterwards. What does an implant over the motor cortex read when a paralysed person intends to move? Neurons in motor cortex still fire with the intention to move even when the spinal cord or the muscles can no longer carry it out, and each neuron fires more for some directions or speeds than for others. An array recording a few hundred of them sees a pattern that changes with the intended movement, and a decoder trained while the person imagines or attempts moving maps that pattern onto a cursor's velocity, a click or, in the speech areas, the sounds a person tries to articulate. What it reads is intended movement, the output of planning, which is why these systems work best for actions the user tries to make.