SWITCHED ON
The daily technology series nobody asked for but everyone needed
The Mind Machine: Brain-Computer Interfaces and the Neural Frontier
A man with a chip in his brain moved a cursor with his thoughts. That was the easy part. Now comes everything else.
Every other technology we have discussed in this series operates outside the body. This one goes inside your skull. The questions it raises about identity, privacy, and what it means to have a thought that is genuinely your own are not philosophical abstractions. They are engineering problems that nobody has solved yet.
Yesterday we talked about CRISPR and genetic engineering — the technology that edits the source code of life, the first approved gene therapy for sickle cell disease, the catastrophic recklessness of He Jiankui's germline experiments, and the designer baby question that the scientific community has been carefully not answering for a decade. Today we are staying inside the body but moving north. Considerably north. We are talking about brain-computer interfaces: devices that connect neural tissue directly to machines, read the electrical signals produced by thought, and translate them into actions in the digital or physical world. The field has been building for decades in academic laboratories and rehabilitation clinics. It arrived in the popular imagination the moment Elon Musk got involved. As with most things Elon Musk gets involved with, the reality is simultaneously more impressive and more complicated than the headlines suggest.
Let's start with what brain-computer interfaces actually are, what they have genuinely achieved, and then work our way toward the questions that the enthusiasm tends to skip over.
01 — What a BCI Actually Is
A brain-computer interface is, at its most basic, a system that reads electrical signals from neurons and translates them into commands for an external device. The brain produces electrical activity constantly — neurons firing in patterns that encode everything from sensory input to motor intention to the experience of reading this sentence. A BCI intercepts some portion of that activity, processes it, and uses it to control something: a cursor on a screen, a robotic arm, a speech synthesiser, a wheelchair.
The approaches vary enormously in their invasiveness and their resolution. At one end, electroencephalography — EEG — places electrodes on the scalp and reads the aggregate electrical activity of large populations of neurons through bone and skin. It is non-invasive, relatively cheap, and produces signals that are blurry in the way that trying to listen to a single conversation in a crowded stadium is blurry. Useful for some applications, limited for precision tasks.
At the other end, intracortical arrays — grids of tiny electrodes implanted directly into brain tissue — can record the activity of individual neurons with high fidelity. The resolution is orders of magnitude better. The surgery required to implant them is brain surgery. The trade-off is not subtle.
The history of BCI research is substantially a history of helping people who have lost the ability to communicate or move. For this population, the invasiveness calculation looks very different than it does for a healthy person considering an elective neural implant.
Between these extremes sits a growing range of intermediate approaches — electrocorticography, which places electrodes on the surface of the brain without penetrating it; various forms of minimally invasive endovascular approaches; and the emerging category of flexible, high-channel-count implants that attempt to combine the resolution advantages of intracortical recording with reduced tissue damage over time.
02 — What Has Actually Been Achieved
The clinical achievements of BCI research are genuinely remarkable and significantly underreported relative to the coverage devoted to Neuralink announcements. The BrainGate consortium — a collaborative research effort involving Brown University, Massachusetts General Hospital, and other institutions — has spent two decades developing intracortical BCI systems for people with paralysis. Participants with tetraplegia have used BrainGate systems to control robotic arms, type on computers, and operate tablets purely through imagined movement. The signal processing required to translate the intent to move a hand into a usable control signal has advanced substantially over this period.
In 2023, researchers at the University of California San Francisco reported a system that decoded attempted speech from neural signals in a woman who had been unable to speak for eighteen years following a brainstem stroke, producing synthesised speech at rates approaching natural conversation. The participant's face avatar, animated by the decoded signals, expressed what her voice could not. The demonstration was, by any reasonable measure, extraordinary.
Separately, work on closed-loop systems — BCIs that not only read from the brain but write back to it, delivering stimulation in response to detected neural states — has produced results in treating conditions including Parkinson's disease, depression, and epilepsy. Deep brain stimulation for Parkinson's is already established clinical practice. The emerging generation of adaptive stimulation systems, which adjust their output based on real-time neural monitoring, represents a meaningful advance over static stimulation parameters.
03 — What Neuralink Is Actually Building
Neuralink, founded by Elon Musk and a group of neuroscientists in 2016, has spent several years generating considerably more media coverage than published research. This is a characteristic of Musk-adjacent ventures and does not, by itself, tell you much about the quality of the underlying work. What it does do is create a gap between public perception of where the technology is and where it actually is.
Neuralink's device — the N1 implant — is a small chip connected to a flexible array of electrodes, implanted by a purpose-built surgical robot. It records from a large number of electrode channels simultaneously and transmits data wirelessly. In early 2024, Neuralink implanted its device in a human patient for the first time. The patient, Noland Arbaugh, has quadriplegia following a diving accident. He subsequently demonstrated the ability to control a computer cursor and play chess and video games using the implant. The demonstration was real. The achievement — a paralysed person controlling a computer with thought — was not new, having been demonstrated by BrainGate participants years earlier. What Neuralink brought was wireless transmission, a fully implanted device with no external hardware, and the company's characteristic talent for presentation.
Later in 2024, reports emerged that a significant proportion of the electrode threads in Arbaugh's implant had retracted from the brain tissue, reducing the number of functioning channels. Neuralink adjusted its signal processing software to compensate, with partial success. The retraction problem — electrodes pulling away from tissue over time due to micromotion between the rigid implant and the soft, moving brain — is a known challenge in the field and not unique to Neuralink. It is also unsolved.
The gap between "a paralysed person moved a cursor with their thoughts" and "a consumer product that healthy people will voluntarily have implanted in their skulls" is not an engineering gap. It is several engineering gaps, a regulatory gap, a safety evidence gap, and a very large cultural gap, all stacked on top of each other.
04 — The Privacy Problem Nobody Is Ready For
Every technology in this series has raised privacy questions. Facial recognition catalogues your face. Social media algorithms profile your behaviour. Data brokers sell your history. Brain-computer interfaces introduce a category of privacy concern that makes all of these look relatively tractable: neural data.
Neural signals contain information about intention, attention, emotional state, and potentially the content of thought itself, at least in the vicinity of whatever cognitive process is being recorded. Current BCIs are designed to extract specific signals — the intention to move a hand, the attempt to produce speech — but the raw data stream contains considerably more than that. What you are paying attention to. How you respond emotionally to stimuli. Patterns that, with sufficient processing, might reveal things about your mental state that you have not chosen to disclose to anyone.
The legal frameworks that govern this data do not exist in adequate form anywhere. Neural data is not clearly covered by existing medical privacy law in most jurisdictions. The companies building consumer-facing neurotechnology are not, in the main, medical device companies with established compliance cultures. They are technology companies with established data monetisation cultures. The question of what a neurotechnology company's terms of service permit them to do with your brain data is not currently answerable in a way that should reassure anyone.
Several jurisdictions have begun attempting to address this. Chile amended its constitution in 2021 to include neurorights — the right to mental privacy, personal identity, and mental integrity. A small number of US states have passed or are considering neural data privacy legislation. These are early and incomplete responses to a problem whose full dimensions are not yet visible, because the technology that will create the problem at scale does not yet exist at scale. This is, as a governance approach, suboptimal.
05 — The Identity Question
There is a deeper question underneath the privacy question, and it is one that neurophilosophers have been circling for years while the engineers have been building: what happens to the concept of a self when the boundary between mind and machine becomes permeable?
This is not as abstract as it sounds. People who use advanced prosthetics report, over time, beginning to experience the prosthetic as part of their body — not as a tool they are operating but as a limb they inhabit. Researchers studying BCI users have noted similar phenomena: the computer cursor controlled by thought begins to feel, neurologically and experientially, less like a controlled object and more like an extension of agency. The brain is plastic. It adapts. It incorporates.
If the device you have implanted in your brain is connected to external systems — the internet, corporate servers, other devices — and if those systems can, in principle, send signals back as well as receive them, the question of where your cognition ends and the machine's begins is not a poetic metaphor. It is a technical specification that nobody has written yet. The prospect of write-access to the brain — stimulation that influences thought, mood, or decision-making — raises questions about autonomy and manipulation that dwarf anything raised by social media algorithms, which at least operate outside the skull.
Tomorrow we are moving from the frontier of what technology does to the human body to what technology is doing to human lifespan itself. Anti-aging technology: the biology of why we age, the serious science behind efforts to slow or reverse that process, the billionaires funding it, the treatments that are already being sold, and the question of what a world with meaningfully extended human lifespans actually looks like. See you then.
Switched On is a daily technology series covering AI, social media, data privacy, and the digital forces reshaping modern life — with no corporate spin, no false comfort, and absolutely no mercy for buzzwords.
