As a lapsed subscriber, I get emails from the MIT Technology Review from time to time, and yesterday's about work with brain computer interfaces caught my eye. References 1 and 2.
The point of the work is to help people who can no longer talk very well because of damage to their brain, to maintain contact with the world - and with their computers. In particular people with what we in the UK tend to call motor neuron disease. Gemini was very firm that this sort of invasive procedure is completely prohibited with healthy people; the risk of complications arising from the intrusion are just too great.
As can be seen from the snap above, lifted from reference 2, the machinery is both intrusive and rather cumbersome. Nevertheless, the main point of the paper is that it is something that can be used at home for months, if not years, and that from this subject's point of view, it was well worth while.
What the machinery does is synthesise speech from speech thoughts in the motor cortex and simulate mouse control - and certain other screen functions - from thoughts about that. To the point that the subject can manage his email and participate in video calls.
With the point of particular interest to me being that if you place arrays of electrodes - for which see reference 4 - on the surface of the brain, with the tips of the electrodes about 1.5mm down, in the middle of the cortical layer, you get signals which can be successively analysed into phonemes, words and sentences. With impressive accuracy.
This is a big step up from telling whether you are thinking about playing tennis or watching elephants in a zoo - which activities can be distinguished using arrays of (much bigger) electrodes placed on top of the scalp, without the intrusion.
This particular subject, once he had got the hang of things, asked for a 'private' feature, whereby he could think in private, without his thoughts being transcribed onto the computer adjacent. I suppose he could also insist on redacting whatever transcripts there might be.
Something which I had not thought of is that the brain appears to drift. This means that the system needs to be regularly recalibrated, either from time to time - or continuously, in the background, if things can be so managed. I had thought that once the brain had learned something, the neurons involved and their connections would be set, more or less for life. Or until recycled by garbage disposal.
There has been a problem in the past with the performance of these arrays degrading over time, say over not very many months. The present work is clearly an improvement in this regard, but I did not work out quite how much of an improvement.
In the course of reading this paper, Gemini was very helpful in providing tutorial material about various tricky words and phrases. But sadly, he lost our conversation and I did not make much sense of his explanation about how this happens - although it seems clear that it does happen often enough. You can, of course, get him to go through it all again, but that is not quite the same - or very convenient.
Nor was I able to make much sense of his advice about how to log out of Gemini direct, from within Gemini. I usually log out of Gemini by logging out of gmail, which does come with that feature - and which I had thought might have been mixed up with loss of conversation.
PS 1: I remembered that Mr. Musk is into this sort of thing too with his Neuralink company (reference 3), which I first came across near ten years ago. So far I have found out that he is using much more cunning, less damaging electrodes, but that this work is a bit behind that described here.
PS 2: in the margins, I came across a report that the wonks of Bond Tower, across the road from Vauxhall railway station, were up last night puzzling over a report from G7 that POTUS had broken wind upwind of our Starmer, in conversation with Macron at the time. What might this mean for the special relationship? A bit tricky for them in that the surveillance footage concerned all comes from their French colleagues.
References
Reference 1: Casey Harrell uses his implants to talk to friends and family, read to his young daughter, and perform his job – Jessica Hamzelou – 2026. MIT Technology Review.
Reference 2: Long-term independent use of an intracortical brain–computer interface for speech and cursor control – Nicholas Card, Sergey Stavisky, David Brandman and others – 2026. Nature Medicine.
Reference 3: https://neuralink.com/. The Musk connection.
Reference 4: https://blackrockneurotech.com/products/utah-array/. Nothing to do with the finance people, but everything to do with the long running Utah arrays.
Reference 5: https://en.wikipedia.org/wiki/ALS. ‘Amyotrophic lateral sclerosis (ALS), also known as motor neuron disease (MND) or Lou Gehrig's disease, is a rare terminal neurodegenerative disease defined by the progressive loss of both upper and lower motor neurons that normally control voluntary muscle contraction. ALS is the most common of the motor neuron diseases…’.
Group search key: aisk.


























.jpg)























