Can someone explain what is new about this generation of the tech?
People have been able to move mouse cursors and type using only their brain and tiny implants for decades... so far Neuralink seems to just be repeating these experiments, but receives a lot more hype.
Here's an article from 2006[1] with someone moving a mouse cursor and clicking things with a similar tiny brain implant.
Is there something new here, or just the Musk train?
>the participant was able to produce about 90 characters per minute, easily topping the previous record for implant-driven typing, which was about 25 characters per minute.
So it appears to enable typing at > 3x the speed of previous efforts. At least for this one particular person.
For me, having also seen the cursors and clicking type stuff in the past, what seems new to me (although I don't know for sure that it actually is new) is that they can now read minds on the fidelity of unique letters rather than just more simple "directions" like up, down, in, out.
Previous efforts were more inaccurate, you may remember hearing jargon like “alpha waves” or “beta waves”. Participants would learn to move cursors by learning to create the neurological activity that was being listened for.
This, on the other hand, is watching for complex neural activity, in that it learns what pattern appears when a participant pictures drawing an A.
Think of the difference like: before the input was controlled by turning your whole body, and now the input can take individual sign language letters.
>hearing jargon like “alpha waves” or “beta waves”. Participants would learn to move cursors by learning to create the neurological activity that was being listened for.
Previous efforts included moving a cursor with the implanted Utah array, which I don't think was just based on broad brain waves was it?
Are you thinking of non-invasive devices? Neuralink didn't invent implanted microelectrode arrays. Some of their press has been around being the first one wirelessly connected to a receiver, but it wasn't first there either. Mainly their roadmap for scaling to more and more electrodes and their custom chip for processing is what's new I think, but they aren't sure if the polymer coating they are using for will hold up long term.
You can do that without implants and just an eeg headset. I am still not sure why people think it’s a great idea to put things in the brain for non life threatening reasons
Agreed. The divorce rate of married people who get these government regulatory approved BCI interfaces is also extremely high. There are also other extremely severe problems that such interfaces cause.
EDIT: To the downvoters, it’s true, and there are other severe issues with these implants. Have a read, since many of you are laymen:
Don’t act like you can’t compare a group of married people suffering from the condition that doesn’t get the implant and a group of married people with the condition that do get the implants. Did you honestly think they were drawing the conclusion simply from comparing the rate of divorce of individuals with the condition to groups without the condition? It sounds like you basically just concluded the other side knows nothing about science because they mentioned divorce rate in a negative light, when you were lacking core details of their study
It is believed that a lot of this is due to a shift in identity. Look at the examples of this used in people with epilepsy (medically stable for years, that has not progressed) for example. The divorce rates are similar with other users of regulatory approved BCIs.
A local publication had one of the best first person accounts of TBI and the resulting rehab I’ve ever read and I always think about this whenever the subject comes up.
> The divorce rate of married people who get these government regulatory approved BCI interfaces is also extremely high.
Divorce rates spike after diangosis of and successful treatment/recovery for/from all kinds of chronic health issues, as they do for all kinds of major life changes. Disruption in living patterns changes the context and patterns of relationships.
Don’t act like you can’t compare a group of married people suffering from the condition that doesn’t get the implant and a group of married people with the condition that do get the implants. Did you honestly think they were drawing the conclusion simply from comparing the rate of divorce of individuals with the condition to groups without the condition? It sounds like you basically just concluded the other side knows nothing about science because they mentioned divorce rate in a negative light, when you were lacking core details of their study
Leggett’s identity changed again once the device was gone. Now she knew great loss, but she also knew things that had been impossible to understand before the device. Like many people with epilepsy, she had often found herself fuzzy for a considerable amount of time after a seizure. That state made it very difficult to notice the signs that preceded seizures which could act as a natural warning light. These days, when she gets a funny, flip-floppy feeling inside, she takes anti-seizure medication. She’s not always sure. Sometimes she gets her husband to weigh in. He says, “Go with your first instinct,” and usually she takes a pill. She is now seizure-free.
Just FYI: the article mentioned Neuralink as context for the general public (and with a bit of snark if I’m reading it right) but this was done by an independent academic research group.
To answer your question, it is merely a continuation of research in the field.
I see the promise of Neuralink as being able to read and write to computers/ai directly at high bandwidth. It's a long way from that, but it's what its going for.
Probably a good analogy is that we had letters and fax machines before the internet, the internet is just much faster with much higher bandwidth.
> Designing the full stack to be minimally intrusive seems unique.
Using the words "minimally intrusive" in the context of open brain surgery is quite amusing to me.
No matter who or what performs the surgical intervention, it still involves cracking open and replacing part of the skull and inserting foreign bodies into cerebral tissue.
The basic idea of the older EEG setups is you have a net of 128 or 256 electrodes on your head, and then the program signal processes all of the waves from them. Problem is your skull makes the waves bounce around, adding a lot of noise and making it hard to parse much signal from the data.
Neuralink uses an implant, inside your skull, so that this noise isn’t a problem.
Not true: You ideally still need as many electrodes as feasibly possible for a more complete BCI interface, implanted or not, regardless of the resolution (constant bitrate) of the electrodes. It is true though that certain implanted electrodes have higher resolution than transitional EEG caps. But at this point, it is not that big of a deal because neurosurgeons can only implant a very limited amount of electrodes. Even if they could implant 256 electrodes, the processing of this would be extremely limited. Even a standard 19 channel EEG cap can require a top-of-the-line computer (think i9 processor, 64 GB RAM, or better) for some use cases, due to latency issues involved especially with the amount of signals being used.
Also: With EEG caps, we have filters that remove the noise from movement and muscle activity very well.
The paper published says they used RNNs to decode handwriting. Similar research from what seems to be the same team used ReFIT Kalman filter around 2011, but this approach looks to produce better results.
As another poster said, they used RNN to decode handwriting. While there was an increased character rate observed, that is substantial compared to other such systems, that is not necessarily what is groundbreaking here. You would probably need a very powerful desktop (or a laptop with a desktop processor, >$4,000) with the best imaginable specifications to pull this off. So, it is of extremely limited use for the disabled individual.
With brain computer interfaces you use motor imagery [1], where you imagine motor movements (e.g. moving your left arm so that your palms face outward, and imagining the position of your shoulder, arm, and fingers) to control the BCI interface, such as in the case with prosthetics or this case handwriting. I imagine since this is fine motor movements (a different brain circuit, too) that it is much more difficult to visualize.
Motor imagery is also used in stroke rehabilitation. It is also used for sports performance.
Graded motor imagery [2] is a variant of motor imagery and there are apps for that. It basically rewires your brain’s pain circuits and makes pain more manageable and more controllable. It works for chronic pain, of any origin.
For example, in graded motor imagery, apps show you images of arms/legs/etc. in contorted (twisted) positions and at various angles in the picture, and based on the positions of the fingers/toes/etc. you are supposed to identify if the arm/leg/etc. is a “left arm” or “right arm” or “left leg” or “right leg” [3][4].
I have tried graded motor imagery (via apps) and I have excellent spatial skills: it is a quite difficult exercise.
There currently is no huge difference between electrode caps and implanted electrodes (there are multiple types) except that electrode caps are super dorky. The resolution (constant bitrate) is higher for some implanted electrodes, but not by a ton compared to an electrode cap. Generally, a ton of electrodes are not implanted, either.
So, everyone drooling about Neuralink really has no idea what they are talking about here. They have a tremendously long way to go with Neuralink and Musk is making claims that may never even be technically possible.
Also, the (marriage) divorce rate of people getting these neural brain computer interface implants (that are regulatory approved already) is extremely high. People should think about this and other social matters, instead of getting all excited about concepts (not technology) such as Neuralink.
/You would probably need a very powerful desktop (or a laptop with a desktop processor, >$4,000) with the best imaginable specifications to pull this off. So, it is of extremely limited use for the disabled individual./
Not necessarily. They have an extremely small-dimensional input, and have a pretty straightforward problem. The fact that they can train a good model with just a few hundred sentences suggests that it's an easy problem...
For other context, I do neural speech generation on phones for my day job, using an RNN that makes 4000 inferences per second. It works fine on a single thread with most phones produced in the last few years. Another helpful point of context might be 'swipe'-style phone keyboards, which are often RNN based, and turn paths into words.
The focus on giant-model work I think hides how effective small models can be, and how much progress has been made on making models run faster in limited resource environments.
(Do you have a reference on the divorce rate? Not sure I understand the causal link there...)
My work in AI revolves around biophysical signals. I do not use AI generally in this case but I use a 19 channel EEG (using an EEG cap) for experiments in closed-loop controls. It requires a lot of RAM (ideally 32 GB if not more) to prevent latency.
Thanks for the article; it was a super interesting read. My takeaway was that poking brains can lead to major personality changes, which can lead to divorce and other bad outcomes. (Just hearing 'leads to divorce' made me wonder if it was due to previously non-communicative people expressing themselves in ways they couldn't before... but sounds more like a 'sometimes you get subtle kinds of brain damage' problem.)
People have been able to move mouse cursors and type using only their brain and tiny implants for decades... so far Neuralink seems to just be repeating these experiments, but receives a lot more hype.
Here's an article from 2006[1] with someone moving a mouse cursor and clicking things with a similar tiny brain implant.
Is there something new here, or just the Musk train?
[1] https://www.nytimes.com/2006/07/13/science/13brain.html