select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
Elon Musk has trailed the announcement of a Neuralink ‘brain-machine interface’ expected on Friday 28th August 2020.
Prof Andrew Jackson, Professor of Neural Interfaces, Newcastle University, said:
“Elon Musk and colleagues posted this preprint last year: https://www.biorxiv.org/content/10.1101/703801v4. If I recall correctly, it showed data collected in rats. It will be interesting to see if this evening’s demonstration also uses animals, perhaps a non-human primate, or whether it will involve a human subject. If the latter that would be an impressive step forward (although worth noting that there have been human ‘Brain-Machine Interface’ trials already). Also, if using a human subject, it will be interesting whether it is an ‘acute’ demonstration of something intraoperative (i.e. placing electrodes short-term during a surgical procedure for something like epilepsy resection) or whether they have actually done a long-term implant in a person. Should know more later.
“It was impressive to see how quickly they got to the previous point (the rat work in the preprint), and will be interesting to see how far they get. Theirs is one of a number of efforts to ‘read’ the electrical activity of large numbers of brain cells. The Neuralink approach is to insert many flexible polymer threads into the brain using a sort of sewing machine. The threads attach to an electronic package implanted under the skin. Other notable approaches involve incorporating electronics onto small silicon needles (e.g. the Neuropixels probe developed by Tim Harris and colleagues) or using many small, wireless implants distributed throughout the brain (so-called ‘Neural Dust’). Only time will tell whether Elon has backed the right horse. One thing this does demonstrate is the potential for commercial investment to advance the field of neural interfaces. Until recently, neuroscientists were using some pretty old-fashioned equipment to record from the brain, so it’s great to see this kind of interest and investment from Silicon Valley.”
Dr Peter Bannister, Biomedical Engineer & Executive Chair, Institution of Engineering and Technology (IET), said:
“While the immediate practical impact of this announcement will be scrutinised over the coming weeks for its scientific and ethical rigour, it’s important to remember that these ‘moonshots’ are essential to inspire the next generation of engineers and scientists. Like the namesake moon-landings, or the Star Trek tricorder, ambitious projects and ideas like these have the ability to influence future technological advances for society at large and often lead to practical innovations in the near term. In the same vein, Covid-19 has set a huge challenge which has motivated and catalysed groups from across multiple industries to come together and produce innovative and out-of-the-box solutions to support us in our everyday lives.”
Prof Patrick Haggard, Professor of Cognitive Neuroscience, Institute of Cognitive Neuroscience, UCL, said:
Controlling movements by thought alone:
“The idea of a brain-computer interface (BCI) for paralysed individuals, where the signals from neurons in the brain are used to control movement of an external device, is an exciting one: many teams around the world have been working on this for several years. These devices offer the exciting possibility of restoring some autonomy of action to people who are unable to move. There have been demonstrations with several patients, but research is still at an early stage.
“New developments in BCI are always interesting. They should be published in the peer-reviewed scientific literature, with a clear and fair measurement of their performance, including any limitations. Scientific reviewers will typically evaluate what tasks can the patient achieve using the BCI, how quickly and how accurately can they communicate their intentions, and how easy it is for the patient to generalise the control from one task to another. The algorithms used to decode the patient’s neural signals should be explained, and the limitations of the system should be clear. Finally, the patients’ voice should be included in the design and evaluation of the systems wherever possible.
The problem of bandwidth:
“Up to now, BCIs can work reasonably well in limited, highly controlled settings. Patients can learn to produce patterns of neural firing that the computer can interpret, and then use to drive a robot or similar device. All BCI systems suffer from what engineers call ‘low bandwidth’ – the patient has to generate the right signals carefully and continuously over a period of time, even to make a simple movement choice like getting the robot to press either one button or another. None of these systems approach the effortless, fluent flow of everyday human voluntary actions. A key question for any new BCI technology will be: can it increase the bandwidth? Can it allow a patient to choose quickly, accurately and easily to perform a wide range of possible actions?
Risks
“The most successful BCIs are invasive – they involve inserting electrodes directly in the human brain tissue – so there will always be risks around the surgical intervention, and risks around long-term viability.
“There are also wider ethical considerations surrounding all neural interface technologies. The patient needs to be protected against the possibility that the algorithm decodes their intentions incorrectly. Responsibility for action needs to be carefully considered: how can we distinguish whether the action reflects the choice of the patient, or a decision of the algorithm, against the patient’s wishes? The communication loop between the patient and the algorithm is likely to influence the patient’s thoughts, in the same way that any change in the environment can influence how people think, so the effects of this influence on mental autonomy must be monitored carefully. While the aim is clearly to increase patient autonomy, it is important to ensure that BCI devices do not in fact restrict it. Finally, a very important distinction must be made between recording brain signals using electrodes, as a BCI does, and stimulating the brain to create a particular pattern of brain signals – possibly through the same electrodes. Neurotechnologies that stimulate the brain in principle offer the possibility for controlling a person’s thoughts. This always raises deep ethical questions about individual volition and mental autonomy.”
Prof Thomas Nowotny, Professor of Informatics in the School of Engineering and Informatics, University of Sussex, said:
“As far as I can tell, there is nothing particularly interesting happening in Elon Musk’s event this evening. From what I have seen he is going to demonstrate live recordings from human neurons using some kind of implanted electrodes.
“Recording from neurons in live and behaving animals is a standard procedure in Neuroscience and is decades old. The most modern incarnations of advanced system for invasive recordings include miniaturised head stages with wireless interfaces that can be worn by animals for prolonged times of months if not years.
“Similarly, invasive recordings with extracellular electrodes are also done in humans in a clinical context, in particular in the context of epilepsy with either acute electrode placements for on the order of days to identify the focus of seizures before brain surgery or permanent implants to provide deep brain stimulation. Other technologies include ECoG electrodes, which are implanted onto the surface of the brain. These techniques have been standard for at least the last 20 years. In terms of restoring function after sensory or motor function loss there is a plethora of work in many areas, too numerous to describe in detail.
“While interfacing with the brain or peripheral nerves will probably need improvements if such technology is going to become mainstream, I believe that the more challenging area is to make the signals that the artificial implants provide to the brain more useful. The brain is unbelievably flexible in interpreting signals that contain information but the effectiveness of for instance an artificial retina would be much improved if the signals fed into the optical nerve were closer to what normally would be received from the retina.
“In terms of brain-computer interfaces (BCIs) where the computer/machine is the receiver and tasked to interpret signals from the brain, it is equally the most challenging part to decipher the brain’s code rather than to measure its activity per se. Of course, there is room for improvement in terms of the number of recorded neurons and the temporal resolution at which they can be recorded, but I believe the real challenges for BCIs lie in the current lack of understanding in the way the brain functions.
“The recent announcements from Elon Musk look like old news to me and many neuroscientists have come to a similar conclusion, in particular colleagues who are doing brain recordings on a daily basis. Unless I’m wrong about the nature of the achievement, it looks like this may be someone who has not done their homework to look what is already well-established by scientists around the globe.”
Prof Christopher James, Professor of Biomedical Engineering, University of Warwick, said:
Is there enough information at the moment to be able to tell how likely this is to be successful?
“Hardly any! Brain computer interfacing, BCI (sometimes brain machine interfacing, BMI), has been done many times, many scientific papers/ popular press articles/ documentaries etc. So this could be “just another” BCI – in which case it could be “successful” – if success is “can we get a control signal out from the brain?” I’ve shown it, my students have shown it, we have controlled Scalextric cars with scalp BCI signals and I did a toy “brain to brain” experiment to show what we can already do. The devil is in the detail – what exactly is this BCI purporting to do? I can comment after the fact.
Is this being published in a peer-reviewed scientific journal as new scientific breakthroughs usually are?
“It doesn’t appear to be – but as said many BCI articles exist in the scientific literature, it’s a “done thing” – but usually in relatively narrow scope – move a cursor, move a wheelchair, etc. – although other more complex BCIs exist.
What are the main issues that would need to be addressed for a paralysed person to be able to move their limbs again or for a blind person to be able to see?
“In part it depends on how/why the person is paralysed – is it due to severed spinal chord with brain intact? Or due to brain damage? It will matter! Assuming an intact brain – we know already we can put electrodes over the motor cortex and detect the intent to move and thus control something – my daughter was moving a wheelchair with her thoughts in “James May’s Big Ideas” about 15 years ago! Moving limbs is “just” a level of complexity above that – more sensitive interface required, being able to decode the signals better and in real time – then how to control limbs – the actual limbs? An exoskeleton? A robotic limb? (the first being the hardest and the last being the easiest in this list!).
“For the blind aspect it’s similar – how is a person blind? If one assumes the visual cortex is intact that stimulating the visual cortex is possible, “Bionic Eyes” are already about (at different levels of refinement).
“Note that moving limbs is “output” from the brain, being able to see or hear again is “input” to the brain – different in scope and in technical requirements.
What does ‘human symbiosis’ mean?
“I am not sure and this sounds open to interpretation to me so I will not even try!
How easy is it to place electrodes in the brain?
“Easy, as in it has been done many times; it’s risky in the sense of any operation but not super high risk – the issue is: where? And how many electrodes? For a narrow scope an array of even a few hundred electrodes is small in reality. If you wanted the futuristic symbiotic man-machine you will need electrode arrays in many different brain areas – as with the brain WHERE you sense/stimulate makes a big difference. This scenario clearly would increase the risk of infection, scarring over the electrodes over time can cause issues too – making them less reliable etc.
What would you need to see to convince you this experiment had worked?
“It depends what is being claimed! Can electrodes be implanted successful? Yes – done. Can a person with such electrodes elicit a control signal to control “something”? Yes, done. Can a person’s brain be stimulated to elicit a response from the person? Yes – done. So, taken in isolation all of these “minor”/narrow in scope results have already been obtained… so what is being proposed here?
“What would impress me? Realtime control of complex actions/movements – repeatedly and with little error (oh and being able to move something whist doing something else like talking or whistling or whatever!) Being able to stimulate the brain to elicit complex responses, repeatedly and accurately (e.g. being able to infer higher order things like colour, intent, or whatever).
Any other comments?
“The above comment is pure speculation based on what is already out there – as I don’t know what is being proposed it is hard to judge but I will be looking with interest at what is being proposed. Perhaps he can do what can already be done but can do it cheaper? Safer? Faster? More accurately? In combination with other devices? All of these would be achievements – perhaps less boring in the eyes of the world but could make BCI steps closer to reality and would be useful nonetheless.”
Prof Sethu Vijayakumar FRSE, Personal Chair in Robotics and Director of the Edinburgh Centre for Robotics, and Programme co-Director for AI, The Alan Turing Institute, said:
“I have known and worked with some data on multi-electrode recordings from primates in late 1990s, with many of the innovations coming from Richard Andersen’s lab at Caltech on miniaturising the form factor of such devices and ensuring the ability to retarget the electrodes to ensure sustained recording capacity. The next decade saw forays into multi-electrode stimulation of the motor cortex in order to enable stroke patients or paraplegics to move various parts of their body or external devices– examples included robotic arm manipulation in real time at the Brown University, in collaboration with DLR, Germany. [https://youtu.be/ogBX18maUiM]
“Besides sustaining live activity (and reduce rejection) of live implantable electrodes, the biggest challenge has always been to interpret data to do something useful with it. The holy grail has been to increase the size of the electrode arrays, with the hope of using machine learning algorithms to identify cross-correlations and higher order information to interpret intentions and even, short circuit (or shunt) muscle activation pathways for action – without passing through the spinal cord.
“However, there are still big challenges in real-time data interpretation to disambiguate fine intentions. Our group has worked on surface Electromyography (EMG) and Inertial Measurement Units (IMU) based decoding of muscle activity to drive real-time adaptation in upper limb prosthetics as well as use of Function Electrical Stimulation (FES) of muscles to partially actuate muscles groups in conjunction with ankle, pelvis and lower limb exoskeletons. However, Neuralink’s approach that is much more invasive (directly reading from and stimulating parts of the human brain) to restore capabilities of the human body, where feasible to do safely, is definitely more challenging and potentially more rewarding that using wearable robotics.
“The biggest challenges I see (for a paralysed person to be able to move their limbs again or for a blind person to be able to see with this technology) are three fold. One is the long term, stable sustainability of implanted devices. Second, is the bandwidth and stationarity of the signals – this can change depending on precise location, cross talk and conditions like temperature and nature of task being carried out, such that quantity of information that can be effectively used may be limited to few bits. And finally, the most challenging is trying to decode a distributed signal without fully knowing or understanding the neural code – recording from more and more number of neurons will not solve this problem. I have only addressed the interpretation challenges, but obviously to close the circle, one needs to either directly interface with the human motor neurons or drive exoskeleton or wearable robotic devices from the output of these devices.”
Dr Dean Burnett, Honorary Research Associate, Cardiff University, said:
“There’s certainly nothing in the video released to tell us whether it’s legitimate or not. Looks like a standard PR video for a lab/technical company to me. But then, you wouldn’t really expect any different. There’s not a great deal about modern neurological techniques and methods that makes for impressive visuals.
“I’ve not heard any sign or mention of this in the general neuroscience news/feeds regarding publication in a legit journal. That itself is a bad sign. I get that Musk may want to keep proprietary tech to himself, but that’s not really how this sort of thing works. To be sure that it works they should have done extensive trials and studies, and ideally these would be under conditions which follow ethical approval rules and where the data produced is submitted to peer review, so relevant experts can determine if it’s been done right, if any bias or inaccurate assessment has creeped in, and so on. It not even being mentioned in the academic press (assuming it hasn’t been, it could have been but I’d almost certainly have heard about it before now if it had) is usually a very bad sign. I even made up my own ‘Burnett’s law’: “If someone claims to have made a groundbreaking advance/discovery, but hasn’t published any peer-reviewed data, but HAVE done a TED talk, then you should be extremely sceptical”. This isn’t a TED talk per se, but it is a big publicity event for the media, not for scientific scrutiny, so that immediately makes me very sceptical.
“It is unclear as to how this development would help people with paralysis. Not that it couldn’t, they just don’t make it clear. People become paralysed because the neural connections between brain and spinal cord to the relevant bodily area have been damaged/severed. The ideal treatment would be to restore these damaged links, by regenerating and reconnecting the nerves. Not sure how brain/tech interface would help that. It could be that they mean they’ll use the new technology as a sort of relay between the damaged sections, so signals from the brain can sort of be ‘texted’ to the damaged regions? Maybe the promised AI interface is what’s translating the neural signals into machine code and back again, in ways that make sense to the body.
“This is just conjecture on my part though. It could be that the chip in the patient’s head is used to transmit movement signals to a sort of technical exoskeleton which moves the individuals legs (for instance) as they would their non-paralysed body, but that’s not really restoring movement, it’s more an advanced prosthetic. Which is fine, and would still be great news, but it would mean the language being used could be a bit questionable.
“People having electrodes embedded in their brains which record neural activity is not a new thing. It’s been done for people with issues like chronic epilepsy for many years now, and such individuals are often recruited into scientific studies because of this rare but very helpful setup. I’m not saying Musk hasn’t done it, but that wouldn’t quite be the breakthrough that’s being implied by some, far as I can see. Unless he means something else, but what that may be is unclear so far.
“Along those lines, you can’t really implant electrodes into the brain of a healthy person for pure research purposes – it’s too risky and invasive a procedure for that. But it does beg the question of, if Musk has volunteers with embedded chips in their brains, where did he find these people? There should be an extensive paper trail for such a thing, to make sure nobody was coerced and everything was above board and done right.
“Restoring sight to the blind would also be a bit of a questionable claim. The eye isn’t a camera, and the visual signals relayed to the brain are terrifyingly complex. And depending on how long the individual has been blind/sight impaired, it could well be that their visual cortex has rearranged itself in response to the diminished or lost signals from the eyes. Ergo, going right back to ‘normal’ sight using tech, even if that were an option, is a very big ask, as you’d have to restore the visual cortex to its original sighted state too. If Musk can do that, he’s considerably more powerful and accomplished than any human in history.
“Final point: you hear claims along these lines a lot, usually in the hypothetical sense, but they typically ignore the variability of the human brain and how it develops organically based on our experiences and development, not just due to what our DNA says it should look like. There’s still this persistent belief that the brain is largely modular, that each specific bit has a specific role, and just that. This suggests that if a bit goes wrong and you can replace it with technology, it’ll work fine. But it’s not like that, because each brain is very unique in structure, and massively interconnected, and often pluripotential. If Musk has indeed found a way to insert technological components into the brain that can mimic or copy this level of sophistication, then he has indeed achieved a major breakthrough. I remain sceptical though.”
https://twitter.com/elonmusk/status/1298494373278220290
Declared interests
Prof Andrew Jackson: “No competing interests.”
None others received.