Aug 31 2020

Elon Musk Unveils Neuralink Pig

Three days ago Elon Musk revealed an update to his Neuralink project – a pig named Gertrude that had the latest version of the Neuralink implanted. (I first wrote about the Neuralink here.) The demonstration does not seem to involve anything that itself is new with brain-machine interfaces, but it does represent Musk bringing the state of the art together into a device that is designed to be commercial, rather than just a laboratory proof-of-concept.

Unfortunately, I have had to cobble together information from multiple sources. There does not appear to be a scientific paper with all the technical details spelled out, and the mainstream reporting is often vague on those details. But I think I have a clear picture now. The device is a coin-sized, 23 mm diameter and 8 mm thick. It was implanted “in” the skull, and also described as being “flush” with the skull. From this I take it that the device is not on top of or inside the skull, but literally replacing a small piece of skull. It has 3,000 super thin and flexible electrodes that connect to 1000 neurons. The device itself has 1024 channels (a channel reads the electrical difference between two electrodes).

The company also reports that it has an internal battery that can last “all day” and then recharge overnight. It also communicates to an external device (such as an app on your smartphone) via bluetooth with a range of 5-10 meters. As an electronic device, this is pretty standard, but it is good to have these features in a small implantable device.

The big question is – what can the Neuralink actually do? The demonstration, in this regard, was not that impressive (compared to the hype for Neuralink) – just the absolute bare minimum for such a device. It was implanted in a pig and was interfaced with neurons that connect to the snout. This demo device was read only; it could not send signals to the pig’s brain, only read from the brain. The demonstration consisted of Gertrude sniffing around her cage, and when she did so we could see signals from the neurons in her brain that were interfacing with the Neuralink.

To be clear, this could be a fantastically useful device. Essentially what we have is a portable brain surface and even deep brain EEG. This could be really useful for evaluating patients with epilepsy, for example. The large number of electrodes and the the fact that they are in contact with neurons would mean a high resolution view of a seizure. Also, the device is ambulatory (you can walk around with it and don’t have to be in a hospital). An epilepsy patient could have the device installed, then be monitored for a month or so for their brain activity, in-between and during seizures. Everything could be recorded portably on their smart phone and then sent to their doctor for analysis in real time. I hope that this purely medical application is not lost on Musk or the neuroscientists working for him (I would be shocked if it were).

Such a device would also be useful for research, trying to figure out how the brain works. It’s likely that similar devices will be implanted in monkey brains for years to come as part of neuroscience research.

But medical and research applications were not what Musk wanted to talk about. For him the “near term” applications were therapeutic. He specifically mentioned treating spinal cord paralysis and one of the early applications that will be researched. In this application a Neuralink  would connect to the motor cortex (or perhaps two, one on each side), and then a second device would connect below a spinal cord injury. The devices would communicate to each other, bridging the gap across the injury.

This seems like an obvious application, but my concern is that spinal cord injury results in much more than just an inability to get a signal from the brain to the muscles. Neurons are not just wires, they are living cells, that have biological functions beyond carrying electrical signals. Spinal cord injury can cause severe spasticity, for example, and this will definitely limit function even if a wireless signal restored a connection. Depending on the level of the trauma, you will also get areas of severe atrophy. You can’t communicate to a muscle fiber that is no longer there because it died. But still, this could serve as one piece to a solution, even a partial solution. Anything that provides some functionality could make a huge difference to independence and quality of life.

Perhaps a more feasible application is communicating to a robotic prosthesis. But as research is showing, these function best when they also provide sensory feedback. This requires not only the read capability demonstrated by Gertrude, but would also require “writing” to the brain.

Disorders in the brain itself would also require not only the capability of reading signals from the brain, but also sending signals to neurons in the brain, to either provide that sensory feedback, or replace or modify lost function due to disease or brain injury. This gets really complicated fast. So even though these medical applications seem to Musk to be the low-hanging fruit, I hope he realizes there are years or even decades of research to bring these applications to fruition.

To Musk the long term applications are currently in the realm of science fiction. He hopes to use Neuralink (obviously in a more advanced form) to interface humans with AI. He feels this is necessary in order to prevent what he thinks is the eventual take over of humanity by our own AI. If we instead merge with AI, then we can compete with them. Regardless of whether or not you think AI is an existential threat, merging with AI would be a great idea and I think is inevitable. The potential benefits are simply too great. But I think this is a project for 50-100 years.

Musk also said (perhaps offhandedly), “You could potentially download [memories] into a robot body.” What I think about this depends on exactly what Musk means. One interpretation is that a copy of your memories could be transferred to a computer. You would then create a “Max Headroom” version of yourself – a digital copy (whether or not it was used to run a robot or just exist as software). This is plausible but would require a much greater knowledge of the brain than we currently have, and a much more mature brain-machine interface technology.

But I hope Musk does not mean something akin to “transferring your consciousness” into a robot. That is not possible. Your consciousness is not a thing that can be transferred. It is your brain. At best you can make a copy of some of the information that your brain represents, but this would not be you.

It may be theoretically possible, however, to so fully merge with an external AI that over time you essentially become the AI. This is the only path, in my opinion, to silicon immortality. Imagine an advanced version of neuralink connected to a supercomputer with computing power greater than that of the human brain, and with neurological-like function (a neural net). The two components, your biological brain and the computer prosthetic brain, would essentially function as one whole, the same way the two hemispheres of your brain function. Your mind would be a brain-computer hybrid mind. Over time your memories and neurological function would be increasingly represented in the AI portion of your hybrid brain. If this were designed to redundantly duplicate all the functions of your biological brain, eventually that biological brain would become itself redundant. You would become mostly the AI portion of your hybrid brain.

While the futuristic potential applications of this technology are interesting to contemplate, I hope they don’t overshadow the near-term realistic applications. Musk has a potentially useful medical device on his hands. Perfect that, and worry about the sci-fi applications a little down the road.

No responses yet