Jul 18 2019

Neuralink To Begin Human Trials

I’m still trying to figure out if Elon Musk is a mad genius or a supervillain. Perhaps that’s a false dichotomy. Seriously, I do like his approach – he has billions of dollars laying around, so he decides that we need some specific technology in order to build the future, and he builds a company dedicated to developing that technology. Wherever he sees holes, he tries to fill them.

SpaceX has been, in my opinion, his most dramatic success. He has pioneered the technology of reusable rockets, and anyone who has seen one of his falcon rockets landing vertically has to be impressed. Tesla cars are impressive as well, but from what I understand he still has to make the company profitable. I’m still skeptical about the hyperloop, but at least he’s trying. It all depends on how cheap he can make tunneling, and the real innovation may be in his Boring company.

Not all of his companies involve travel. He also wants to change humans, in order to ultimately keep up with the AI he thinks we will inevitably create. In 2017 he tweeted, “If you’re not concerned about AI safety, you should be. Vastly more risk than North Korea,” along with a picture declaring, “In the end, the machines will win.” The existential threat of AI is a separate question.

Now for most people, if you are worried about AI you talk about it with your friends and colleagues. Perhaps you have a blog where you can share your concerns with the world. But if you are Elon Musk you can start a 100 million dollar company designed to thwart the perceived threat. So that’s what he did.

Neuralink is a system for interfacing the human brain with a computer (brain-computer interface, BCI). Musk’s goal is to develop the technology for merging humans with artificial intelligence, so at least we will be able to keep up. As with the hyperloop, the technology to get there may be more important than Musk’s goal itself. There are many uses for a BCI other than merging with AI.

There are already several designs for such an interface, and many labs and companies working on the technology. In fact we already have some very crude versions in clinical use, such as deep brain stimulation. So the idea is nothing new. There are two basic approaches that have been successfully tested in labs. Either the subject wears a cap of electrodes, which read brain signals through the scalp, or you implant electrodes on the brain surface. Implanted electrodes are more invasive, riskier, and eventually develop scar tissue which limits their lifespan and can potentially cause problems. They are also rigid, while the brain is pulsating, and so they tend to move with respect to the brain. The scalp electrodes are much easier and safer, but their resolution is much lower.

One approach to solve these limitations is to develop flexible electrodes. They will be less irritating to the brain, and will be able to move as the brain pulsates in order to preserve their relationship to brain tissue. Another approach is to implant electrodes within veins inside the brain. This gets them a lot closer than scalp surface electrodes, but not as close as brain surface or implanted electrodes, but may be a good compromise for now.

The electrodes also have to communicate with a computer. For now this computer is outside the skull, which means you need a connection, or the electrodes can communicate wirelessly. You could alternatively have computer chips inside the skull so the system is self contained. But then you have to deal with the heat generated by the chips, and you need a way to power them.

I believe it was in the book 2100, a sequel to Clare’s 2001: A Space Odyssey, where it had become common practice for people to get their “skull cap.” The top of the skull was replaced with a supercomputer, which then projected tiny hairlike electrodes down into the brain. This may be the ultimate solution – the computer does not have to be inside or outside the skull, it is the skull. It could them be connected for firmware updates or recharging, which could also be done wirelessly. Heat dissipation may still be an issue, but computer chip designs are beginning to prioritize greater efficiency in order to produce less heat. Perhaps quantum computers will do the trick.

In any case, we are a long way off from that vision (although 2100 is probably reasonable), so what is the next step? The Neuralink system involves many tiny hair-like electrodes, 300 in all, which are placed through the skull through tiny drilled holes. They are also flexible, so that they can move with the brain. These would connect to an external computer, although future designs may be wireless. Right now the electrodes connect to a chip that sits behind the ear. That is the hardware vision, which has already been successfully tested in monkeys.

The software for BCI is already fairly advanced, and with deep computer learning it seems will remain far ahead of the hardware. AI algorithms will need to interpret signals from the brain in order to know what the subject is thinking. This process could then be reversed, to communicate information to the brain. Given the incredible advances in AI, I am confident this aspect of the BCI will not be a problem.

So how is Neuralink doing? What experts are saying so far is that Musk has put together a good team. They have gathered the technology already developed, and put it together in a unique way. There are no breakthroughs in what they are doing, but they are pushing the ball incrementally forward.

Now Neuralink wants to start testing in humans, and they have applied to the FDA to do this. They hope to begin trials next year. They admit there are years of research ahead before the system is functional. There are a lot of potential applications, however, long before we get to augmenting humans with AI.

Anytime there is a circuit in the brain that is not working and resulting in a disorder, there is the potential to replace or augment the function of that circuit with a computer chip. One example in Parkinson’s disease – the extrapyramidal system, which is a feedback loop that adjusts the connection between thoughts and movement, fades away because of the death of dopamine secreting neurons in the substantia nigra. The pharmacological treatment is to replace the dopamine. This works for a while, until so many neurons have died that the circuit is essentially gone. You can still use dopamine to simulate the effects, but there is no feedback buffer, so patients constantly go through too much and too little movement as their blood levels vary.

Deep brain stimulation can be used to reduce the resulting tremors, but not replace the full function of the lost circuit. If, however, we could connect to the circuit with tiny electrodes, and replace their function with a computer chip that uses deep learning AI to learn how to modulate movement, this could be a literal cure for Parkinson’s disease.

The same can be true for spinal cord injuries, or even stroke. Computer chips could replace lost brain function, as long as they can be integrated into the brain’s communication network. Some types of coma could theoretically be treated, ones in which the brain loses the stimulation critical to keeping it awake.

The most interesting application will be how the brain interfaces with additional memory. Will remembering information stored in a computer chip feel natural? If so, we could put a calculator in your brain, or give you a perfect sense of time, or location. Now we are getting into brain augmentation. This, of course, is what Musk is hoping for.

Finally, could such a BCI make you smarter – not just by giving you access to more information, but enhancing your ability to think, problem solve, and create? At some point will you be more the computer than your biological brain? If so, then perhaps the death of your biological brain will become inconsequential – your digital brain will live on, and at some point that may be more you then your meat brain. Musk’s vision of merging with our own AI will be complete.

Like this post? Share it!

No responses yet