Neuralink: An Integrated Brain-Machine Interface (quick overview)
Today I’ve read a white paper written by Elon Musk presenting his company Neuralink. The paper’s source can be found at the bottom of this article.
Neuralink is a private founded by Elon Musk and eight partners in 2016. The company aims at developing brain-machine interfaces that would be able to connect humans and computers.
Brain-machine interfaces have a lot of potential in healthcare: from allowing people with paralysis to easily communicate with others to curing Alzheimer’s disease. The principle used by Neuralink and allowing us to imagine all these applications rely on few facts about how does our brain work.
The indivisible unit of communication in all animals is the neuron. The neuron is a specialized cell that is able to receive information (physically or chemically) and to react to this information through a pulse called an action potential. Action potentials can be seen as electrical spikes that can propagate along the neuron membrane and be detected by surrounding neurons. By organizing themselves in networks and by specializing themself in specific input and output information, neuron networks are one of the core components of intelligence.
Our brain contains a lot of neurons (around 15 billion) and these neurons are intermediates in all our physiological interactions with the world (information we receive like images or smells and information we send like movements or speech). Since everything (or almost) passes by our brain, being able to interact with it through neuron action potentials open the way to any applications interfering with how we interact with the world.
Neuralink’s approach to the brain-machine interface as described in Elon Musk’s white paper consists of 3 components. The first component is a probe (an array) which contains a lot of very thin polymer threads, each of them having electrodes attaches to them. The white paper precisely describes a probe containing 96 polymer threads of 32 electrodes each. Each thread is supposed to be directly inserted into the brain which requires using a material (a special polymer here) that is not too rigid to not trigger an immune response or bleeding and rigid enough to go through brain tissues. Each electrode is able to analyze action potentials from its surrounding neurons and the input data from the 96*32 electrodes are analyzed in order to identify neuron spikes and map them to specific activities such as communicating or movement.
The second component is a surgical robot that is used to insert the probe’s threads through the brain. This is done by first removing a piece of the scalp (the human skin under the hair), digging a hole through the skull, and inserting each thread precisely enough to avoid piercing any significant blood vessel.
The third component is the electronics which are supposed to detect spikes with high resolution and high accuracy while being energy efficient. This poses many challenges since the device must be able to perform signal compression and possibly wireless data transmission while detecting spikes in real-time (while most spikes detectors run offline which allows for powerful and accurate algorithms) while working in a very high signal density region.
For further information, Neuralink’s website provides comprehensive explanations about the company and its devices: https://neuralink.com/. The white paper source is at the bottom of the article.
Source: Musk, E., 2019. An integrated brain-machine interface platform with thousands of channels. Journal of medical Internet research, 21(10), p.e16194.