By Link Daniel
Original Publication Date: January 21, 2021

The problem is how do we get information out of the brain or record neural activity and how do we get information into the brain or stimulate neurons in such a way that neural patterns are changed. It needs to be a two-way feedback loop between the brain and computer, whereby we can increasingly get more information out of the brain and more information into the brain. We want to change neurons or stimulate them in certain ways in order to perform certain actions outside the brain. We need to specifically increase the number of neurons we can interface with and we need to accelerate its growth curve in order to start exploring the brain’s frontier in a real way.

If we look at current brain computer interfaces, there are three useful metrics suggested by scientific author Tim Urban. First, scale or how many neurons can be recorded at the same time. Second, resolution, or the depth that the brain computer can record. This is divided into spatial resolution which measures closeness of how individual neurons are firing and temporal resolution which measures when activity of the recording is happening. Third, invasiveness, or as the name suggests how invasive do we have to go into the brain in order to install the brain computer.

There is fMRI, which relies on magnetic resonance imaging technology, in order to generate images of the brain and body. While it can scan the whole brain and be operated non-invasively, its resolution is rather low.

There is EEG, which puts an array of electrodes on the head in order to record electrical activity in different regions of the brain. An EEG, however, only records brain activity and does not produce an output. While its scale is high and it is also non-invasive, it has low spatial resolution and medium temporal resolution.

Up next is ECog, which is similar to EEG, except the electrodes are placed under the skull. It is therefore more invasive with a high scalability, yet spatial resolution is still low while temporal resolution is high.

Even more specialized, Local Field Potential is even more invasive as it relies on micro electrodes. A needle is inserted into the brain which then is able to read the electric charges from the neurons in that area. Its scale is therefore small as it only measures the region of where the needle is placed. Single-Unit Recording, which is even more invasive, has a very high resolution. It only captures the resolution of a single neuron, and therefore the scale is tiny.

We have brain computer interfaces that can restore sensory motor functions. For example, we have those that can give artificial eyes or ears. There is deep brain stimulation which has had some impact on people with mental health problems.

For the computer industry to take off we had Moore’s law. For the neuro revolution to take off, we need something similar. Thus far, we have what is known as Stevenson’s law which measures the number of neurons we can simultaneously record. Unfortunately, it only doubles every seven years. That is just too slow.

Fast forward to today, we have a series of startups that develop their own technologies. DARPA has already funded academic teams for non-invasive brain computer interfaces. To list a few companies that develop brain computer interfaces, in no particular order: Neuralink, Kernel, OpenBCI, Openwater, etc. Indeed there is a whole list of private companies exploring the uncharted territory of brain computers and each company has their own unique way of doing it. As capital intensity is lowered to build a brain-computer interface, there will be more startups exploring this frontier. It is too early to tell which brain computer interface will become the most successful. We must focus on succeeding as an industry together.