top of page
  • samrodriques

Some considerations on brain-computer interfaces

I was speaking recently to a friend who is interested in starting a brain-computer interfacing company about the challenges and opportunities associated with BCIs. Here are some basic considerations:

Basic considerations:

Consideration 1: You have to compete with the peripheral nervous system. Any central brain interface that will achieve widespread adoption must be at least some reasonable fraction as good as hands, or eyes. This is a high bar.

Moreover, there is a lot we could do with BCIs that leverage the peripheral nervous system, that we are not currently doing. For input: we could be leveraging touch far more than we do. Why not have a watch that communicates through braille rather than through a display? Why not have a device that converts images in real-time into embossed patterns on a 128 x 128 or 256 x 256 grid? (I looked for this extensively, it seems like it should exist for blind people, but I didn’t find it. The learning curve would be significant but it seems like low-hanging fruit.) For output: I suspect we could be using EMG much more than we do already. Why hasn’t there been any serious adoption in this area already? I suspect a combination of steep learning curve and weak demand.

Central BCIs will not overcome the steep learning curve problem. They can overcome the weak demand problem if they vastly improve a human’s capability relative to what that human can do with a peripheral interface. So the fundamental question is: how much better can a central interface be than a peripheral interface? This is a basic neuroscience question that is currently unknown. It’s possible that it could be much better, if we could directly write memories into synapses. It’s also possible that AI alignment will solve most of our computer interfacing challenges.

Consideration 2: We already know that brain interfaces are very hard to learn after the critical period closes. Babies who are born with cataracts, who do not have the cataracts removed prior to the closure of the critical period for vision, never regain their sight at the level of their cataract-free peers. This is an extremely instructive point for brain interfaces. The hardware is perfect: the eye functions, the wiring is there, and yet they still do not achieve anything like natural vision. For this reason and others, I (and many others) suspect that the difficulty of learning to use a brain-computer interface is a plasticity problem, not a hardware problem. A sufficiently plastic brain will learn to use any interface, but a non-plastic brain will struggle to use even the best-engineered interfaces. (Moreover, the interfaces do not necessarily need to be high bandwidth. How many bits do you need in order to control a robotic arm? Maybe a kilobit per second? It’s not high throughput. The issue is much more an issue of plasticity than it is an issue of hardware.)

There will always be demand for brain interfaces among disabled patients, and I expect many more impressive demos from Neuralink. However, for central brain interfaces to achieve widespread adoption, they either need to be implanted before the critical period (i.e. in children), or we have to figure out how to solve the plasticity problem. I suspect implantation of BCIs in children is probably the way things will go in the long run.

Consideration 3: No one wants a hole in their head. Seriously. We need to figure out how to get into the brain non-invasively. The vasculature is a good option. I am also a big fan of the cribriform plate (see below), which is a piece of the skull that already has many millimeter-scale holes in it for olfactory nerves. I also think it should be possible to drill thousands or hundreds of thousands of microscopic holes in the skull ultra-rapidly for interfacing.

Considering this, what are the opportunities?

Opportunity 1: BCIs in children, before the critical period closes. This is hard for obvious reasons, but it is probably the way things will go in the long run.

Opportunity 2: Fix the plasticity problem. Figure out how to reopen the critical periods. This is an active area of research. Note, however, that there is a basic question here about whether the critical period can be reopened without destroying memories.

Opportunity 3: Reduce invasiveness. If we could install central brain interfaces in a much less invasive way, it would dramatically reduce the barriers in the field. However, it does not solve problems 1 and 2 above.

Wild speculation: Teleportation

I suspect that the path to true widespread BCIs will be that we begin to implant BCIs in children who are missing limbs or other body parts. Those children will learn to use their robotic limbs or eyes with as much proficiency as a fully able-bodied human can use their biological limbs; but those children will have an advantage, which is that they will be able to switch their feed at will. A child with a prosthetic eye could switch their feed to see through some camera halfway across the world. Ultimately, the advantages gained that way will cause parents to clamor for BCIs in healthy children. And this is the way we will invent teleportation: when you learn to use prosthetic limbs from birth, you will be able to teleport to other bodies by just switching your feed.

221 views0 comments

Recent Posts

See All

Protocols are long (and full of horrors)

I think the reason automating science is hard is that protocols often have 20+ steps; but each individual step will fail or require modification in 5% or 10% of cases. So the probability you get all t


bottom of page