Long story short, a couple friends and I — our unifying trait being persistent optimism that with the best possible tools and minimal red tape, more things are possible than most people realize — have started a new company, which we've named Science.
I explain a bit about what we're up to below, and if these kinds of challenges are interesting to you, please consider coming to work with us!
It’s pretty shocking how much better life science tools have become over the last decade or so. I mean seriously, just look at this:
Multi-beam FIB-SEMs basically enable direct digitization of blocks of matter at single-digit nanometer resolution, enabling us to “scan” tissue and automatically reconstruct the dense anatomy using deep learning. We’ve gone from microarrays to spatial transcriptomics. We can directly visualize single Cas9 molecules cutting a molecule of DNA using high speed atomic force microcopy. More scalably, single-molecule localization microscopy using off-the-shelf superresolution confocals is now becoming widely available. In 2000 we had GCaMP, and now we have compartment-localizable ultrafast voltage indicators and light beads microscopy.
Consider the following interpretation of the dichotomy between science and engineering: if you can directly see and manipulate the things you’re dealing with, you’re doing engineering, and if not — you need to ask clever indirect questions to get at what you’re really interested in — you’re doing science.
Because of this intervening cleverness, science is “hard” and often advances serendipitously, but improving technology has steadily transferred more and more content from the realm of science to the realm of engineering, just as similar progress produced a conveyor belt from natural philosophy to modern science during the Enlightenment.
Probably to no great surprise, “Track 1” at Science is Neural Engineering. Specifically, our initial focus is on the cranial nerves, using some exciting ideas we think are promising that, while definitely known in the literature, don’t seem to have caught on yet. This scope reduction yields a much simpler set of problems to solve, as the general problem of BCI is tough and bordered on all sides by imposing physics barriers, while still enabling some pretty dramatic products.
To understand this plan, it’s important to see the divergence in the goals of neuroscience overall and the sub-field of brain-machine interfaces in particular.
At the highest level, neuroscience is concerned with understanding the structure and function of brains: what they are and how they work. In this pursuit, you find yourself asking questions along the lines of “how does this particular circuit in the hippocampus underlie the formation of memories?” or “how do varying levels of protein expression in the accumbens relate to addictive behaviors?” or the like. To answer these questions, you need to get inside the brain to where these phenomena are happening and there’s no way around it — peering into the brain is in some sense the whole game.
Naturally, the field of neuroscience has produced a wide variety of tools for observing things happening inside the brain. This is where brain-machine interfacing comes in: if we can record or drive neural activity, maybe we can connect the brain to external systems. But the goal of brain-machine interfacing isn’t necessarily to understand the brain, at least not any more than required to add I/O channels. Brain-machine interfacing inherited a roadmap from neuroscience, because the field came out of neuroscience, but ultimately these are different endeavors.
While it’s extraordinarily difficult to reach into the brain, all of the information that flows in or out goes through only a handful of nerves in the head and spine. These form the complete “API” of the body: if you can connect to them with single-unit resolution — still extremely difficult — you can provide exactly the same senses and motor surface that your nose, eyes, ears, hands and so on do.
Crucially, you don’t need to place anything into the parenchyma, the sensitive bulk tissue, to do this. It’s not “non-invasive,” but there are early products on this road that should have a safety profile approaching true mass-market consumer devices, and others may be able to do some truly wild things for patients with few alternatives. The future isn’t better smartphones or AR glasses: it’s making the sensorium itself directly programmable, and maybe even adding new senses entirely.
To be clear, these are aspirations. I’m not saying we have some secret device behind the curtain you should be hyped about. What I am saying is that this is technically possible to do, and we should do it. We’ve raised the money to get started — actually significantly more than what’s been publicly reported, but that’s the wrong thing to focus on — and now we have to go make things. We believe we have a clear view through to a first big defining product, built on an overlooked technical approach that hopefully we can deliver to patients in a timely manner.
As for Neuralink, I was being genuine when I said that I learned a ton there and remain significantly invested in their success. The truth is that it wasn’t my choice to leave, and there was a lot left there I still wanted to do. But as usual, Elon was right: it was time for me to go. The background is complex and, in some sense, doesn’t really matter. In what should be a surprise to no one, Elon is by far the most effective operator I’ve ever met, and the four years I had there were, in retrospect, the best education on how to turn difficult technical problems into businesses I could have possibly found. Going forward, though we both work with the brain, our goals are different: Science is not working on products that will likely be very useful for merging with AI, though that is a mission in which I hope Neuralink is successful. From a technology standpoint, what we are pursuing at Science is quite different.
For as long as I can remember, I’ve dreamt of a world of bits. Building things with atoms is expensive and difficult, but beyond that, Earth is small and intensely contested. Space is vast and open, but utterly inhospitable for humans. Maybe someday we will terraform Mars, but we will never terraform Europa. The promise of neural engineering is that, one way or another, we won’t have to.
See you in the Matrix.