Why we chose the harder constraint

There is a version of this company that would have been easier to start.

Implanted brain-computer interfaces work. That is not in dispute, and anyone who tells you otherwise is arguing with a decade of published results. Put electrodes close enough to cortex and you get a signal clean enough to do remarkable things with. The hard parts are real, but they are engineering problems on a known path.

We chose the other path, and it is worth being precise about why.

The ceiling is the point

Every implanted system inherits the constraints of the operation that installs it. It needs a neurosurgeon. It needs a theatre. It needs a hospital with a programme, and a patient healthy enough to undergo a craniotomy, and a regulatory case measured against the risk of opening someone’s skull.

Those constraints do not go away as the technology improves. A better implant is still an implant. You can make the electrodes smaller and the surgery shorter and you still cannot ship one to somebody’s house.

What changes when surgery comes out

Three things, and they compound.

Iteration speed. If improving the device means another operation, you improve it slowly and carefully and rarely. If it means shipping a new one, you improve it the way software gets improved.

Reach. Most people who have lost speech do not live near a neurosurgical BCI programme. Many live in countries that have none at all.

Data. A decoder gets better with exposure to more people, more sessions, more variation in how a brain actually behaves outside a lab.

What it rules out

Being honest about the trade: the signal is worse. Substantially. Everything between the cortex and a sensor outside the head degrades what you are trying to read, and no amount of wanting it to be otherwise changes the physics.

That is the problem we work on. We think it is the right hard problem, and we would rather spend a decade on it than build something only a handful of people will ever use.