Speaker unverified: commitment

Transcript context

…Yeah, better than before. That’s a story on its own of what took the BCI team to recover that performance. It was actually mostly on the signal processing. And so as I mentioned, we were looking at these spike outputs from our electrodes, and what happened is that four weeks into the surgery we noticed that the threads have solely come out of the brain. And the way in which we noticed this at first obviously is that, well, I think Noland was the first to notice, that his performance was degrading. And I think at the time we were also trying to do a bunch of different experimentation, different algorithms, different UI, UX. So it was expected that there will be variability in the performance, but we did see a steady decline. And then also the way in which we measure the health of the electrodes or whether they’re in the brain or not, is by measuring impedance of the electrode. So we look at the interfacial, the Randles circuit they say, the capacitance and the resistance between the electrode surface and the medium. And if that changes in some dramatic ways, we have some indication. Or if you’re not seeing spikes on those channels, you have some indications that something’s happening there. And what we noticed is that looking at those impedance plot and spike rate plots, and also because we have those electrodes recording along the depth, you are seeing some sort of movement that indicated that threads were being pulled out. And that obviously will have an implication on the model side because if the number of inputs that are going into the model is changing because you have less of them, that model needs to get updated. But there were still signals, and as I mentioned, similar to how even when you place the signals on the surface of the brain or farther away, like outside the skull, you still see some useful signals. What we started looking at is not just the spike occurrence through this BOSS algorithm that I mentioned, but we started looking at just the power of the frequency band that is interesting for Noland to be able to modulate. Once we changed the algorithm for the implant to not just give you the BOSS output, but also these spike band power output, that helped us refine the model with a new set of inputs. And that was the thing that really ultimately gave us the performance back. And obviously the thing that we want ultimately and the thing that we are working towards, is figuring out ways in which we can keep those threads intact for as long as possible so that we have many more channels going into the model. That’s by far the number one priority that the team is currently embarking on to understand how to prevent that from happening. ong as possible so that we have many more channels going into the model. That’s by far the number one priority that the team is currently embarking on to understand how to prevent that from happening. The thing that I will say also is that, as I mentioned, this is the first time ever that we’re putting these threads in the human brain. And a human brain, just for size reference, is 10 times that of the monkey brain or the sheep brain. And it’s just a very, very different environment. It moves a lot more. It’s actually moved a lot more than we expected when we did Noland’s surgery. And it’s just a very, very different environment than what we’re used to. And this is why we do clinical trial, we want to uncover some of these issues and failure modes earlier than later. So in many ways, it’s provided us with this enormous amount of data and information to be able to solve this. And this is something that Neuralink is extremely good at, once we have set of clear objective and engineering problem, we have enormous amount of talents across many, many disciplines to be able to come together and fix the problem very, very quickly. But it sounds like one of the fascinating challenges here is for the system on the decoding side to be adaptable across different timescales. So whether it’s movement of threads or different aspects of signal drift, sort of on the software or the human brain, something changing, like Noland talks about cursor drift, they could be corrected. And there’s a whole UX challenge to how to do that. So it sounds like adaptability is a fundamental property that has to be engineered in.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.