It's silly that it's hidden, because 18 WPM for a quadriplegic still seems extremely impressive to me, given that it's just working from brain electrodes and not some muscle the person can control. Yes, the headline is silly, because by hiding it it ended up causing the snark about 18 WPM being slow.
I'm surprised, though, that the most efficient way to do this is still to have the person imagine physically drawing the letters by hand. I know motor neurons are probably our most reliable output, but I would still think that, with all the advances in training from noisy data in the past decade, that training what the thought of "A", "B" etc look like in the head would be doable.
Or even what the thought of hearing or saying "A", "B" etc looks like. The auditory cortex is activated when we imagine sounds. Or, if they wanted to stick to motor neurons, could they have the person imagine saying the letters with their mouth?
I'm sure they've thought about this stuff and it's harder than it seems, of course. But I would just predict that brain-computer interfaces 20 years from now won't involve imagining using your hand to write letters.
Great point! The electrode arrays used in these studies are implanted in the participants' motor cortex in an area called the "hand knob". Neurons here are highly modulated during motor output, not only for the hand, but also for the rest of the body. This area is close to the output though, and abstract higher-level features like the "idea of an A" isn't likely to be decodable from neurons there. Future research is looking at exactly this, but will likely require recording in a different area.
For context, I did my PhD in the lab that did the work in this article.
When typing in passwords on a phone I have to visualize the keystrokes by thinking about the finger movements and concentrating on my forearms without really moving my fingers. Generally once fast and once slow to think about the actual activations.
> 18 WPM for a quadriplegic still seems extremely impressive to me
Honest question: how so? We should expect a direct neural interface to far exceed the speed of any manual input device, especially after 40-50 years of research.
Because I'm basing it on where things were 5-10 years ago (almost nothing like it actually working yet), not some hypothetical place where the research could be after 50 years.
GPT-3 is also very impressive to me, even though 30 years ago I thought we'd have Hal by now.
Some problems just turn out to be way harder than anyone anticipated, and so when they make advances I'm impressed.
Because manual input devices aren't available to quadriplegics and earlier solutions (e.g. sip/puff controller used to select single letters from an on-screen keyboard) are relatively very slow.
I'm surprised, though, that the most efficient way to do this is still to have the person imagine physically drawing the letters by hand. I know motor neurons are probably our most reliable output, but I would still think that, with all the advances in training from noisy data in the past decade, that training what the thought of "A", "B" etc look like in the head would be doable.
Or even what the thought of hearing or saying "A", "B" etc looks like. The auditory cortex is activated when we imagine sounds. Or, if they wanted to stick to motor neurons, could they have the person imagine saying the letters with their mouth?
I'm sure they've thought about this stuff and it's harder than it seems, of course. But I would just predict that brain-computer interfaces 20 years from now won't involve imagining using your hand to write letters.