Neuralink has caused a sensation on the technological side of neural implants, but has not yet shown their practical use (unless you are a fan of Pong).
This week, the academic community provided a rather impressive example of the promise of neural implants.
With the help of a neural implant, a paralyzed individual was able to type around 90 characters per minute simply by imagining writing those characters by hand.
It's a crazy leap forward.
A neural implant with immediate feedback
Previous attempts to provide typing skills to paralyzed people via implants involved virtual keyboards and cursors to be moved with the mind.
An effective but slow process, which requires the subject to follow the progress of the cursor and determine when to perform the equivalent of pressing a key. Not to mention the time it takes to learn how to control the system.
But there are other possible ways to get the words out of our heads. Somewhere in our writing thought process, we form the intention to use a specific character, and a neural implant that tracks this intention can potentially work. A process still not very clear.
But, but ...
Downstream of this intention, a decision is transmitted to the motor cortex, where it is translated into actions.
Again, there is a phase of intent, where the motor cortex determines who will form the letter (typing or writing, for example), and which results in the specific muscle movements required to perform the action.
Instead, these processes are much better understood, and on these the research team has targeted for its new work.
Neural implant research (here it is)
Specifically, the researchers placed two implants in the premotor cortex of a paralyzed person. This area is thought to be involved in the formation of intentions to perform the movements.
With the implants in the right place, the researchers asked the participant to imagine writing letters on a page and recorded neural activity as he did so.
What came out
Overall, of the approximately 200 electrodes in the participant's premotor cortex, not all were useful for letter writing. But for those that were, the authors performed a principal component analysis. By converting these recordings into a two-dimensional texture, the team noted that physically similar characters, eg, eg boh, ner, formed homogeneous neural record clusters.
Overall, the researchers found they could decipher the appropriate character with an accuracy of just over 94%, but the system required relatively slow analysis after recording the neural data.
To make things work in real time, the researchers trained a neural network to estimate the probability of a signal corresponding to each letter.
The results of the neural implant for writing with thought
Despite working with a relatively small amount of data (only 242 sentences), the system worked very well.
The delay between the thought and the appearance of a character on the screen was about half a second, and the participant was able to produce about 90 characters per minute, outperforming the previous record (25 characters per minute). The crude error rate of the neural implant it was only about 5%. With the addition of an automatic corrector it was reduced to 1%.
The tests were all performed with preset sentences. At a later stage, the researchers asked the participant to type in free-form responses. Here, the speed dropped a bit (to 75 characters per minute) and the errors increased by up to 2% after the autocorrect, but the system still worked like a charm.
Neural implant, we are at year zero
As the researchers themselves have stated, this "is not yet a complete and clinically valid system". For starters, it has only been used on a single individual - we still have no idea how well it works for others. In addition, the alphabet used did not contain digits and punctuation.
That said, the system showed a tremendous speed increase over previous neural implant guided systems, with excellent accuracy.