Neuralink Elon Musk over the past couple of years has achieved very noticeable success in the development of implants and their implementation in the brain. But, as they say, not by the Mask alone. Now it has become known about a neuroimplant developed by a group of scientists, and the capabilities of this system are impressive.
This week, a group of scientists published an article about a new development that allows a paralyzed person to type text on a PC at a speed of about 90 characters per minute. To do this, the volunteer just needs to imagine writing a similar text with a pen on paper.
Imagination can do anything
Previous attempts to "teach" a paralyzed person to type on a PC using a neuroimplant have been relatively successful. But almost all such projects were a virtual keyboard on the screen, at the buttons of which the patient had to look, or individual symbols highlighted on the same screen, which also had to be selected using a thought-controlled cursor. It all really worked and works, but the typing process is extremely slow. It is simply impossible to compare it with the speed of typing on a keyboard by an ordinary person - the difference is huge. In addition, in order to learn how to work with such a system, the patient has to spend a lot of time.
As it turns out, there is a more efficient alternative. When we write text on paper, we write down every letter we think about, at least for a fraction of a second. This short thought provides an electrical signal that can be identified. True, only a signal can be identified, which is transmitted to the region of the brain responsible for controlling the body - in particular, with the hand and fingers, with which we hold the pen and write.
The area of the brain mentioned above is called the motor or motor cortex. It is believed to be involved in controlling body movements. After installing implants in this area, the scientists asked a volunteer to imagine how he writes letters on a page. They then analyzed the brain activity during the thought process.
A little more details
In total, about 200 electrodes were introduced into the motor cortex. Not all of them turned out to be useful or hit right on target. But there were enough of those who were in the right place, which allowed scientists to record the electrical activity of the desired area of the brain. The signals were converted into a regular chart, dividing it into patterns for each specific letter. As it turned out, letters similar to each other stimulate the generation of signals with similar configurations. These are, for example, the letters p and b or h, n and r.
The researchers asked the volunteers to imagine and add punctuation marks as they “write” the text. The ">" sign was asked to be used in those places where there should be spaces between words and symbols.
As a result, the electrical signals received from the participants in the experiment were deciphered with an accuracy of 94%. The system could already be used for work, but it was not particularly fast. Therefore, scientists have trained a recurrent neural network to assess the likelihood of a signal corresponding to a certain letter.
Despite the fact that the training base was small (only about 242 sentences imagined by the participants), the system worked very well. It took only 0.5 seconds to play the symbol after the participant thought about it. Of course, this is a significant delay, but still, this method is much faster than conventional methods. As a result, as mentioned above, the participants were able to type at a speed of 90 characters per minute. The best result of other similar research is 25 characters per minute. Errors are only 5%, which is not much more than people who type with their fingers. After the introduction of the auto-correction system, the error rate dropped to 1%.
True, this result was shown for a previously prepared text. When test participants were asked to write something in free form, the speed dropped to 75 characters per minute, and the accuracy, with the automatic correction system turned on, to 2%. But this is also an excellent result.
It's not even an alpha version
Yes, we are talking about a system that is clearly not even an alpha version, although the results are excellent. However, the texts that the volunteers recruited did not contain numbers, capital letters, and most punctuation marks, except for the main ones. Also, the system had to be calibrated at least once a week. The settings were regularly lost - perhaps due to the accumulation of scar tissue or the movement of the electrodes, albeit minimal.
Anyway, as a prototype, the system showed just excellent results - both in terms of typing speed and accuracy.
Hopefully, such systems will eventually be refined and made available to patients with motor problems.
