What are neurocommunicators and should you be afraid of artificial intelligence? Will we be able to read other people's thoughts in the future and how to make an analogue of Google Glass glasses using neural interfaces? What technologies did Stephen Hawking use to control his body and how to print text on a computer with the power of thought? The famous biologist and psychophysiologist Alexander Kaplan told us about this.
Alexander Yakovlevich Kaplan
Doctor of Biological Sciences, Psychophysiologist, Professor of the Department of Human and Animal Physiology, Head of the Laboratory of Neurophysiology and Neurocomputer Interfaces, Faculty of Biology, Moscow State University. Lomonosov. Laureate of the State Prize of the Government of the Russian Federation, developed the first neurocommunicators in Russia.
Alexander Yakovlevich, what are neurocommunicators?
- There is such a method, well known in medicine, is available in almost every clinic, called electroencephalography. It allows you to record the electrical activity of the brain (EEG) directly from the scalp. It is very convenient: put electrodes on your head, connect them to a computer through a potential amplifier and observe curves on the screen that reflect the life of the brain. This method is used to diagnose various kinds of diseases, epilepsy, and tumors. We have been doing this for a long time, but at some point it became interesting, why not enter the electrical signals of the brain into the computer to control the keyboard?
It only remains to learn to recognize in the EEG the moments when a person intends to type this or that letter, and immediately transmit this command to the keyboard. You will get a neurocommunicator: letters are typed by mental efforts without the help of a voice or hands, directly from the brain! The use of this technology also suggests itself: to establish communication with patients who are speechless and motionless, for example, after a stroke. Such a neurocommunicator will help not only to type letters, but also to press command buttons on the keyboard: call a nurse, turn on / off the TV, etc.
In our laboratory, we have brought the reliability of such a communicator to 95%, that is, typing with mental efforts, a person makes only 5% of errors. True, the word "decryption" hides a rather ingenious algorithm. The user of the neurocommunicator does not just sit at the computer - he is shown a matrix on the screen, in each cell of which a letter or some symbol is drawn.
All cells are in random order, but they are highlighted very quickly sequentially - 5-6 times per second. At the same time, electrodes fixed on the back of the head record the EEG of the visual areas of the brain. Here in this recording, by computer methods, you can find the reactions to the highlighting of each letter. The reactions are all different, but not because the letters are different, it's just that the brain reacts differently at every moment, even to the same signal. But the physiological trick is that if a person focuses his attention on a letter, then the reaction to the illumination of this particular letter will differ from the reaction to all other letters. Thus, the neurocommunicator detects the focus of a person's attention to specific characters on the computer screen and sends them to print. No mysticism!
What you are talking about would be very helpful, probably, to such a famous scientist as Stephen Hawking …
- Stephen Hawking has been using a communicator for a long time, however, this communicator decoded signals not from the brain, but from some muscle allocated for this purpose. Straining the signal muscle to a greater or lesser extent, Hawking used such a communicator to type the texts of all his books. In any case, a person is much more accustomed to controlling muscles than brain reactions. And the process of decoding a muscle signal is simpler than a brain one. Therefore, the speed of the muscle communicator is higher than that of the neurocommunicator, in which the choice of one letter takes 5-6 seconds.
True, according to unofficial information, the progressive disease has deprived Hawking of the activity of the last muscle. It is known that Hawking later received a new communicator, but what its principle of operation is, we do not know. I asked my American colleagues - they say that this is a commercial secret. Despite the insufficient performance, neurocommunicators are the only way to communicate in patients who are deprived of speech and movement. For about a year and a half, we have been adapting this development to work with real post-stroke patients at the Pirogov First City Hospital.
While we are at the development stage - since we are talking about medicine - we need reliable clinical trials in many areas. On healthy people, we have worked out our method very well.
Is there any hope that it will be possible to make neurocommunicators faster?
- Everything eventually becomes faster and more powerful. But there are things that cannot be overcome, such as the speed of light. There is also a ceiling here. In the whole world practice of using neurocommunicators, there is a delay of 2-6 seconds, depending on how many commands need to be processed. If you need only six commands, for example, to control a wheelchair, then the choice of a specific command from six: "left", right "," forward ", etc. will take 1-2 seconds. But if on the screen the entire alphabet, and even commas and periods, that is, 36 characters, then processing will take 5-6 seconds. And so far, it has been impossible to overcome this threshold for twenty years already.
In connection with what?
- It's very simple: to decode a signal, you always need a piece of record equal to at least the length of the message being decoded. Otherwise, there will simply be nothing to decipher. You can, of course, take a very short section of the encephalogram and just try to guess about the whole message, that is, about the intended letter. But on such fortune-telling, the probability of error appears. If you use a recording fragment of 1 second, you get about 50-60% errors.
If high reliability is needed - and we achieve 95% reliability - then it still takes 5-6 seconds. Meanwhile, despite the relative simplicity of the entire technological chain of the neurocommunicator, probably only we were able to achieve 95% accuracy of its work in our country. It's just that we have been doing this business for a very long time and have polished many different nuances. But many developers could already achieve 70% accuracy.
While we are talking about more or less simple commands. Is there a mind-reading perspective?
- Scientists argue about this. But most of my colleagues, including myself, are of the opinion that it is theoretically impossible to read thoughts using instrumental methods. If only because thought is the cumulative result of the activity of many parts of the brain. And no matter how many electrodes we put on the head, we will still cover only a small part of what happens in the head when a thought is born. We cannot connect to all nerve cells at once - there are too many of them, almost a hundred billion. Even if you come up with some kind of ultra-thin wire, you will probably need a whole wagon of such wires. Because one hundred billion connections is a lot.
Today, only 100-200 electrodes are used in the development of neurointerfaces. And even if there are one hundred thousand of these electrodes, this is far from 100 billion.In addition, in each pair of nerve cells, obviously, there is some kind of communication code. How to unravel these codes if there are three orders of magnitude more such pairs in the human brain than there are nerve cells themselves.
Are there any other areas, besides medicine, where such technologies could be applied?
- So far, the main area of application of neurointerfaces is still medicine. Both here and abroad. Primarily to help paralyzed patients. And here, in addition to communication, there is one more task - the restoration of motor function, that is, neurorehabilitation. One of the main ways in this matter is training the impaired function: fingers move poorly - you need to work them as much as possible. What if they don't move at all? The patient has an intention to move, but not. It turns out a vicious circle: training is needed to restore movement, which is impossible, since the hand is paralyzed.
But, as we know, the intention to move can be deciphered by the EEG using the same neurointerface, but with a slightly different algorithm than that of the neurocommunicator. The decoded intention can be immediately transformed into a command for a special design with motors, an exoskeleton, which is attached to the hand and sets it in motion mechanically. Thus, the patient's intention to move is translated into movement. Training begins! Therefore, there is a chance that after some time the natural movement will begin to recover. The use of neurointerfaces in rehabilitation is already a serious trend in medicine. Several laboratories are engaged in these technologies in Russia, in addition to ours, for example, the laboratory of Professor A. A. Frolov at the Institute of Neurophysiology and Higher Nervous Activity.
We talked about medicine. The use of neural interfaces in other areas is a difficult question. Because this technology is not magic and has its limitations. Of course, I would like to control, for example, an airplane with the power of thought. Or at least a car. Alas, this is not possible yet, because, as we said, neural interfaces are not very fast. On a busy highway, this is, as you know, an unaffordable luxury.
Now the biggest business with neurointerfaces is the use of the Neurointerface brand in all sorts of toys equipped with a couple of electrodes, some kind of amplifier and a microcontroller to control this very toy: furry ears on the rim, a ball in a vertical wind tunnel, etc. All of this, of course, works with minimal reliability, but it is quite suitable for a fun party.
There is another area of possible application of neural interfaces: control of robots (not those autonomous robots that work according to a written program, because you cannot write a program for all occasions). This is especially true for robots working in hazardous conditions, for example, during demining or in an area of radiation pollution. There needs additional control from the side of the person. And it is here that technologies based on neurointerfaces are possible. And it is already starting to be used today.
There is also an idea to use neurocommunicators as a tool to complement the brain's activity. By analogy with glasses that augment reality. Only in the case of neurointerfaces, not reality is supplemented, but the fulfillment of intentions: a person will be able to flip through, for example, Internet pages with one intention, that is, even before the birth of the thought itself, in search of yet unconscious but necessary information.
What do you think about artificial intelligence - are the fears described in the movie "Terminator" justified? Many say that one should not be afraid of this, because no machine has its own desires, which means that it will not "rebel" against humanity
- All technologies that a person invented are dangerous. When a man invented a stone ax, it was already a dangerous technology.A car is a means of increased danger. Therefore, if a person comes up with artificial intelligence, he will also carry a certain threat. But, as we can see, with all this, a person manages to make sure that these dangers are minimized. Therefore, I do not see any problem in being able to limit dangerous tendencies in the behavior of robots and artificial intelligence.
Not to mention the fact that a good half of scientists consider the creation of intelligence, similar to human, simply impossible. There are a lot of reasons for this. One of them is that a person is a natural being and his brain was not loaded with programs of behavior, these programs are formed by themselves as the organism grows up. Therefore, human intelligence is much richer than any artificial fake. Artificial intelligence will become equal to that of a human, if only it grows next to a person, has senses, a heart, etc. Is this possible?
As for the statement that the machine has no will or desires of its own, here I would not agree with you right away. After all, what does desire mean in the first approximation is the actualization of some need: you are thirsty and ask for a glass of water - this is a desire, the computer has a lack of charging - is it not a desire? Instead of the inscription that appears on the laptop: "Low battery", you can display the inscription: "I want electricity." What is the difference?
Tell us about the experiments carried out in the West with neurointerfaces on animals
- The fact is that all this was originally done on animals, mainly on rats and monkeys. In working with animals, the experimenter has more options, for example, to place electrodes directly into the brain itself. Such experiments are impossible with a person, even if it is a very sick patient - such things can be done only for medical reasons. Therefore, monkeys with interfaces based on implanted electrodes in the brain achieve much more.
For example, they can easily operate the cart they are sitting on. The neural interfaces are connected directly to the monkey's brain cells, which makes it possible to decipher its intentions faster and easier. When the electrodes are on the surface of the head, we get a mixture of different electrical signals, almost noise, so it is more difficult to recognize them. Experiments on animals make it possible to find out possible prospects in the development of neurointerfaces.
Moreover, in the USA there are already laboratories in which the implantation of electrodes into the brain of patients, developed in animals, is practiced, of course, strictly for medical reasons. Paralyzed patients with implanted electrodes can literally “use the power of thought” to control the manipulator in order to feed themselves a container with a drink or a bar of chocolate. Together with our American colleagues from the University of Southern California, we are now trying to create a similar technology in Russia.