Let's remember the whole Wikipedia

Is it possible to improve memory with an electronic chip

Roman Fishman, N+1

A month ago, the American startup Kernel announced the beginning of the development of an implant to improve the memory and learning ability of people with disorders of these functions – for example, in Alzheimer's disease. The scientific director of the project was Theodore Berger from the University of Southern California, who proposed a model of activation of hippocampal neurons in the process of perception and memorization of information, and also showed the ability to purposefully stimulate such activity in the hippocampus of mice and even primates.

At Kernel, Berger will create "hippocampal prostheses" for people who need them. We talked about his work with a neurophysiologist from the Center for Neuroengineering at Duke University, Mikhail Lebedev.

– Before we deal with the mechanism of a possible "memory prosthesis", let's deal with the memory itself. There are neurons, there are synapses, contacts between them, and the more often synapses excite, the easier they work afterwards. This plasticity of synapses is, as it were, memory. But this is not the end of the picture. What do we know today about the mechanism of encoding memory at a higher level than the synapse?

– The story of synapses was "invented" by Donald Hebb, and he also "invented" a higher level, called a neural ensemble, a neural network. According to Hebb, synapse amplification occurs only if both one of the neurons that receives a signal from the synapse and the axon, the nerve fiber of the "incoming" neuron, are activated simultaneously. Actually, this is how memory is formed: if a neuron "shows interest" and at this moment receives a signal, then the communication node is strengthened. It's like an entry is being made in a phone book. If the neuron subsequently receives a signal through this channel, it will respond with a higher probability, since it "remembers the previous conversation."

You can recall Pavlov's experiments with dogs, light bulbs and bells, almost the same thing happens there. A light bulb or a bell produces a signal that comes to neurons excited by food. These connections are strengthened, and the next time the call is enough for the dog to remember about food, and her mouth watered.

Pavlov called this strengthening of connections a conditioned reflex, meaning a fairly simple neural network. The Hebbian ensemble is much more complicated – it is already a neural network that includes a large number of nodes and connections. Such a network uses its own principles of information encoding (still unknown to science) and generates different types of activity (for example, neural oscillations). In addition, the same network can store many "records" in its very configuration. These records are quite resistant to damage to individual elements – a micro-stroke can kill a certain number of neurons, but the memory will remain.

This kind of memory is often compared to a hologram, bearing in mind that each small part of the brain stores the entire record, and due to a large number of sections, only the detail of these records is improved – that is, almost like in a hologram. But the elementary basis of such a holographic memory is still neurons and synapses.

It is worth adding that there are many different types of synapses in the brain: there are chemical synapses that use neurotransmitters for transmission, but there are also electrical ones in which ions are transmitted between neurons – just as it happens when the heart is excited. In addition, there are a great many neurons, they can be excitatory and inhibitory. We should not forget about the glial (auxiliary) cells of the nervous tissue, which can also play a role in the formation of memory.

To complete the picture, we add that memory can be conscious, as a memory of last night, and unconscious, as a motor cycling program, short-term (it is also "working") and long-term, declarative (memory of concepts) and episodic (about events). In general, it is clear that nothing is clear with memory. The distributed memory representation code is unclear. It is unclear how the memory read request is formed, how it is returned in response to the request and in what form.

– What is called an "engram"?

– Not really. In the description of the German zoologist Richard Zemon, who introduced this term, an engram is a "record" that remains in excitable tissue after an exciting stimulus passes through it. Zemon also coined the term "ecphoria", meaning the process of reading memory from an engram. In general, these ideas gained popularity after the publication of Carl Lashley's work "In Search of an Engram".

The story boils down to the fact that Lashley was looking for an engram, but he never found it: he destroyed various parts of the brain of experimental animals, after which memory deteriorated, but did not disappear altogether. Lashley came to the conclusion that memory is not stored locally, it is distributed throughout the brain. Actually, approximately the same ideas exist now. Although, of course, we have made significant progress in understanding the biophysics and physiology of synaptic changes, the main mystery remains: how is distributed code built and how is it read? We don't know that.

– What happens when memorizing? What role does Pavlovian reinforcement play in this?

– Attention mechanisms are important for memory formation: only what we pay attention to is remembered. At every moment of time, there is a context that is determined by both external factors and the state of the brain. For example, I'm in a museum and looking at paintings; I'm in a sports club and doing sports; I'm in a library and reading a book. Depending on the context, we pay attention to certain external information coming to us through the senses.

As a result, neural networks are excited, which can be maintained for a short time due to reverberation – signal transmission from neuron to neuron. If this information is significant, it enters the hippocampus, which helps translate it into long-term memory. But although the hippocampus promotes memorization, in the end, long-term memory is not stored in it, but distributed throughout the brain.

As we memorize, most of the information is lost; we usually remember the most essential, or something that attracted our attention and was remembered. Well, the most essential and memorable thing for a biological organism is food. That is why Pavlovian reinforcement is so strong, and everything connected with it is strong: taste, smell. However, in addition to Pavlovian reinforcement, there are also other important factors, danger signals and the desire to reproduce, which also contribute to memorization. These factors can be described by the general concept of motivation, in the formation of which subcortical nuclei play an important role. It is motivation and emotions that determine the brightness of memory.

– What determines the neural composition of the engram? What approaches are used to install it?

– First of all, there are neurophysiological approaches that allow you to register the activity of neurons during memorization and memory reproduction. Pharmacology helps to understand the biochemical mechanisms of these processes. In addition, you can look for traces of memory under a microscope, examining changes in neurons and synapses. Sleep research is of great help: it is believed that it is during sleep that consolidation, consolidation of memory occurs. To this end, the brain "plays" the daily recordings again and translates them into long-term memory.

We can recall a rather curious study performed 40 years ago: worms were trained to move along a certain route, and then allowed to mince and feed other untrained worms. It seemed to the researchers that such cannibalism allows the transfer of memory and "memory molecules". However, later it turned out that it was a mistake.

– Based on our knowledge of memory, which approach – purely theoretically – could lead to the creation of "electronic memory stimulators"? Let's say we are able to make any device. What exactly should it do?

– As already mentioned, the hippocampus plays an important role in memorization. Its structure is well studied, there is a general understanding of how exactly it performs memorization. For simplicity, let's assume that there is a department A and a department B in the hippocampus, and the transfer of information from A to B is important for memorization. Let's say we studied the relationship between the activity of A and the answers of B and trained a mathematical algorithm to calculate these answers. Now we can do without A by simply electrically stimulating B and triggering the responses predicted by the algorithm in it. This is roughly how an electrical memory stimulator developed by Theodor Berger works.

By the way, Berger is not the first to activate memory by electrical stimulation. Several groups have noted that stimulation through electrodes implanted in people in areas of the brain close to the hippocampus either evokes memories or improves memory. Although I will add that the electrodes in these studies were implanted not to improve memory, but to treat various neurological diseases.

– The engram forms a more or less stable ensemble of nerve cells whose activity is associated with memorization. But we cannot predict in advance which neurons should get into this ensemble. Isn't there a chicken and egg problem here? To consolidate the connections in the engram, we need to stimulate the neurons of the engram, but what kind of neurons will enter it, we do not know until the engram itself has formed without our help?

– We don't know this only because we have a bad idea of how memory occurs in the brain. But the brain itself probably "knows" in advance which neurons will be in which ensemble. At least, it is known that it has strict and orderly maps of the body, sensory and motor. It is more difficult to understand abstract memory maps, but this does not mean that they do not exist.

– How exactly does the device that Berger's team is working on work? What is it already capable of?

– The Berger device itself is not too difficult: several electrodes in the hippocampus area A, several electrodes in area B. First, the rat remembers something; activity A and B are recorded, a mathematical algorithm is trained to translate activity A into activity B. Then you can cut the connections between A and B, but stimulate B and form the necessary memory.

The scheme of operation of the device

– Berger reports that the system is based on their MIMO mathematical model, which predicts the activity of hippocampal neurons. How does it work?

– MIMO is short for Multiple Input, Multiple Output, "Many inputs, many outputs". It means that several electrodes are used in area A, and several in B. Based on signals A, signals B are output. The mathematical algorithm uses a nonlinear Volterra model, which is suitable for such operations.

– How easy is it to transfer the very artificial experimental conditions in which Berger and his team worked into real life? Is it even possible?

– Something is possible, and something is impossible. It is impossible, for example, to record the contents of a book or movie into memory by means of electrical stimulation. We just don't know how to encode such information. But as for improving memory in patients, this is undoubtedly possible. By modulating the natural process of memorization with electrical stimulation, it is possible to reduce the influence of pathological processes that interfere with the work of memory. Pathological processes in different areas of the brain, as a rule, manifest themselves in the form of flashes of activity similar to epileptic ones. Such outbursts can be suppressed by electrical stimulation.

Similar techniques are used in other cases, for example, for the treatment of Parkinson's disease; only here it is not the hippocampus that is stimulated, but the basal ganglia. Such artificial ordering of their activity makes it possible to cope with the strongest tremor from which such patients suffer.

Unlike Parkinsonism, it is not necessary to place electrodes deep into the brain to improve memory. It can also be stimulated from the surface of the head, transcranially. However, all these developments are quite new and the results are still ambiguous. For the same purpose, pharmacological drugs, nootropics, which have become so popular with students, are also used.

– How universal are the memory mechanisms that are studied at the level of the hippocampus and its parts? And if some approaches work for him, will it be possible to transfer them to other areas of the brain?

– I will note that, in fact, there are no firmly established methods for the hippocampus yet, just as there is no truly powerful memory prosthesis for the hippocampus. Therefore, the problem of "transferring or not transferring" is not worth it yet. The general principles are undoubtedly portable, since any part of the brain is an electric machine, although we are still very far from prosthetics of entire departments.

Prosthetics of peripheral parts of the nervous system is much easier: for example, cochlear implants have proven to be surprisingly effective, they are used by hundreds of thousands of people with hearing problems. This is because the coding of signals in peripheral nerves is quite simple, it is enough to stimulate them with a suitable frequency. But with the higher departments, everything is more complicated. There are billions of neurons operating here, performing subtle operations. It would be naive to hope that they will be able to control filigree with the help of inserted wires-electrodes.

So currently, the best way to improve memory is to use electronic reference books – such as Wikipedia. In ancient times, it was considered unworthy to rely on records instead of one's own memory, this is mentioned in Plato's dialogues ("... memory will be deprived of exercise: they will remember from the outside, trusting the letter, by extraneous signs, and not from the inside, by themselves" ["Fedr", A.N. Egunov lane] – N+1). But until we have effective memory prostheses, we will rely on the Internet. And they will appear in the future – we will remember the whole Wikipedia.

Portal "Eternal youth" http://vechnayamolodost.ru  30.09.2016