Neurobiology and artificial intelligence: part one - educational program

It just so happened that I have been reading Habr and especially the section on artificial intelligence for a long time, but so far I have not dared to contribute to the general level of entropy.

Lively discussions in the comments show a keen interest in the topic and at the same time a wide variety of points of view, opinions and levels of training. After reviewing the history of publications, I somehow did not find an important starting point for many arguments, namely, any description of the signal transmission mechanisms in the brain. Those who write about neural networks and progress in computer models of intelligence usually casually mention synapses and mediators (which is quite enough for their purposes), while those who try to understand the nature of natural intelligence are mainly reasoned in philosophical categories. As a result, many comments contain references to popular speculations and myths that are not confirmed in modern studies.

In this article, I will try to summarize the answers to the following questions:
- What is a neuron, how is it arranged and works?
- What happens in synapses when neurons communicate with each other?

And in the following (s):
- how are intelligence and consciousness related to the activity of neurons? (here it is about how information is processed by the brain, neuroplasticity, quantum theory of consciousness, sleep, etc.)

So what is a neuron, how is it built and working?

Fig. 1. Various forms of neurons. Hereinafter, the pictures are taken from “Neuroscience”, 3d edition, Dale Purves et al .

Firstly, it is worth mentioning that in the human body there is not one kind of neurons, however, for all their differences (Fig. 1), they have much in common, including from the point of view of mechanisms for ensuring their direct functionality. Secondly, in addition to neurons, there are many auxiliary cells in the brain - neuroglia cells, or simply glia. These cells are on average 3 times more than neurons, and they provide neurons with nutrition, energy and essential substances. Recently, more and more attention has been paid to them due to the fact that their influence on the work of neurons (for example [1]) in the field of learning and memory has been shown. Here it would be worth noting that being a support cell, neuroglia cannot but affect the functioning of neurons, because a hungry neuron cannot work just like a well-fed one, but do not overestimate their role in processing information: Yes, they affect the work of neurons and synapses, but they do not determine it. Hence the proverb: a full hungry one does not understand.

Fig. 2. Schematic representation of the structure of a neuron.

But back to the neurons and signals that they produce. The abstract averaged structure of a neuron is something like the one shown in Fig. 2. and consists of the cell body (soma), in which you can find all the elements of an ordinary living cell (nucleus, Golgi apparatus, mitochondria, etc.), dendrites - processes that serve as input from other neurons, and usually one long axon - by which a neuron broadcasts its opinion to other neurons. From this structure all mathematical models of perceptrons went to be - many inputs, black box, one output, profit. But what happens in the black box? How does a neuron process incoming signals and on what are the mathematical models of neurons based? It turns out that not everything is as complicated as it could be. Consider, for example, the axon mechanism of signal transmission.

Fig. 3. Axon structure with the main active elements for signal propagation: proteins responsible for the transfer of ions from the inside of the cell to the outside and vice versa.

1. Axon structure

As can be seen from Figure 3, the axon is an ordinary cell membrane, consisting of a double layer of lipids, alternating with various proteins, plus what is not shown in the picture is a cytoskeleton consisting of protein micro-tubes and connecting proteins. It is important to note here that the chemical composition of liquids is different inside and outside the axon. And, as we will see later, concentration gradients of sodium, potassium, calcium, and chlorine ions play a very important role. A different concentration of ions on both sides of the membrane is actively supported by just those proteins that are in large quantities in the membrane. Some of them support a certain membrane potential, the so-called resting potential (RP): in various organisms, it varies from -40 to -90 mV. It turns out thanks to that special proteins (called active transporters) unilaterally pump ions against the concentration gradient, while other proteins (called ion channels) allow certain ions to flow in the opposite direction. The balance of these two processes affects both RP and action potential (AP) - the main signal transmitted from one neuron to another, as well as synaptic and receptor potentials, which provide AR.

2. What is a signal for an axon and where does it come from?

So, since AP is what is transmitted from one neuron to another, it will be most important for us, so let's look at it in more detail. The work of proteins in pumping ions and controlling the level of RP within certain limits is quite well organized and even noise-resistant. This can be demonstrated by applying certain potentials to the membrane (Fig. 4).

Fig. 4. Schematic representation of the conditions for the generation of passive and active signals.

As can be seen from the diagram, at the equilibrium value of RP, the current does not flow through the membrane. If an external potential is applied to the membrane, then, within certain limits, its response will be linear. However, at the intersection of a certain threshold value (Threshold), an “explosive” membrane depolarization occurs - the very action potential. Moreover, it is important to note: when applying greater potential to the membrane, the AP does not become larger or longer, but the repetition rate becomes larger! And there is no magic: in a quiet time, the sodium channels are closed and the concentration gradient of sodium is provided by ion transport proteins. However, when crossing the threshold value of the membrane potential, the sodium channels turn on and very quickly pass a large number of ions into the cell, quickly changing the total membrane potential. Potassium channels of such treatment will not tolerate and will also open, but the equilibrium gradient of potassium concentrations is opposite to the gradient of sodium concentrations, so potassium will flow out of the axon, thereby neutralizing the effect caused by sodium and restoring the equilibrium value of RP. Due to the different speed of these two processes for a short time, an excessively negative membrane potential is formed, but it is compensated by the work of ion transport proteins.

So in fact, the axon is a simple ADC that encodes the amplitude of the incoming signal with the frequency of the output signals, plus a certain threshold below which it does not strain at all, and passes this signal on. The signal propagation along the axon is regulated by the same proteins: when AR appears in one part of the axon, it changes the concentration of ions in a certain neighborhood, and the same thing happens, only with a certain time delay, hence the completely finite speed of propagation of nerve impulses: from 0.5- 10 m / s for short neurons up to 150 m / s for long axons surrounded by myelin.

What happens in synapses when neurons communicate with each other?

Probably, it is worth starting with the fact that the synapse is the connection of the axon of one neuron with the dendrite of the other and the synapses are of two types: electrical and chemical. The former are quite rare, but actually represent the ability to transmit an electrical signal directly from one neuron to another. This happens when the synaptic connection of two neurons is so close that they “merge” and the ion channels of one neuron are directly connected to the ion channels of another neuron, allowing small molecules to move between them. Thus, electrical synapses simply allow ionic currents to flow between neurons. An interesting feature of this connection is that it works in both directions. Another feature - they are extremely fast. Therefore, such synapses are used in populations of neurons,

Fig. 5. The full cycle of signal transmission through the chemical axon: 1 - synthesis and storage of the transmitter; 2 - the arrival of the AR; 3 - opening of calcium channels under the influence of AR; 4 - flow of calcium ions through an open channel; 5 - fusion of bubbles with a transmitter with a membrane under the influence of an excessive concentration of calcium ions; 6 - release of the transmitter into the synaptic space; 7 - binding of the transmitter to receptors on the dendritic side of the synapse; 8 - opening of ion channels under the influence of a transmitter; 9 - the formation of AR under the influence of ion current through the channels; 10 - return of bubbles for filling with a new transmitter.

Chemical synapses work a little differently (Fig. 5). In short, when AR arrives, a neurotransmitter is released from the axon, which diffuses through the synaptic space and binds to receptors on the dendrite side, activating them. Activated receptors depolarize the dendrite membrane and thereby include the process of AP propagation through its membrane. In theory, it is simple, but in practice it became clear that neurotransmitters are for 100 different types. The question is, why so much? Studies have shown that many neurons simultaneously produce several non-transmitters, which, in turn, affect the receptors of the post-synaptic (signal-receiving) membrane in different ways. For example, small-sized neurotransmitters that are usually secreted by rarely coming ARs, quickly flow through the synaptic space and cause a quick reaction, while the large molecules released during frequent ARs move longer, but also bind to receptors more strongly, thereby causing a longer membrane depolarization. All this creates the prerequisites for better transmission of the incoming signal. It is also important to note that some neurotransmitters can excite the receiving neuron, while others can inhibit excitation, on the contrary.

Something like that.
So far, not much has been done about artificial intelligence, but we will get to it.

- [1] google "mystery-of-the-human-brains-glia-cells-solved-key-to-learning-information-processing"

Also popular now: