The theory of human memory, the beginnings of AI

The theory of human memory, the beginnings of AI

Surely all of you are very well aware of such moments when you need to remember something, but it becomes a big puzzle to extract information from the brain.

Why is this happening. First, a little theory of the work of a neuron, you can read here or here
')
Suppose, and maybe it is, all neurons are combined into one very large graph with a complex structure. This structure is complex and cannot work randomly, i.e. the transmitted pulses are transmitted strictly in a certain order, so there are 2 options here:

1. Count edges have only positive weights.
2. Graph edges can have both positive and negative weights.

Considering the second case in the real work of a person’s memory, it can be assumed that this situation occurs when a person’s memory fails, i.e. the neuron containing the information that we need or does not receive enough signals to accumulate and transmit further, or these signals do not exist at all. In the case of graphs, this can be represented as a node with few paths, either they are negative or they do not exist at all (Figure 1).


As for the first case, when all edges have positive weights, i.e. human brain is not damaged. Then why can't a person remember moments from his childhood? The answer is simple: “Any body strives for peace”, and so our neural network tries to optimize its work. (Owners of navigators should be familiar with the fact that laying a route is based on the principles of the graph, finding the shortest path, etc.). The human brain is a more sophisticated system and its optimization consists in breaking ties with small weights, and building new connections with higher ones. (Fig. 2). This explains the numerous breaks and new connections of neurons. The more nodes there are connections, the easier it is to remember the necessary information.

Practical example

You went to the cinema, to a regular movie, at first you remember almost all the moments, especially those that emotionally shocked you. A month later, a year, who has brain activity loaded for you, if you ask about this film (given that you have not watched it all this time) you can tell a short plot of the film and those moments that you emotionally shook you. This is the optimization of neurons .
It should be noted that each node has its own bandwidth, i.e. if there is a lack of connections, information can and not be remembered until the required rebuilding of the graph. During optimization, the transmitted pulse has a stronger signal, and the node itself does not change its bandwidth, or not significantly, which is why we remember very important things, as they say at the subconscious level.
In the process of remembering, you can feel a lack of information, and continue to remember, then your brain is looking for a longer path in the graph (with a lower weight), and passes along its edges into nodes that are not “involved” in everyday life, because insufficient accumulated pulses for further signal transmission.
That is why the best way to remember this association . The more edges in the graph go to the top, the easier it is to remember the information.

Optimization or partial loss of memories

DREAM - switching the brain from active memorization to active optimization, in a dream we see what we have seen before or the synthesis of these things. Synthesis in this case is the breaking of connections and it is possible to build a new connection with another vertex.
You can try to give an explanation of why it is better remembered before bedtime. Firstly, the information is fresh, and quite complete (not optimized), and our brain considers it important, it gives weight. During the night optimization, the information received shortly before sleep does not have time to mix with other information and is lost in the “infinite” graph, which means its connections remain almost unchanged. Of course, this is only a theory, each person’s brain works in its own way, each has its own scales of “useful” and “useless” information.

The rules of AI built on this theory

• All accumulated information over a period of time should be optimized, i.e. establish a causal relationship directly from the starting point to the end, passing intermediate (breaking the network, and creating new connections with a large weight).
• After optimization, all data with low weight should be destroyed, and all weights should be reduced to the minimum weight from all destroyed vertices.
• New information must have a weight greater than the minimum existing to properly assess the importance of information.
• All destroyed information should build a new graph, and be optimized. The reason is that the constant optimization of the main graph will lead to the fact that there are no details that play an important role in a person’s life (emotions).

Problems

In this case, the problem is optimization, i.e. selection of the information that will live, the most significant for the life of man and the entire planet, and that which should go to the second, third, fourth, etc. plan.