Artificial intelligence, who are we?

Introduction

The problem of creating artificial intelligence has been actively discussed for the past few years, experiments are being conducted, supercomputers are being used, sufficient funds are being spent to study the problem. In this article I will not be able to answer the question whether it is good or bad, and what can I expect from a rebellious car, but I will try to clarify the current state of affairs.

Neuron

What is a neuron? Let's see if those “bricks” of which our consciousness consists are so complicated.


If we imagine the cell as a black box, then we can say that through the “dendrites” (attachments of the “synapses”) the neuron receives data, and through the “axon” it gives the result. Very similar to a regular function, we pass the values ​​from the “dendrites” as parameters, and return returns the calculated signal value on the “axon”.
If we discard everything that supports it as a living organism from the cell, in the bottom line we get:


N digital inputs -> input signal adder -> activation function -> output value.

where:
1 - input signals
2 - adder
3 - activation function (for simplicity - any function)
4 - axon (output signal of a neuron)

So, trepanation

At school, in a biology class, Mary Ivanovna told us what the brain is and why it serves the person.
Let's look in more detail: what can we see after trepanation? Pink something - a full-fledged "supercomputer", weighing a couple of kilograms.

	Brain	Intel core i7
Elemental base	Neurons, size: 3-100mkm	Transistors, manufacturing process: 45nm
Amount of elements	100 billion neurons	731 million transistors
power usage	15-20 watts in wake mode	130 watts
Number of links per item	tens of thousands of connections per neuron	Proprietary technology
External Interrupt Response Time	About 1 second	<1 microsecond *
Number of Cores"	Simultaneous execution of up to 5 threads of “meaningful” commands + background threads	4
Self healing	Partially possible	SC 12months

* ~ 3000 cycles x 4 cores

Everything looks beautiful, we buy 150 processors, assemble a server, start it, and ...

In series, in parallel? Perpendicularly.

What prevents us from reconstructing the brain on today's supercomputers?
Modern processors are focused on complex sequential calculations, while inside the brain, signals are transmitted in parallel. This difference does not allow the use of conventional computers to effectively simulate brain processes.

Can I find a solution to this problem?
Yes, partly solutions exist; a microchip has been developed at the Massachusetts Technical University that can replace a living neuron. At the time of creation, the chip was able to interact with living neurons, however, due to some technical problems, the project did not receive large cash injections.
One of the most active, at the moment, projects in this direction is the Blue Brain Project, but its goal is to simulate the work of only 10,000 neurons (for which almost 8,000 processors are already involved).

"Ancient" and "new" brain, or ten percent problem

Where does consciousness come from? One could be upset if it arose from nothing, fortunately - this is not entirely true.
The human brain, unlike the brain of animals, can be divided into two, quite different in structure, zones: the "Ancient" and "new" brain.

The “ancient” brain is what remains of our ancestors, its structure is similar to the structure of the animal brain, there are reflexes, instincts, signal processing centers from all the periphery (vision, hearing, smell, etc.), the internal structure of this parts of the brain are formed in humans from birth.

The “new” brain - this is what makes us human, it accounts for about 10 billion neurons. Here our personality, memory, thinking is fixed. It happenedfrom birth, this part of the brain is not filled with information.
Because of the percentage of the “new” brain, the myth arose that people use their minds only by 10%.


Why this section? As long as we don’t have a disassembler of the “ancient” brain, human DNA (850MB of information) has not yet been deciphered, and, as we have noted, consciousness does not arise from scratch.

Reflection

The last of the main problems is related to the ability of the human brain to create, as necessary, new neurons and the connections between them. In the current implementation, when each neuron is considered to be practically a single physical processor - such a “dynamics" is rather difficult.
The main advantage
Creating an exact copy of our mind would not be very promising, most likely productivity will not allow it to think much faster than us. But there is one important advantage. Creating a mind - we design communication interfaces with the outside world, nothing prevents us from connecting the nerves of our newly loaded Masha, Dasha, Sasha, Pasha to powerful computer networks (including the Internet), and classic storage systems. The advantages of this approach are obvious, just remember any films about cyborgs.

Conclusion

Obviously, x86 architecture, like any of the most common, is unsuitable for reconstructing the mind. One can imagine, under the assumption that somewhere in the military laboratories more successful experiments are carried out.

If you exclude all the "secret" data and take into account technological difficulties:
No one bothers in the next 5 years to create artificial intelligence.
No one will create artificial intelligence in the next 5 years.