AI today Is it possible to build artificial intelligence today

Is it possible to build artificial intelligence today?

The article discusses various aspects necessary for the construction of artificial intelligence, in contrast to most of the articles on AI, the topic of neural networks is not considered.

One should not confuse the problem of artificial intelligence with the question asked by one scientist: what do you think when the machine becomes smarter than a person, to which he replied: you know a washing machine in my house, it is also smarter than some. Today, computers are smarter than man not only in purely arithmetic operations, but also where man considered himself to be the king - in logic games. To date, the computer is able to beat even Kasparov in chess.

Let's try to clarify the definition of artificial intelligence. In the literature, I found the definition of how to solve complex problems, creative problems, with which I can not agree. Some authors argue that the computer is not able to write interesting stories, write beautiful poems, draw beautiful pictures. But today it is not. Take for example fractals and see for yourself using the Fractal Explorer program. In a matter of minutes you can draw unique pictures that no one has ever seen, thanks to the infinite sets of pictures generated by the program. In addition, these pictures can be drawn only on a computer in a graphic editor, and it is almost impossible to draw with paints and pencils.

When solving most everyday tasks, a person uses mainly memory, slightly loading his brain, recognizing the problem that has arisen and extracting a solution from memory. And the information gets into the memory in the process of many years of study. But when we are faced with a new task, we cannot solve it immediately, and then we fully use our intellect. Based on the initial information about the task, we mentally, that is, inside our virtual machine, try various options, set up experiments and try to choose a method or sequence of methods to achieve the goal. And thus you can solve almost any problem, and this feature of the brain is especially interesting for us. Let's call this feature as a universal tool for solving any problems, that is, intelligence.

In the human brain, there are about 12-14 billion neurons. The neuron is able to perform a separate operation, has tens of thousands of inputs, and the output goes to tens of thousands of other neurons. Very similar to the function in the program, it is not true, we have several parameters at the input, we have the function value at the output, and this function can be called from hundreds of other places within the program. A neuron can operate at an average frequency of 10 Hz, from these numbers it is possible to determine approximately the brain productivity of 120 GHz.

The performance of modern desktops is approaching the 1 GHz mark, which is clearly not enough for solving problems similar to humans, real-time speech recognition, pattern recognition, recognition of dynamic images and three-dimensional scenes. RO is the highest priority and time-consuming task for our multitasking brain. For example, for an unprepared user, it’s enough just a few minutes to look at a game of professionals in 3-D shooters similar to Quake III, and he begins to feel dizzy and nauseous due to the fact that his brain does not cope with processing unnatural fast movements in the game and other less priority tasks CPU time remains, such as digestion and coordination of movements in space.

Similar phenomena can be observed while riding on a swing or circling in place. But modern supercomputers, consisting of approximately 5,000 Pentium III 1000 MHz processors, have a capacity of about 5,000 GHz, and this already allows solving tasks similar to those of an average person’s daily life, but such computers are mainly designed for simulating nuclear tests.

Modern network projects unite more than 1,500,000 computers, and this, with an average performance of 500 MHz, is a whopping 750,000 GHz, and no supercomputer can withstand the global network. But this project Seti @ Home is designed to search for extraterrestrial intelligence and analyzes information from the radio telescope around the clock. That is, today humanity has the computational ability to create artificial intelligence, but for some reason it still does not exist, we go further.

What do you think, what vocabulary is stored in the memory of the average person, well, at least approximately? According to my humble estimates, about a million, with this calculation it is necessary to take into account that all the so-called different forms of one word are nothing but completely different words having a connection between themselves, and pointing to one object or a similar object with excellent properties. Since for each such word we can give out different information, such as gender, number, case, etc., therefore these albeit slightly similar words are stored in different cells of a person’s memory. With an average word length in Russian of about 10 characters, the dictionary occupied about 10 MB. The program read several hundred megabytes of texts from various electronic libraries, after which I and my friends could not find a word that is not in this dictionary.

So imagine the performance needed for the intensive conversation of two people. We formulate our thoughts, then collect the necessary words, form words from the words that we then pronounce, and an unintelligible sentence must be formed before the phrase begins to be pronounced. Then our interlocutor listens to our phrase, determines its main characteristics, such as loudness, speed and other individual characteristics for each interlocutor, including the pronunciation of individual letters. Sounds are recognized and obtained letters, words are collected and recognized from letters, phrases from words, and only from phrases the interlocutor recognizes our thoughts and proceeds from these thoughts and produces a reaction. And then the cycle repeats several times without interruption. It is interesting that the entire cycle of generating a response and the formation of a response for reproduction occur in hundredths of a second.

For hundredths, it is necessary to process the entire available dictionary (of characters, words, phrases, thoughts) and work out a reaction. Unfortunately, I can’t even approximately give the volume occupied in the memory only by dictionaries, because I have never seen a complete dictionary of the Russian language. I mean not those dictionaries that are used in Russian lessons, but a complete dictionary that includes absolutely all the words that people use. After all, in order to use a word in a conversation, it is necessary to have accurate information about what it means, and if it is not in our dictionary, then it does not exist at all.

Now let's talk about memory resources. 64 MB of RAM is on my computer, and 10 MB is required to store the dictionary, we still have to remember Window NT 4.0 operating system. That is, there is not enough memory, as you need to store the program for processing the dictionary, and other information. For any recognition of words, the dictionary should be in RAM, since recognized words must be checked for presence in the dictionary and, if an error occurs, make corrections to the recognition algorithm or obtain additional information to create a new concept and save it in the dictionary. In addition, for each word it is necessary to have its sound image, and from the point of view of modern multimedia programs, this is a WAV file with a duration of about 1 second 22 KHz, that is, about 22 KByte.

That is, for the recognition and playback of sounds, we need about 22 GB, and it will not fit even on my hard drive. Of course, you can put as many memories on a supercomputer, but it will be very expensive for a toy that can read the words correctly and in response to one spoken word will pronounce another or several words. One should not forget that a dictionary of phrases and sentences is also necessary for conducting a conversation, because only a sentence expresses a complete thought.

But these are only technical problems; they demonstrate that today it is impossible to create even a simple talking program on a desktop computer, but this is quite possible on a supercomputer or on a network supercomputer that for days uselessly break demo unwanted encrypted messages. But in the near future in 10 to 20 years, such computing power will be on the desktop and solve our problems with you if Bill Gates, whose name even Word knows, doesn’t load them, with annoying animated videos, since in 10 years productivity increases on average 1000 times.

Today, the most powerful tool for solving problems, that is, a person has intelligence, and we’ll see how a person solves problems and where he gets these solutions from. Let's not forget that we have been taught for 10 years at school, then for 5 years at the institute, but by and large people have been learning all their life. Let us take as a basis a young engineer who is fully capable of solving various tasks, but he was intensively programmed to solve various problems for 22 years and as a result there are a lot of tasks in his head and for each of them one or several solutions, well or not one for example perpetual motion. All this heap in the head represents chaos in varying degrees, because of the whole heap of solutions, many decisions are not correct, or not rational, or not optimal.

Because different people teach us at different times, and often when one teacher says “you must do this”, another says with the same confidence “you just can't do this” or today we are building communism, and tomorrow we are breaking it. And thus, each of those who program us is mistaken in his own way, and we ourselves have to pay for errors.

If we take humanity as a whole, then at the beginning of its development it knew nothing, and at the moment we see a completely different picture, humanity has escaped beyond the limits of the Earth and beyond the limits of the solar system. Modern technologies help humanity in everyday life, in the treatment of diseases, in work and in war, science with the help of modern technology has approached the borders of the universe. Where does all this progress come from? After all, the information that we receive today from newspapers, magazines, television, radio, libraries, from friends and acquaintances at that time was nothing.

From here there is only one conclusion: all this information was obtained from the environment, by trial and error, that is, by search - the only method that allows us to acquire new knowledge. What makes humanity and today the production clearly performs the technological process to produce high-quality products from high-quality materials, while scientists are researching new, more advanced methods, spending a lot of money on analyzing all possible production options to find the most optimal option, which is brute force.

After we have completed a full course of study, the algorithm for solving any problem on the example of the passage of a maze looks like this. The problem that has arisen can be compared with where we are in the maze, and the solution of the problem can be compared with our actions aimed at getting out of the maze. If you hit the square of p783, then the solution will look like this: go right, up, left, right, ... down, right, up right and there you will see a way out. And so for several points in our maze, we have standard solutions. For all tasks it is impossible to have standard solutions, since there is not enough space in the memory to describe all the tasks and to store all the algorithms for solving, and here non-standard problems arise.

An example of solving a non-standard problem based on a standard one: if you hit the square p782, then the solution looks like this: the shortest way to get to the nearest point that has a standard solution, that is, p783, and then perform an action to solve this problem. But such an algorithm works only if the distance from our task to the standard task having a solution is rather small, and this is not always the case. Sometimes there are fundamentally new tasks that can not be solved on the basis of their experience, because it simply does not exist. This is also an interesting point because in order to make such a conclusion it is necessary to analyze all the existing many years of experience. For example, a non-standard problem about a lamp and three switches published on the main page, then you have to look for doors that are closed, and pick up keys for them, and sometimes you have to break walls to get new opportunities to find solutions in space.

Only in such non-standard situations does what I call the intellect turn on - the slowest and at the same time the most powerful tool for solving problems. And the solution looks like this: we are in the room with three exits and we don’t know where they lead, because we have never been here, in order to find out where they lead we need to consistently go through these different ways and only when we have them let's go we will know where to go. But at the beginning of the journey, all the doors can be absolutely identical, and we cannot get at least some useful information from them that will help us make the right choice, except for the sequence number. But the same door arrangement may refer to a completely different part of the maze.

Well, here with natural intelligence it seems dealt. Fundamentally other more complex tasks I do not know. And the complexity of practical tasks is connected before with a large amount of information, insufficient computing power, inaccurate formulation or lack of necessary information to solve the problem, and as a result an inaccurate, probabilistic solution.

And what we lack to create artificial intelligence. The construction of AI must begin with the creation of a vocabulary of natural language words, because natural language is the basis for building natural intelligence. As the alphabet is the basis for the construction of words, and no one gets the idea to come up with a few new letters. Nevertheless, I have not seen a single program that would know all the words of the Russian language, not to mention the program-translator from one language to another, which is much simpler than AI.

Despite the fact that there are all the necessary technical conditions for creating a translator, even on a desktop computer, that is, enough processing power of the processor and memory, there is still no program that could at least compare to a person with the quality of translation. The program speeds ahead of the person, but the quality remains to be desired, often unfamiliar words for the program, the translation often makes no sense and is harder to understand than the original, sometimes words fall in a sentence from a completely different context.

In my opinion, today the main problem of creating AI is not technical, but lies in the fact that all the information people have to enter manually using the keyboard. Using this not very convenient, low-speed device for entering information, we need to enter huge amounts of information, those that we introduce in the process of learning about 22 years in natural language. It is quite another thing to teach a computer to get information from the environment (listen, see, understand), put him behind the TV and let him learn by collecting the necessary information from newspapers and magazines and asking the necessary questions in natural language. Which, in principle, makes the intellect natural for a very long time, when a child learns to listen, see, speak and asks an infinite number of questions. On the basis of this simple and natural algorithm, we will be able to build artificial intelligence, which is so necessary for us.