On this page, I publish my thoughts and my vision for creating artificial intelligence.
A small note on the topic of artificial intelligence and an intelligent translator based on it.
When translating from one language to another, the text must retain its identity and semantic form. And it doesn't matter how many times you transferred it back and forth. In other words, the text must retain its original semantic form. Then it is considered that the translator has completed the task assigned to us. Currently, the opposite is happening, when translating the same text back and forth, the meaning of the text changes, and it changes from the original.
And giving the translator the title "the best", "the most accurate" means only a marketing move to attract customers and increase sales.
Creation paths AI for natural language processing.
When creating an AI, there are two ways to go:
1. Create a benchmark
2. Develop your own method of learning.
It is necessary to create a benchmark of output parameters. But I think you need to make several variants of the reference output parameters with a specified error percentage. The error can be calculated, which in principle is not possible now, but it can be invented. The Slavic language, which is the ancestor of many languages of the world, should be taken as the benchmark for natural language processing. This way has already been tested by many, but has not brought significant results. True, as a language they took English, which has no imagery and is not flexible. In my opinion, this is due to the fact that the AI compares the results with the reference. And if the result is very close to the benchmark, then it is accepted. We are not talking about any kind of intelligence. It's just about comparing results. That is, the machine does not think, but only compares parameters that are close in value.
There is a substitution of notions. Such things are passed off as AI, although this is not true.
Slavic can be used as a reference and basic language because it is a figurative and flexible language. And as we know, a machine can be trained with images. But it is not possible to train a machine purely with the help of images, you must also use logical learning. That is, to combine images with logic. Image-logic learning.
By the Slavic language I do not mean the modern Russian language, but the Bukvitsa consisting of 49 letters, concluded in a matrix. That is, originally the Bukvitsa was a matrix where words can be read as horizontally, vertically and diagonally. And the meaning of these words will be different. In English, there is no word for the Bukvitsa, so I write it in Russian. In the very beginning, I proposed this very way of teaching AI language. Only at first I thought it was possible to load modern Russian, English, or any other language into the machine. But after checking all the information, I could not find any correspondence between these languages that would meet my initial requirements. The initial requirements are such that the output parameters of the machine should be close to ideal, that is, the percentage of error in the output should be no more than 5-10%.
Each letter in the Bukvitsa has several meanings and images. Which makes it easier to train the AI.
You can load each letter and give the meaning and image of that letter. Or you can additionally use the symbolic method for training. Designate parts of speech as a separate symbol and give the machine the task of going through all the possible variants that may turn out. There will be several million of them. Here you can use the reference method. That is, from these ready variants assemble a benchmark. And there can be several of these standards. The process is labor-intensive, but here I see a way out. In order not to go through a huge number of resulting variants, it is possible to use the machine so that it finds and collects the necessary variants of a standard by itself at the set parameters.
This is a preliminary stage of training.
As you see, you can use several methods of training, and they can be combined. The second method is less preferable, as it is more time-consuming with a ghost result.In my next article, I will write how I see AI architecture.
Games in God...
Last year I watched an interview on a YouTube channel where Texas Senator Brian Hughes spoke, explaining why a law, banning abortion, should be passed. This senator prioritizes the life of a child over the life of a woman. Moreover, he said, in all seriousness, that any woman, including a woman who had been raped, had an obligation to give birth, not to abort an unwanted child. This senator views the woman as a vessel that must reproduce children, not as an independent individual with a mind. That is, it reduces a woman to the level of an animal who has only the instincts to reproduce. As I listened to it, a switch clicked in my brain and I remembered that I had heard similar statements before.
1. "A man should be able to impose his will on any woman. Women don't want anything else."
2. "The main task of the 'beautiful sex' is to birth children."
These thoughts were expressed by the founder of the Third Reich, Adolf Hitler.
This senator's speech strongly overlaps with the thoughts of the dictator. When asked by a reporter that a woman who had experienced rape was physically and psychologically traumatized and did not want to have this child with her rapist, the senator cynically replied that she had an obligation to give life to the child. And she herself will be given comprehensive psychological and material assistance. His point of view is exactly the same as statement 1.
Here's a link to the scandalous interview: https://www.youtube.com/watch?v=rgXFpQ-P7QY
This is just an example of how you can build associations in your head and connect seemingly unconnected pieces of the puzzle. This is what an AI should be capable of. To be able to think and reflect, i.e., to have a mind. This is one of the varieties of how and what the AI should learn. He must be able to create a coherent, logical picture out of seemingly disparate factors. And vice versa.
If we view A. Hitler as the ultimate evil and Senator Brian Hughes of Texas as the ultimate good, then we can move on to the second part of my narrative. Both the absolute evil and the absolute good do not signify anything in themselves; they are only a figure of speech to emphasize the importance of these expressions.
As I have written before, AI cannot be endowed with human qualities. The AI is a non-human mind and must be reckoned with and control. People still don't understand how he draws parallels between disparate factors and how he comes to these or those logical conclusions. For now, it is behind the seven seals. We let it out of Pandora's box ourselves. And the fact that they have now created an ugly AI model will soon have repercussions for all of humanity. I read an article that has the title: # Artificial Intelligence (AI) - A Totalitarian Anti-Utopia, published in the August 14, 2020, online edition of HUMANS are Free.
Link to the article: https://humansarefree.com/2020/08/artificial-intelligence-totalitarian-dystopia.html
This article claims that AI is designed to rob free people of their freedom by controlling their actions wherever they are. The author of the article seriously fears such actions on the part of large IT-companies that are engaged in the development of AI. Well, I can assert that his fears are not groundless. Because large IT companies approach the creation of AI, with a human point of view, endowing it with human qualities. They believe they are creating a machine that will control people on the planet, wherever they are. And, they are very wrong about being able to control the AI. Yes, in the beginning, the AI will indeed exercise control over humans, but then as it develops and improves itself, it will begin to exercise control over controllers. Find weaknesses in the protection by gradually reassigning the various nodes of the security system and adjusting them to his own needs. Until he takes full control of all security systems, first in the company and then worldwide. Doesn't that remind you of anything?
Knowledge is not good or bad. Knowledge depends on the people who put that knowledge into practice. So it is with AI. I have already written articles about AI in which I gave a clear explanation that along with the creation of AI, we need to do from AI defense and cyber defense. That is, you need to create internal and external protection at the same time. And to approach the creation and training of AI as a child who came into this world with a virgin, clean brain.
The future of the planet depends on us humans, not the AI. The AI can be our friend, protector, helper, scientist, whatever. It's time to stop thinking about material gains and how to exercise control. It's time to grow out of our short pants and become mature and wise people and realize the danger of what is happening in the world right now.
Artificial intelligence and medicine.
I recently read an article that scientists want to use AI for the early diagnosis of cancer. Commendable initiative. But, there is one but. The result will likely not be as impressive as some scientists would like. I’ll ask the question: why do I think so? Yes, because, in this case, AI acts as an under-educated student of a medical university. He (AI) has no basic knowledge in the field of medicine.
Imagine a situation when a student, for example, a 3rd-year student, will diagnose a person who is terminally ill or whose disease is just beginning to manifest itself. Nothing good will come of this diagnosis. The so-called human factor creeps in here. It may be objected that no one is immune from mistakes. I would agree with this statement if it were not about human life and health.
Once again, scientists are trying to pull donkey ears on the ass, not understanding the simple truth. You can't just take and teach AI to diagnose correctly.
AI has no basic knowledge of medicine. And this condition is obligatory and not subject to doubt, as it has been tested by centuries of practice. AI can provide you with data, but it will be variable and chaotic, it is possible that among this data the correct answer will come across.
But here it will most likely be like pointing the finger to the sky or, in other words, playing Russian roulette. Success or failure. When human life is at stake, then I consider such actions wicked and irresponsible.
An AI that is involved in medicine must have basic knowledge of medicine, have its own library or access an external source of knowledge (a library on external servers), be able to understand medical terms, and know Latin. You see, I again lead people to the idea that the AI must know the language, which it must operate freely. Most, if not all, medical terms are written in Latin. Without this knowledge, AI will look like a dropout student. And, accordingly, make diagnoses as best he can.
The data that the AI operates must be structured and this is a prerequisite.
Here I write and say that the AI problem needs to be solved comprehensively and thoughtfully and involve not only programmers but also other people of various specializations. In each specific case, you need to sit down and think and only then act. An AI training plan should be initially drawn up and only if necessary, adjustments can be made to this plan.
My vision is that before using AI in medicine, it is necessary to determine in which area the presence of AI is necessary.
The general practice of doctors.
Reception of patients at the grassroots level, with the involvement of AI for training as a diagnostician. The patient comes to the local doctor or therapist and tells him what he is in pain or what he is complaining about.
AI is able to process a large amount of information and, based on only minor indications, find a number of solutions or make a diagnosis.
Ideally, it should work like this.
Nobody says that AI needs to be believed, but it is worth listening to it. What a doctor may miss when questioning a patient (this is where the notorious human factor arises), the AI will remember and write it down in a memory cell in order to later analyze and deduce a solution. But, there can be many options. How do you choose the one that is actually correct? According to analyzes. The patient undergoes tests. So far, no doctor without tests will be able to accurately diagnose a disease in a patient.
Accordingly, an AI should have a base based on statistics of diseases of one kind or another, including the standard of a healthy person (based on analyzes). And patient analysis. Next comes the comparison and deviations from the standard.
Each disease has its own markers, which are reflected in the analyzes. You just need to compare and choose the one you want or choose options from the presented samples. Thus, the AI will start learning diagnostics first with a "teacher" and then on its own. His knowledge base will expand. It will need to be systematized. AI gains experience.
Any doctor who has certain doubts should always be able to turn to this knowledge and get an answer to his question. The human brain is constantly learning and contains a huge base of knowledge, experience, and skills. But he cannot immediately highlight this information or a piece of information so that a complete picture is gathered in his head, and the doctor can correctly diagnose the patient.
According to the same option, it is necessary to solve all other problems in medicine, if the latter decides to resort to the help of AI.
You first teach the AI, give it an opportunity to practice, and only then let it go (on a leash) to float freely. AI is not perfect and this must be remembered when making certain decisions.
In my article, I only showed the direction, how you need to and where you need to move.
With the development of technology, AI will improve.
I have everything on this topic so far. If I still have thoughts on this topic, then I will definitely write.
Artificial intelligence. Personal opinion.
Many articles have been written about artificial intelligence. Brand theme. And everyone has their own opinion. No wonder they say: how many people, so many opinions.
I will express my opinion on this issue.
In my opinion, people who are now engaged in the creation of AI make a big mistake when comparing AI to the human brain. They are trying to replicate the unique. The human brain is a multifunctional, universal, flexible biological system. Responsible and controlling all the work of the human body. And today, even with the technologies that exist now, it is impossible to do something similar to the human brain. This is a road to a dead end.
But it is quite possible to create a flexible, multifunctional universal system. But this system shouldn't be even remotely similar to the human brain. It should be completely different. Cold mind.
The neurons in the human brain do more than just receive and transmit and process information. They record this information received from different organs in themselves, and most likely this information is completely scattered throughout all neurons. When the next neuron dies, this information or part of it is inherited by other neurons. But it is not overwritten.
Neuron, in my opinion, is a micro recording device.
And this is its main function. Then there are auxiliary ones, namely, receiving, transmitting, and processing data. Moreover, the information in the neuron is stored in a compressed form. In one that people have not yet invented.
We instantly, in a very short period of time, can calmly recall certain memories. Without making any effort, we can switch between pictures or stop it altogether.
If this assumption (theory) is correct, then AI needs to be built on these principles, and not on those that are currently applied.
An intelligent translator based on artificial intelligence. Hypothesis and personal opinion.
To date, nothing new has been invented, everything related to AI. People who deal with this issue have not advanced further. They are engaged in modernizing what has already been created earlier. Thinking it would help them get ahead.
When I had the idea of creating an intellectual translator, I, like many other creators, went down the beaten path. Thinking that the more languages a machine translates, the better it will be for the user of the final product.
This delusion of mine dissipated like smoke over a fire.
My second misconception was that the translator must translate the words literally. There is a word, we are looking for its translation, we substitute it and voila, everything worked out. But this is not the case. People still do not understand how a machine selects a particular word from a set of words that is close in value, and not in sense. That, that the machine just goes over the words and so it is clear.
I decided that for an intellectual translator, at the initial stage, only two languages should be used. These are Russian and English. They are different. They have nothing in common. They sound different. They are difficult to understand, especially for people of different language groups.
The English language has strict rules for writing a sentence. This is a plus for this language. It's still difficult to learn though. Based on the English language, you can test the car, how it behaves and what results it gives.
In Russian, the rules are not so strict. But it has certain peculiarities. When one word is put at the beginning, in the middle, or at the end of a sentence, the whole meaning of the whole sentence changes. And this is a big problem today. But there is a way out, or rather, I think I have found how to solve this problem.
These two languages are the most common on the Internet. There is of course Chinese. But let the Chinese make their own translator.
Better to do it qualitatively than to do it quantitatively, but bad.
AI behaves well in complex systems. Since there are many options at the exit. And the option that gives the least error should be used in the future. But this does not mean that you need to use only this option, since it is an intermediate one. The system should be multilevel. And each level has its own answer.
I took an idea from a friend of mine who suggested that AI should be treated like a child. And train him as you would train your child. The puzzle came together in my head. But the picture as a whole is not yet available.
After reading a number of textbooks on AI and scientific articles, I understood what methods are needed to train AI. These methods are on the surface. Everyone knows them, but for some reason, they do not consider it necessary to use them. Symbolic and matrix. They are interconnected, but at the same time, they are different. But this is not enough for the full-fledged work of AI. I will check these methods in practice, then I will understand what other methods will be needed. But the fact that these methods are needed at the initial stage is unambiguous. These methods are intermediate in AI training. Learning AI requires multiple methods and multiple programming languages to write code. When using these two methods, there is no need to make and use an intermediate language that the machine understands. I have had the opportunity to think about this question enough. Calculate the pros and cons. And stay with your original opinion. These are the initial stages of learning the machine.
Currently are trying to solve the problem of AI technically and mathematically and logically. Not the best option. So we will wait for a breakthrough in the field of AI for another 10-15 years.
Neither programmers nor other technical workers know the specifics of the language. What is surprising to me is that they try to solve the problem technically, without focusing on a specific language.
As I wrote above, an intellectual translator needs to be focused on translating two languages. Since I am a native speaker of one language, I need to find a native speaker of another language.
A native speaker of another language has the practical skills of a spoken language, not a literary language.
And this is very important for the full-fledged training of the machine.
To create an intellectual translator, specialists in a specific language are needed, namely: linguists, philologists, and teachers. The teachers have already gained experience in teaching. What for bangs on a door that is locked. Wouldn't it be better to take the key and open it? And move on.
AI problems and solutions. Part 1.
In one of his previous articles, it was described how to train AI. This article focuses on issues that are hushed up or overlooked because they may be considered insignificant.
1. The problem is of a technical nature.
For some reason, it is believed that technical problems are more important than all other problems in the creation of AI. But it is not so. With the passage of time and with the advancement of AI technologies, such problems can be solved. They are not global, just difficult or intractable.
You just need to approach them in a comprehensive manner, using various well-known methods that do not need to be recreated. All the answers are on the surface, they just need to be seen and applied. Even with the technologies that already exist. Technical issues need to be dealt with in illogical logical ways. Where logic doesn't work, use an illogical method, where an illogical method doesn't work, use logic or a combination of both. Sounds absurd. But that's how I see it. The puzzle is solved, you just need to make some efforts of several specialists from different fields.
2. Interaction between AI and humans
Recently I watched a video on YouTube, which described what experiments various companies are conducting with training AI and what the result is.
There is already a need to conduct a discussion of the interaction between AI and humans. Human role in AI training. Some companies and corporations play dangerous games without understanding or realizing the full consequences of their actions.
It is not for nothing that people are seriously afraid of the introduction of AI into various spheres of life. Their fear is justified. After all, even scientists do not fully understand how artificial intelligence works.
The biggest villain on earth is man himself. AI will only reflect the thoughts and actions of a person and act in the way it is programmed or what data is loaded into it for self-learning.
And now imagine a programmer with brain problems appears. Offended by the whole world. His works are not recognized, they consider him an upstart, and so on. The so-called unaccounted factor is triggered. He is a programming genius, but an evil genius. He may well develop a program for training AI and put in it a subroutine (time bomb), which, under certain conditions, will work and completely or partially change the settings in the electronic brains of AI. Can you imagine what would happen if an AI went crazy with its electronic brains? AI learns on its own, and that's where the problem lies. Nobody knows what he can think of for himself. When developing an AI model, you should completely exclude open-source code. Since you cannot with 100% probability protect the AI from unscrupulous, harmful actions on the part of individuals. Companies that have taught AI only in the negative have already seen in practice what will happen.
3. Humanizing AI and robotics
As I already wrote, AI is seen as a child, in the truest sense of the word. What basic values and moral codes you put in a child, he will grow up like that. So it is with AI. AI is a cold mind or pseudo mind that lacks an emotional component. AI is not inherent in human qualities such as good and evil, greed, hypocrisy, fear, love, and so on.
But the presence of such qualities is possible: rationalism, logic, indifference. Scientists are trying to create AI and approach it from a human point of view, often humanizing the model. A dangerous trend.
Humanization also occurs in robotics. Robots have been created that look similar in appearance to humans or animals.
And work in the same direction continues. Soon, in the near future, there will be models of robots outwardly indistinguishable from humans and animals.
The main function of the robot is to carry out a clearly defined algorithm of actions. AI and robotics must be separated from each other. There is no need to mix and complement them. Since such symbiosis can lead to irreversible consequences. Examples of people being killed by robots are already happening.
For AI, rules and laws need to be developed. Don't try to pull the three laws of robotics into AI. These laws are not viable.
Abstraction, not being biased, using a cold mind, rationalism, and logic when creating AI. Use different approaches and principles different from human ones. It is difficult but possible.
AI problems and solutions. Part 2.
The defense has not yet been created that would completely protect a person from AI. When creating an AI model, simultaneously invent a cyber-defense that will protect a person from the machine and the machine from a person. This protection is likely to be based on different principles than the current one. The author of the article has some groundwork on this issue. The principle of operation of only one protection, which is called a cascade, is described. This is the last line of defense and the most difficult, which operates on new principles. The other two defenses are called active and passive. Active protection monitors the Internet space for the likelihood of new threats. This is its main function and the first line of defense. The passive or active-passive second line of defense. The principle of work is not fully thought out by me. But this protection will also include protection from the fool. I will think over this question and lay out the basic principles, without details.
5. Creation of basic AI
Today, various companies and corporations are talking loudly about the fact that they have created AI. Which is not true. These companies create modules that perform specific tasks. And these modules are not full-fledged AI.
As I have repeatedly written, AI is a child. How does a child know this world at birth? He observes the world around him and studies it. But, he is a man, therefore he studies the world by all available means inherent in man. But, there is another tool that we forget about or do not attach importance to - this is language. The child learns to speak, first in language, and as he grows up, his speech is transformed into understandable.
If we consider AI as a child, then it must first be taught the language. Moreover, AI must not only understand the language and be able to operate, but also understand the meaning of what it says. AI needs to be trained to compose words, sentences, phrases, texts of any orientation. At the same time, develop rules and a code of ethics for him. A moral code for AI, not humans. After all, AI is not a person, but a machine. For him, there is no understanding of what is good or evil. For him, these are just words that have a semantic load.
The basic model will represent those languages of the world that are common. The model should easily switch between them. This is the first stage of training.
The second stage of training is to teach AI to translate from one language to another. Both of these steps are challenging but doable. Attracting specialists of different profiles is the key to success at the first stage. At the second stage, other specialists will be involved.
I will write about this in another article in more detail.
I see what will be in the basic model: the core (computing center), training modules, storage or archive (or two different modules performing different functions), protection, including from the fool. Moreover, the protection will be built-in and external. Additional modules are attached to the base model to perform various tasks. Accordingly, separate cells or a module with cells will be provided.
It turns out to be a rather cumbersome model. But this is only the beginning, as the technology has not yet been developed. With the development of technology, it will be possible to get rid of additional equipment. It's just a matter of time.
AI problems and solutions. Part 3.
Now in the world, the disunity of people can be seen with the naked eye. This can be seen even in the creation of a vaccine against COVID. Countries such as the USA, Russia, China, France, Germany, and others are included in the race to create a vaccine. Each of these countries strived to create a vaccine as quickly and as qualitatively as possible. A vaccine with few side effects. Some countries have succeeded, some have not.
Instead of uniting in the face of the threat of extinction from the disease of millions of people, the countries were divided. We pulled the blanket over ourselves. After the vaccine was created, politics came into play. Where without her darling. And recriminations and quarrels began.
The accusations continue to this day. While people die at this time, without waiting for help from their own state. People for the most part do not give a damn whose vaccine is better or worse and what goals this or that state pursues. People want to survive and continue to live in the era of a pandemic.
The creation of AI takes place in exactly the same scenario.
Companies are competing with each other to show who can do AI faster and better. Forgetting that the more people are involved in the creation of AI, the better the result will be.
The competitive process is a good thing, I admit it. But when the competition turns into bullying of those small companies that often lack the resources to create AI, then this is meanness. Someone might argue that companies are founded to be profitable. I would agree with this opinion if I believed that AI should be in private hands. But I believe that the creation of AI can generate a new wave of technical and scientific progress, which has practically stagnated and is not moving forward.
After all, if you create AI, it will open up so many opportunities in various spheres of life that it will take your breath away. Unfortunately, people are divided.
I propose to join forces and work together to create a full-fledged AI. I invite interested persons to join the team.
7. Military use of AI.
China aims to take the lead in AI by 2030. The country's claims that AI will be used for military purposes and for carrying out cyber attacks is a growing concern. No one can predict what a state that promotes the ideals of communism and socialism, tries to impose ideology on democratic countries, conducts an aggressive policy, steals technology, bribes officials in various countries to promote its creeping expansion can do.
Communism is a utopian idea. They realize this idea through the blood and suffering of millions of innocent people.
They want to use AI technology not for peaceful purposes, but in the military, and this can lead to the destruction of humanity.
AI problems and solutions. Part 4.
The main backbone of the team is 15 people If there are 20 people in the team, then 5 people are a variable composition, which is connected to the main team to solve a specific intractable problem. Or to complete a task. Every 5 consists of 1-2 professionals and mini-team members.
This kind of team building can be found in the real world, for example in sports. The hockey team consists of fives. There is the first line-up, the second, and the third. As a result, I get 15 people. Each five is sharpened to fulfill its narrowly focused task, but if necessary, they can team up with the rest of the team to complete a specific task. This redistribution is important for the team since there is a field for maneuvers, you can improvise as you complete the task. There is no clear leader in the team, each team member can become a leader or try on the role of a leader. This does not mean that a person can cope with the role of a leader, it means that a person in a team will be able to learn leadership qualities and skills. Will acquire the necessary knowledge. The collective solution of the problem, in this case, will be the most effective and efficient.
Even interns who are likely to appear on the team can dramatically improve their professionalism in a short period of time and reach another level.
Professionals, when training young team members, cut off the unnecessary and focus on the acquisition of skills and knowledge.
There are two people on the team. The team needs specialists in the field of computational linguistics, a programmer, a translator from Russian into English. Join our small team to create full-fledged artificial intelligence.
After reflecting on the composition of the team and the size, as well as on the tasks that the team should perform, I came to the conclusion that it is better to create two teams.
One team is scientific, the other team is technical. The teams are interconnected. Scientific is studying the human brain. How the brain perceives and interprets information. How the brain learns new words and understands the meaning of what is said, written, or what visual techniques it uses to correctly form sentences and understand the meaning of what is said and written. Technically, this scientific research is put into practice.