This text is translated into Russian by google automatic human level neural machine.
EastRussia is not responsible for any mistakes in the translated text. Sorry for the inconvinience.
Please refer to the text in Russian as a source.
And became an AI
Artificial intelligence (AI) is already involved in various aspects of human life, but they continue the process of its creation. In the Far East, this is done at the FEFU.
The state has allocated 7,3 billion rubles for the study of artificial intelligence (AI). The money, among other things, will be used to create 24 AI research centers in Russia within four years on the basis of universities and scientific organizations. Where artificial intelligence is used now, what it is capable of, and what problems humanity can face with its active implementation, EastRussia asked Ilya Mirin, the director of the FEFU School of Digital Economy.
FORECASTS AND GENERATION
- Ilya Gennadievich, the School constantly implements projects related to AI. What are the areas of interest?
- The large area of "artificial intelligence" has its own subject areas, among them, for example, computer vision, which is very different from the generation of text. At the same time, there are a huge number of different branches within the directions. And here you do not need to artificially restrain yourself, because often the knowledge and results obtained are applied in completely different areas. The same convolutional (a special architecture of artificial neural networks aimed at effective pattern recognition - ed.) Networks that are now used in many places - in its pure form, computer vision. The technology allows you to well identify faces on video cameras - at different times of the day, with different lighting, with and without headdresses.
Therefore, we undertake almost everything that is interesting. Some things work out, some don't, but within the curriculum, this is normal.
Our undergraduates made a forecast for the development of the coronavirus pandemic in Primorye and the state of the post-covid economy in Russia. Both worked. They took information on the number of COVID-19 cases by country and date for the previous few days and offered the machine to process them. This is not an algorithm that we understand how it works, but training. There is data in and out, and by repeating simple selection operations many times, the result is obtained. The more numbers and the more accurate they are, the closer it is to reality. According to calculations, at the peak of the pandemic in Primorye, about 19 people should receive a COVID-260 diagnosis per day, in reality the figure was around 250. A small error, I think, is due to the peculiarities of statistics, where not all sick people get.
For the sake of fairness, I must say that the economic forecast, although it was generally correct, is still a little inaccurate. Among other things, we predicted a decrease in sales of new cars by 300 - 400 thousand per year in Russia, an increase in gasoline prices, the dollar rate to 75 rubles. Now we consider these projects completed, and we will definitely use the experience gained.
- If this goes on, then the machines will be able to replace the experts?
- Competent forecasts, as a rule, are made for a lot of money, but what we see and read in the public domain in the form of public speeches is of little value. There is also an interesting point, which I would also definitely note: experts very often change their minds, but with the machine, "all moves are recorded." With the passage of time, one can understand how the correct methodology was chosen for forecasting, "unscrew" it back and see how the neural networks would behave in other situations.
So, predictive computer models, despite the disadvantages, will occupy and already occupy their niches.
- Where else in the near future machines can they find application?
- One of the most popular areas is the generation of texts and leads - the first paragraphs of texts, where the essence of the text is collected. There are similar and well-implemented solutions. For example, Yandex collects from the mass of texts a kind of "Frankin news". It seems to me that this is not very interesting, because it is strongly tied to the primary sources, which may be a reprint of one press release.
It is much more interesting to take some data and generate news from them. So, we taught the car from a set of numbers - temperature, wind speed, pressure, the possibility of precipitation - to compose a coherent folding text about the weather. Now we turn to work on sports matches.
In the United States, there is already a commercial product for hockey - AI determines the situation on the field from video and, based on this, quickly creates a report. We have started to create a similar program, and it is quite difficult. But I think that such solutions will appear already this year, so we are waiting for the era of electronic commentators.
- Can the AI predict the outcome of the match?
- In such games, the role of accidents is high, and the computer must take into account many unknown factors - the injury of one of the players, which he may not know about, but which will affect his game, and the like. I would venture to suggest that artificial intelligence will quickly learn to calculate the likely scenario in the course of the game, but it is unlikely in advance. But if the result of the match can be predicted, then the series is unlikely, since the complexity does not increase multiple or exponentially, it becomes qualitatively different. Therefore, the main threat to AI has already repeatedly manifested itself - smart algorithms are quite predictable, they are guided, first of all, by the existing experience. Of course, there are certain manipulations. Suppose, during learning, one can "put" some non-existent options that, as we suppose, can happen so that the machine has them in mind.
But she still cannot take into account all the options, and she can do stupid things at a speed inaccessible to people. That is why it is undesirable to connect machines to vital systems.
- Otherwise, we are threatened with the implementation of the script for the film "Terminator"?
- I think a negative scenario is more realistic, when decisions that are vital or fatal for a person are made in a machine way without the possibility of their cancellation. As a social rating system in China. This is 100% a source of problems in the near future.
A huge number of experiments with ratings were carried out - by banks, insurance companies. But the human brain is sophisticated, and there are always followers of Ostap Bender, who read the law, who invent legal ways to deceive, receive unreasonable, from the point of view of the organizers, bonuses, thereby discrediting the system. So a person cannot be excluded from these chains, the final decision must remain with him - this time. Second, the consequences of the decision should not come too quickly, it is necessary to leave time to have time to react. And third, a balance must be struck between convenience and dehumanization, when a person is treated like a set of documents.
THE GENIUS OF MARKETING
- When we enter a request on the Internet, then we see similar advertising offers for a long time. Isn't that what AI does too?
- Of course. Marketing and sales is a colossal niche that will continue to grow. But now it works pretty roughly. A person wants to buy a car, goes to YouTube, watches some videos, and then a huge multicomponent car begins to rebuild in such a way as to show materials on the requested topic, thereby pushing the purchase.
There is no narrative yet, and this is just a restructuring of the information space. It is necessary to change the content of the message. Netflix has been trying for a long time to make a constructor from plots - when different people are shown different endings of series. Not surely cost effective in filmmaking, but may work well in terms of sales. For example, endings can be changed in technical reviews. And if set up well, people won't even guess.
But so far this whole system does not work perfectly, because the distribution of the commission to all those involved in the sale is not ideal. If everyone got their pretty penny, even if they contributed a month ago, things would have worked better already. Perhaps this is the case for the next 2-3 years.
- Where else can this approach be applied?
- Of course, in politics. The candidate of the future is just a passport. If a person's consumer behavior is predictable - either 100 years ago or now, he needed the fastest, most convenient and cheapest car, then political sentiment is unstable.
Now we are working on a paraphrase - when the machine returns your thoughts to you in other words. For English, this was done a long time ago, but with Russian, due to the peculiarities of the language, it is difficult.
Paraphrase is part of the overall mechanism. Given the impressions, beliefs, desires of people, AI will be able to design and sell a policy that everyone likes. In principle, this is a good old prompter, but much more technological. Imagine: there are many people in front of a politician. With the help of high-definition video surveillance cameras, the machine understands what age and social group they belong to, who they are, after analyzing facial expressions, draws a conclusion about their emotions and mood, and, based on the spectrum of data received, suggests the direction of the conversation.
In this sense, politicians are the easiest - as a rule, they already have several basic options in their heads, they need to choose the right one in time. But in general, this applies to any marketing - one and the same smartphone is sold as a media device for fans of photography, a device with a large screen for fans of TV series, and so on.
The problem is that political marketing is quite corrupt, therefore, innovations are the most difficult there, but such solutions will appear there over time.
POET, ARTIST, MUSICIAN
- Recently, the School of Digital Economics spoke about the creation of a digital poet who creates in the spirit of Vladimir Mayakovsky. Does he really write real poetry?
- Poems are the result of a huge number of experiments made by our graduate students, most of which went into marriage. Somewhere tenses are not agreed, the sentence structure is incorrect, the rhyme is lame. Mayakovsky's poems for teaching AI were chosen for several reasons, one of them is that his vast creative heritage is in the public domain, and besides, I am almost sure that the poet would be happy with such a turn.
But no robot can replace a real poet, but, for example, one who composes simple congratulations for anniversaries and holidays is easy. The project has achieved a certain result, but I would strengthen some things. Here's a small example
the east threw in one flaming vase
Take the cute ones, but only right away
The sheets of waters under the belly were
in the blizzards of midday dust
So someone wants them to be
They were ripped into waves by a white tooth
Rejoice louder or
love and lust copper pipes
the ground tramples squeaky and rough
The city's hell broke the windows
a mountain of gazes was thrown into the coffin
really not bored
There was a trumpet howling as if it was pouring
(punctuation and spelling of the author preserved - ed.)
In general, in my opinion, digital art is the most promising peaceful direction for using AI, and Mayakovsky is not our first creative synthetic personality. Undergraduates commissioned by Sber to work on a digital musician performing light pieces in the lounge style. It consists of two unequal parts - one teaches how to compose a neural network music, and the second reproduces it, sounding it with different instruments. The musician was trained on the basis of already existing melodies composed by people.
We have had exhibitions of paintings created by AI on several occasions. A new one is scheduled for April.
There is an AI that develops architectural solutions. A man does, of course, but he has never had such a clever brush. A work of traditional art is a hell of a job, and there is a monstrous amount of sketches or sketches behind the final result, but with the help of AI they can be done in a reasonable time.
Plus, we tried to work at the intersection of types of creativity - we started experimenting with the Youth Theater in the field of creating an artificial actor. This is a kind of interactive character that would help the creative team work on stage. How it will look is the task of the director and director. Unfortunately, due to the pandemic, the experiment had to be suspended, but I hope we will return to it.
- Does AI have prospects in the social sphere?
- Yes. Now we have switched to recognizing the language of the deaf and dumb. This is not a monetary, not a hype, but a demanded and humanly correct task. The first prototype has already gone to colleagues in Moscow for examination.
We are observing a not widespread, but quite noticeable transition to a contactless interface - where you used to press a button and pull a handle, you can often give a command by voice. Covid and related events spurred the transition. For many deaf people this is not very convenient. If in our country there are still remnants of the Soviet system of teaching speech, then in many other states they are not engaged in this, and people with hearing problems were cut off from such interfaces in public spaces.
Another closely related issue is neurodegenerative diseases. Arthur Biktimirov, a surgeon, works at the FEFU medical center, who has set himself the goal of creating a certain technological concept that makes it possible to understand how such diseases develop. While doctors see people suffering from parkinsonism with a delay, patients cannot go to an appointment every day or lie under observation in a hospital for years. We taught AI to recognize gestures, and now we are trying to understand the dynamics of the development of the disease based on the movement of the limbs.
ALL IS ALLOWED THAT IS NOT PROHIBITED
- What ethical questions arise with the increasing use of AI?
- One of the most difficult is deep mining (search and analysis of information - ed.). There is personal data: last name, first name, patronymic, medical history, and in general this information is protected by law. But there are those who definitely do not fall under the personal, but, being public in one form or another, can tell a lot about a person.
You go to some kind of online store. These platforms themselves, as it is spelled out in the agreements, absolutely legally collect information about you. The browser on your computer or phone knows what interests you. On the basis of such data, especially if you combine those obtained from different sources, it is possible, for example, to establish pregnancy, the presence of some, including stigmatizing, diseases, to conclude that a person is going to get divorced, get married. Analyzing this, at the right moment it is easy to slip the appropriate advertising - clinics, travel ... But here the critical question arises, how legal it is.
This is a very slippery story, and there are several approaches to it. The first, conditionally American - everything is bought and sold, everything is allowed. The second, conditionally European - with a whole range of strict prohibitions.
Last year our undergraduate student worked on an e-lawyer. Having large statistics, machine learning can be used to build mathematical models that make it possible to predict court decisions, to understand what factors affect it. So, monitoring of ships is prohibited in France. In Russia, de facto, this is still allowed.
But we, as a public organization, are not engaged in deep mining of individuals. Although in our country, as in the United States, there are already companies whose business is built on this.
- Where, in your opinion, AI should definitely be used?
- It is used in different ways in the world. There are experiments when people who have problems with movement and live alone are provided with an electronic interlocutor. These are different, very well-made characters. On the one hand, it is better than nothing, on the other - ersatz. You need to be careful with such things, because they can intensify the dehumanization of society. Meanwhile, AI can help find real people to communicate with.
You can fire journalists and put in a car that will generate news. Or you can help a journalist who writes on specific, narrow-profile topics using AI to find his audience.
Artificial intelligence should unite people, help them communicate, interact. I advocate the use of it for humanitarian purposes, and I hope that over time, this approach will dominate.