“Keep your hand on the pulse and look around” - an interview about AI with Intento co-founder Gregory Sapunov
The other day we decided to talk with our main teacher on the Deep Learning program , Grigory Sapunov, and discuss with him topical issues related to the field of artificial intelligence (AI). Grigory several years ago was the head of the development of Yandex.News. Currently he is a CTO and co-founder of Intento. For 15 years, he has been engaged in data analysis, artificial intelligence and machine learning, since 2011 he has been engaged in Deep Learning, participated in the projects RoadAR (neural network object recognition on the road), Icon8 (neural network filters), etc.
- I have long wanted to talk with you on the subject of artificial intelligence. You have been in this topic for a long time, you are well-versed and follow it, consult the n-th number of startups ... What did you initially attract about this topic?
- I have always been partial to cross-disciplinary topics, especially at the intersection of such interesting areas as programming, biology, mathematics, brain science, psychology, linguistics, philosophy. Artificial intelligence is just at the intersection of all these areas. There are a lot of difficult challenges here, there is a constant feeling that you are on the outskirts of the unknown, and with all of the rather long history of this trend, you still regularly find yourself in places where few have visited you before. It is very interesting.
')
- You started to get involved in this when it was not yet fashionable. What do you associate with the boom that we are currently experiencing with the development of AI? And is he really? Maybe this is another hype with high expectations?
- As usual, there is both a constructive history and a non-constructive one.
The constructive part is that in recent years there has been a breakthrough in working with neural networks. The paradox is that many of the ideas, algorithms and methods related to this breakthrough have been developed for a long time and are not a special novelty. About five years ago, the growth of computational capabilities and the emergence of large datasets successfully came together at one point. And then it turned out that the old algorithms, in general, work quite well, but before that they simply did not have enough resources to train them properly. Since then, quite a lot of conceptually new things have appeared, and in several areas (for example, speech recognition and computer vision), deep learning has been successfully completed and has supplanted traditional methods. Most books on AI older than five years are already dangerous to read because they are outdated. For example, they constantly exploit an example of a task that a computer is not capable of solving, but a three-year-old child can solve - to distinguish a cat from a dog. This has long been (several years as) not true. The computer perfectly recognizes the image, and does it with a higher quality person. Or just recently (a year or two ago) it was believed that the game of go would remain impregnable for another twenty years. In the spring of 2016 it became clear that this is no longer the case. Few expected. That is, in this place the boom is fully justified.
Moreover, there is a feeling that fantasy is not keeping pace with technology at the moment. The current developments are very universal, they can be used for very different tasks, and no one has yet thought of a multitude of such applications. Or thought of it, but have not yet devoted this necessary amount of time. I have my own shortlist of such tasks, so I’m waiting for myself to get to them, or someone will do it :)
HYIP with high expectations, unfortunately, is also present. And by the example of the history of AI, we already know that everything repeats and what it leads to. Even at the dawn of the birth of the AI field in 1956, one of the founders of the region (John McCarthy) stated that serious progress could be made in one summer, and a little later a steep person (Herbert Simon) later in 1965 stated that during 20 years of the machine will be able to do all that man can. And these high expectations have already led to “winters of artificial intelligence”, when everyone was disappointed by the lack of the promised results and covered the financing of AI projects. There is a risk of another winter now. Because along with the existing successes, there are still a lot of unresolved issues that remain beyond the brackets in the enthusiastic press publications. But as usual, I want to believe that this time everything will be different :)
- Many now fear that AI will destroy a lot of professions. What tasks can AI solve now at a high level? And who in the end should be wary of AI competition?
- Various international research groups and media already regularly release lists of professions ranked by risk of being replaced by computers. Some of these lists look weird, some quite plausible. You yourself can safely find a few of these on the Internet.
AI can already replace or strongly press people in a heap of those places where people were needed only to recognize visual or voice images. Or where people are forced to shovel vast amounts of information (this is a more native environment for computers).
I think the tragedy and the comic of the situation will be that first of all the AI will wash a lot of people from the same field and more widely from IT. At the moment, there are a large number of data scientists who deal only with the use of ready-made recipes and the search for parameters for models. This stupid non-creative work should and will be automated and different companies are already engaged in this (see, for example, FBLearner Flow ). Programming also has a huge number of tedious and boring activities that are long overdue to automate (see the wonderful words of Sassman, one of the SICP authors: “Programming today is more like science: you take a part of the library and poke into it - look at what it does. Then you ask yourself, “Can I set it up to do what I need?” The “analysis through synthesis” approach used in SICP when you build a large system of simple, small parts, became irrelevant. Today we are programming “At random”, habrahabr.ru/post/282986 ). By the way, in the current Intento project we aim at automating one of such programmer areas. Follow the announcements :)
- And if you dream for 5-10 years? The question may be urgent for those who are just entering the university.
“First of all, mediocrity will go everywhere.” Mediocre teachers, mediocre programmers, mediocre translators, mediocre lawyers. If you learn something, become the best. Well, do not be afraid of new technologies. The strength is still not in replacing a person with an AI, but in supplementing his AI. Those who will constantly improve and will be able to successfully complement themselves with the information technologies of the new century will never lose.
Some professions will certainly go into oblivion, but even if you are now studying for one of these professions, keep your finger on the pulse and look around. Part of the professions will go away, but how many new ones will appear, which we still do not guess? When horses disappeared from everyday life, a whole sea of new professions arose around cars and vehicles. One could even guess about these new professions by analogy, but when computers appeared, professions arose, many of which could not even be thought beforehand. If you are always open to new things, then without problems you will find a place in this new world.
- Now, by saying “artificial intelligence”, we almost immediately imply or are building the “deep learning” association. How justified is it at all? What else can be sewn inside the AI, besides deep neural networks?
- Deep learning (DL) supplanted all other AI methods in the media field and practically became synonymous with AI. But this, of course, is not true. There are a huge number of methods outside the DL. There are, for example, evolutionary computations and swarm intelligence methods capable of finding solutions to very complex optimization problems (often NP-complex). There are symbolic methods for representing knowledge and logical inference, which are very strong in the field of automatic reasoning and can, for example, explain how one or another conclusion was obtained. DL is hardly able to do this at the moment (the recent recent DeepMind publication about Differentiable neural computers opens the way for DL in this direction). There is an interesting area called Probabilistic Programming . Open any sane book about AI, there are a huge number of different methods in the table of contents.
I can advise a relatively recent my presentation about the current state of AI, which I did for schoolchildren in the framework of the GoTo School project. It is far from complete, but there are many examples of success other than DL approaches.
- Why do you think deep learning is good at solving many tasks that a person performs? Could something be thought of as an alternative?
“This is actually a hard question: why does DL work, and why does it work so effectively.” A parallel question in fact is why the human brain is capable of solving the same tasks well. Scientists are trying to explain this with the help of physics .
Apparently, there is some kind of commonality of principles, even though artificial neural networks are rather far from biological.
Complementary interesting question, what tasks a person performs poorly and is capable of performing well a machine (perhaps some kind of new architecture). In addition to the obvious quick counting and storage of large amounts of data. In this sense, I like Hamming's words very much: “I can’t like you. . “Then you can’t think,” you couldn’t be surprised. Evolution, so far, may be blocked from being able to think in some directions; there could be unthinkable thoughts. ” This is where AI can potentially revolutionize us.
It should be remembered that there are still many other tasks for which DL is not suitable, but other methods are suitable. See, for example, the character methods that I talked about earlier.
- According to your feelings, the market is now experiencing a shortage of specialists who are able to engage in the training of deep neural networks? What trend do you expect?
- The request for specialists is much more than they are available. I see this by the number of calls, the dynamics of open vacancies, and also by the presence of unplowed fields, where DL clearly has to make a revolution, as he did in sound and image processing.
From the trends, I still expect the washing out of the “surface” data scientists. If you decide to enter this area, do not stop after the first success, dig further. This is an area in which you need to constantly learn. If you are uncomfortable, DL / ML / AI is not for you. But what do you do in IT?
- I understand that this is one of the reasons why it is worth going to our educational program on deep learning. Tell us a little more about her - how do you see her? What is its use?
- The region is huge, you can not become a specialist in either two days or three months. I see my goal in providing participants with a basic understanding of the DL area, terminology and intuition, which is behind the basic methods, as well as a framework for orientation in this area, which they can then meaningfully fill in, based on their interests.
Another goal is to remove the fear of a new area and offer easy entry into it, teach fast practical things and show how you can build working solutions from almost ready-made components, or build your own using powerful libraries available.
Although the course is very rich and the amount of material is huge, I did not want to stretch this course for months, but to make it compact so that even very busy people could afford it. Time is a very important resource that I strive to optimize.
- What are the requirements for the person who is going to go for it?
- Interest in the field, the desire to understand, the ability to make time for it, as well as the basic ability to program in python, basic knowledge of linear algebra (vectors, matrices) and mathematical analysis (it is useful to be friends with derivatives, if you want to understand more deeply the essence of the processes, but learn how to work DL is possible without it, now all modern frameworks are able to independently make differentiation for all types of layers and no longer need to program it manually for a long time). Familiarity with the field of machine learning is also necessary (we will not re-tell the basics of machine learning, for example, we will not re-analyze in detail what logistic regression is, but restrict ourselves to a small reminder of important points).
“It sounds very interesting to me.” The last question is probably a classic for this topic. Is the uprising of machines threatening us? The community somehow adheres to different points of view. Some people think that cars can become smarter than us, but they will not threaten us. Others, including well-known technology companies, are concerned and want to somehow prevent possible threats.
“I’m not more concerned with the uprising of machines, but with human stupidity, greed, short-sightedness, lack of agreement and so on, which leads to permanent wars and conflicts.” By the standards of evolution, too little time has passed since the advent of civilization, we are still cavemen. I hope computers can compensate a little for our weaknesses. I believe in augmented intelligence.
The car is smarter than a person sooner or later appear. She reads this text with all the comments :) Such a machine will definitely change our world and I cannot predict how it will behave, nor to what it will lead. These are risks. But we are not indifferent players here, something depends on us. By this point you need to prepare.
But even earlier there are a lot of other risks from the fact that we are strongly tied to new technologies, not having time to properly understand them . We create complex systems (which can be very far from AI), about which we do not understand how they work . We do not think well about the risks of the technologies being introduced. We finally just use the new technologies as tools of our black business.
But besides the risks, I would consider potential gains. For example, in biology, where huge data arrays are being accumulated that cannot be embraced by the intelligent head of a single superexpert, or even by a group of experts. The potential of using this data is enormous - it is a better understanding of all aspects of the living cell work, a cure for many diseases and a radical extension of human life, it is a deeper understanding of the principles of the brain. In any other science there are also many problems and unsolved problems that could be solved by AI, from physics to history. It would be possible to radically change education by increasing its efficiency by orders of magnitude and transferring it to a completely new level. AI could help solve one of the worst "invisible" tragedies - the untapped potential of each person.
- I have long wanted to talk with you on the subject of artificial intelligence. You have been in this topic for a long time, you are well-versed and follow it, consult the n-th number of startups ... What did you initially attract about this topic?
- I have always been partial to cross-disciplinary topics, especially at the intersection of such interesting areas as programming, biology, mathematics, brain science, psychology, linguistics, philosophy. Artificial intelligence is just at the intersection of all these areas. There are a lot of difficult challenges here, there is a constant feeling that you are on the outskirts of the unknown, and with all of the rather long history of this trend, you still regularly find yourself in places where few have visited you before. It is very interesting.
')
- You started to get involved in this when it was not yet fashionable. What do you associate with the boom that we are currently experiencing with the development of AI? And is he really? Maybe this is another hype with high expectations?
- As usual, there is both a constructive history and a non-constructive one.
The constructive part is that in recent years there has been a breakthrough in working with neural networks. The paradox is that many of the ideas, algorithms and methods related to this breakthrough have been developed for a long time and are not a special novelty. About five years ago, the growth of computational capabilities and the emergence of large datasets successfully came together at one point. And then it turned out that the old algorithms, in general, work quite well, but before that they simply did not have enough resources to train them properly. Since then, quite a lot of conceptually new things have appeared, and in several areas (for example, speech recognition and computer vision), deep learning has been successfully completed and has supplanted traditional methods. Most books on AI older than five years are already dangerous to read because they are outdated. For example, they constantly exploit an example of a task that a computer is not capable of solving, but a three-year-old child can solve - to distinguish a cat from a dog. This has long been (several years as) not true. The computer perfectly recognizes the image, and does it with a higher quality person. Or just recently (a year or two ago) it was believed that the game of go would remain impregnable for another twenty years. In the spring of 2016 it became clear that this is no longer the case. Few expected. That is, in this place the boom is fully justified.
Moreover, there is a feeling that fantasy is not keeping pace with technology at the moment. The current developments are very universal, they can be used for very different tasks, and no one has yet thought of a multitude of such applications. Or thought of it, but have not yet devoted this necessary amount of time. I have my own shortlist of such tasks, so I’m waiting for myself to get to them, or someone will do it :)
HYIP with high expectations, unfortunately, is also present. And by the example of the history of AI, we already know that everything repeats and what it leads to. Even at the dawn of the birth of the AI field in 1956, one of the founders of the region (John McCarthy) stated that serious progress could be made in one summer, and a little later a steep person (Herbert Simon) later in 1965 stated that during 20 years of the machine will be able to do all that man can. And these high expectations have already led to “winters of artificial intelligence”, when everyone was disappointed by the lack of the promised results and covered the financing of AI projects. There is a risk of another winter now. Because along with the existing successes, there are still a lot of unresolved issues that remain beyond the brackets in the enthusiastic press publications. But as usual, I want to believe that this time everything will be different :)
- Many now fear that AI will destroy a lot of professions. What tasks can AI solve now at a high level? And who in the end should be wary of AI competition?
- Various international research groups and media already regularly release lists of professions ranked by risk of being replaced by computers. Some of these lists look weird, some quite plausible. You yourself can safely find a few of these on the Internet.
AI can already replace or strongly press people in a heap of those places where people were needed only to recognize visual or voice images. Or where people are forced to shovel vast amounts of information (this is a more native environment for computers).
I think the tragedy and the comic of the situation will be that first of all the AI will wash a lot of people from the same field and more widely from IT. At the moment, there are a large number of data scientists who deal only with the use of ready-made recipes and the search for parameters for models. This stupid non-creative work should and will be automated and different companies are already engaged in this (see, for example, FBLearner Flow ). Programming also has a huge number of tedious and boring activities that are long overdue to automate (see the wonderful words of Sassman, one of the SICP authors: “Programming today is more like science: you take a part of the library and poke into it - look at what it does. Then you ask yourself, “Can I set it up to do what I need?” The “analysis through synthesis” approach used in SICP when you build a large system of simple, small parts, became irrelevant. Today we are programming “At random”, habrahabr.ru/post/282986 ). By the way, in the current Intento project we aim at automating one of such programmer areas. Follow the announcements :)
- And if you dream for 5-10 years? The question may be urgent for those who are just entering the university.
“First of all, mediocrity will go everywhere.” Mediocre teachers, mediocre programmers, mediocre translators, mediocre lawyers. If you learn something, become the best. Well, do not be afraid of new technologies. The strength is still not in replacing a person with an AI, but in supplementing his AI. Those who will constantly improve and will be able to successfully complement themselves with the information technologies of the new century will never lose.
Some professions will certainly go into oblivion, but even if you are now studying for one of these professions, keep your finger on the pulse and look around. Part of the professions will go away, but how many new ones will appear, which we still do not guess? When horses disappeared from everyday life, a whole sea of new professions arose around cars and vehicles. One could even guess about these new professions by analogy, but when computers appeared, professions arose, many of which could not even be thought beforehand. If you are always open to new things, then without problems you will find a place in this new world.
- Now, by saying “artificial intelligence”, we almost immediately imply or are building the “deep learning” association. How justified is it at all? What else can be sewn inside the AI, besides deep neural networks?
- Deep learning (DL) supplanted all other AI methods in the media field and practically became synonymous with AI. But this, of course, is not true. There are a huge number of methods outside the DL. There are, for example, evolutionary computations and swarm intelligence methods capable of finding solutions to very complex optimization problems (often NP-complex). There are symbolic methods for representing knowledge and logical inference, which are very strong in the field of automatic reasoning and can, for example, explain how one or another conclusion was obtained. DL is hardly able to do this at the moment (the recent recent DeepMind publication about Differentiable neural computers opens the way for DL in this direction). There is an interesting area called Probabilistic Programming . Open any sane book about AI, there are a huge number of different methods in the table of contents.
I can advise a relatively recent my presentation about the current state of AI, which I did for schoolchildren in the framework of the GoTo School project. It is far from complete, but there are many examples of success other than DL approaches.
- Why do you think deep learning is good at solving many tasks that a person performs? Could something be thought of as an alternative?
“This is actually a hard question: why does DL work, and why does it work so effectively.” A parallel question in fact is why the human brain is capable of solving the same tasks well. Scientists are trying to explain this with the help of physics .
Apparently, there is some kind of commonality of principles, even though artificial neural networks are rather far from biological.
Complementary interesting question, what tasks a person performs poorly and is capable of performing well a machine (perhaps some kind of new architecture). In addition to the obvious quick counting and storage of large amounts of data. In this sense, I like Hamming's words very much: “I can’t like you. . “Then you can’t think,” you couldn’t be surprised. Evolution, so far, may be blocked from being able to think in some directions; there could be unthinkable thoughts. ” This is where AI can potentially revolutionize us.
It should be remembered that there are still many other tasks for which DL is not suitable, but other methods are suitable. See, for example, the character methods that I talked about earlier.
- According to your feelings, the market is now experiencing a shortage of specialists who are able to engage in the training of deep neural networks? What trend do you expect?
- The request for specialists is much more than they are available. I see this by the number of calls, the dynamics of open vacancies, and also by the presence of unplowed fields, where DL clearly has to make a revolution, as he did in sound and image processing.
From the trends, I still expect the washing out of the “surface” data scientists. If you decide to enter this area, do not stop after the first success, dig further. This is an area in which you need to constantly learn. If you are uncomfortable, DL / ML / AI is not for you. But what do you do in IT?
- I understand that this is one of the reasons why it is worth going to our educational program on deep learning. Tell us a little more about her - how do you see her? What is its use?
- The region is huge, you can not become a specialist in either two days or three months. I see my goal in providing participants with a basic understanding of the DL area, terminology and intuition, which is behind the basic methods, as well as a framework for orientation in this area, which they can then meaningfully fill in, based on their interests.
Another goal is to remove the fear of a new area and offer easy entry into it, teach fast practical things and show how you can build working solutions from almost ready-made components, or build your own using powerful libraries available.
Although the course is very rich and the amount of material is huge, I did not want to stretch this course for months, but to make it compact so that even very busy people could afford it. Time is a very important resource that I strive to optimize.
- What are the requirements for the person who is going to go for it?
- Interest in the field, the desire to understand, the ability to make time for it, as well as the basic ability to program in python, basic knowledge of linear algebra (vectors, matrices) and mathematical analysis (it is useful to be friends with derivatives, if you want to understand more deeply the essence of the processes, but learn how to work DL is possible without it, now all modern frameworks are able to independently make differentiation for all types of layers and no longer need to program it manually for a long time). Familiarity with the field of machine learning is also necessary (we will not re-tell the basics of machine learning, for example, we will not re-analyze in detail what logistic regression is, but restrict ourselves to a small reminder of important points).
“It sounds very interesting to me.” The last question is probably a classic for this topic. Is the uprising of machines threatening us? The community somehow adheres to different points of view. Some people think that cars can become smarter than us, but they will not threaten us. Others, including well-known technology companies, are concerned and want to somehow prevent possible threats.
“I’m not more concerned with the uprising of machines, but with human stupidity, greed, short-sightedness, lack of agreement and so on, which leads to permanent wars and conflicts.” By the standards of evolution, too little time has passed since the advent of civilization, we are still cavemen. I hope computers can compensate a little for our weaknesses. I believe in augmented intelligence.
The car is smarter than a person sooner or later appear. She reads this text with all the comments :) Such a machine will definitely change our world and I cannot predict how it will behave, nor to what it will lead. These are risks. But we are not indifferent players here, something depends on us. By this point you need to prepare.
But even earlier there are a lot of other risks from the fact that we are strongly tied to new technologies, not having time to properly understand them . We create complex systems (which can be very far from AI), about which we do not understand how they work . We do not think well about the risks of the technologies being introduced. We finally just use the new technologies as tools of our black business.
But besides the risks, I would consider potential gains. For example, in biology, where huge data arrays are being accumulated that cannot be embraced by the intelligent head of a single superexpert, or even by a group of experts. The potential of using this data is enormous - it is a better understanding of all aspects of the living cell work, a cure for many diseases and a radical extension of human life, it is a deeper understanding of the principles of the brain. In any other science there are also many problems and unsolved problems that could be solved by AI, from physics to history. It would be possible to radically change education by increasing its efficiency by orders of magnitude and transferring it to a completely new level. AI could help solve one of the worst "invisible" tragedies - the untapped potential of each person.
Source: https://habr.com/ru/post/314160/
All Articles