Welcome!
Восстановление пароля
Введите адрес электронной почты или телефон, указанные при регистрации. Вам будет отправлена инструкция по восстановлению пароля.
Некорректный формат электронной почты или телефона
Is (Artificial) Intelligence Possible Without Mathematics?
8 June 2019
10:00—11:15
KEY CONCLUSIONS
AI capabilities are reaching human level and being superior in some areas

If a human being can do something, we are absolutely sure that a machine will ultimately be able to do it too, especially if it is a very specific task — Alexander Kraynov, Head of Computer Vision and Artificial Intelligence Technologies, Yandex Group of Companies.

Artificial intelligence is now capable of speech recognition, image recognition, natural language processing, but all of it is based on the same general principles that we see in our brains. Not the details, but the general principles — Terrence Sejnowski, Professor, Laboratory Head of the Computational Neurobiology Laboratory, Salk Institute for Biological Studies; Distinguished Professor, The Biological Sciences at University of California San Diego.

Talking about computer vision, computers are now better at recognizing humans than humans themselves. Humans are unable to recognize a criminal in disguise, whereas software does it with no effort. <…> I think artificial intelligence will be similar to ours in many ways, but not identical, because of hardware limitations, and other limitations, but it will in many ways be superior to the human brain. Performance of niche-specific tasks shows that it is already true — Artem Yamanov, Senior Vice President, Business Development Director, Tinkoff Bank.

Science soon to become impossible without artificial intelligence

Science today, and mathematics in particular is, of course, still possible without AI. But it will become impossible in the years to come. It is the same as the arrival of computers transformed the work of scientists. Advances in AI will change the work of a scientist in the same manner. It is actually already partially happening — Dmitriy Vetrov, Research Professor, National Research University Higher School of Economics; Head of Machine Learning, Artificial Intelligence (AI) Center in Russia, Samsung.

Artificial intelligence enhances cognitive capabilities of humans

We should think of AI now as tools, tools that will enhance our cognitive capabilities. And not just scientists, but also doctors, engineers, lawyers. Help us do our job better — Terrence Sejnowski, Professor, Laboratory Head of the Computational Neurobiology Laboratory, Salk Institute for Biological Studies; Distinguished Professor, The Biological Sciences at University of California San Diego.

ISSUES
Insufficient understanding of AI principles

We must understand what we mean by artificial intelligence. There is strong artificial intelligence and weak artificial intelligence. The strong AI is something that does not need our full support to develop; that is, we provide some starting points and then the system begins self-improving. The weak AI is something that serves as an extension of a human body, as our tool. <…> If we want to co-exist with this new reality, it is important to understand from the very beginning what rules and what laws will govern this new reality. We need it, so that we do not find ourselves in a complicated situation which will require all our energy and resources to neutralize it. It is an issue of our security — Andrey Fursenko, Aide to the President of the Russian Federation.

As is often the case, technology starts working before we understand all details, all nuances of this technology. Therefore, unawareness of all details, of how the technology functions, inevitably leads to various incidents. It happened, for example, in the nuclear energy industry. <…> I am afraid that similar things are inevitable in AI. <…> Today, the technology is of course raw. <…> We do not fully understand how one or another technology functions and why a computer is capable of producing way more information from the data than we have expected — Dmitriy Vetrov, Research Professor, National Research University Higher School of Economics; Head of Machine Learning, Artificial Intelligence (AI) Center in Russia, Samsung.

Without understanding [the principles of AI, – Ed.], we cannot speak about actual security, especially in critical areas — Arutyun Avetisyan, Director, Institute for System Programming of the Russian Academy of Sciences.

If you take a look at the best research that is being done now, it is all focused on interpretations of outcomes, because consumers do not want answers from a black box, they are just not ready to accept it — Stephen Brobst, Chief Technology Officer, Teradata .

SOLUTIONS
Awareness of technology deployment risks

Today, taxis in Moscow, and in St. Petersburg, do not drive without GPS. <…> We should realize another thing: one day it all would shut down. A research in the US shows that 90 percent of Americans would not be able to find their way home. <…> In Moscow the number would be from a third to a half, I think — Andrey Fursenko, Aide to the President of the Russian Federation.

We have to think about the consequences now, because it will be too late when we are rolling out the AI — Terrence Sejnowski, Professor, Laboratory Head of the Computational Neurobiology Laboratory, Salk Institute for Biological Studies; Distinguished Professor, The Biological Sciences at University of California San Diego.

Development of AI research methods

We need some new mathematical models including those that account for security — Arutyun Avetisyan, Director, Institute for System Programming of the Russian Academy of Sciences.

We need new science, new mathematics. We need science in which the results will be clear not only to scientists, but to general public, because we see new technology and we want to understand how it works — Stanislav Smirnov, Fields Laureate; Professor, University of Geneva.

We cannot improve the performance of neural networks without mathematics. We cannot use them to solve practical problems — Terrence Sejnowski, Professor, Laboratory Head of the Computational Neurobiology Laboratory, Salk Institute for Biological Studies; Distinguished Professor, The Biological Sciences at University of California San Diego.

If you do not provide explanations, people will not accept the result. Therefore, we need to know how AI functions from the mathematical point of view — Stephen Brobst, Chief Technology Officer, Teradata .

Perhaps the most important thing now is to reduce the role of humans in machine learning, that is we need technology aimed at solving this problem. <…> Not only mathematic, but also engineering approaches — Arutyun Avetisyan, Director, Institute for System Programming of the Russian Academy of Sciences.

There has to be as little human as possible, because it slows down the research, makes it not as reliable and fast as you want. <…> Almost all algorithms used in AI are biased. That bias comes from humans — Stephen Brobst, Chief Technology Officer, Teradata .

Programming AI to explain its decisions

An important goal is to get neural networks not only to make a decision, but to make them provide an explanation why the decision was made that humans would understand — Dmitriy Vetrov, Research Professor, National Research University Higher School of Economics; Head of Machine Learning, Artificial Intelligence (AI) Center in Russia, Samsung.

Pointing out technology that is critically important to understand

There are products for which it is not important how they work, whereas it is the result that is important. For example, music recommendations. I do not think you would be interested in the algorithm. <…> If we take something serious, say, healthcare. You do not want to bring your test results to the doctor who would upload them to a black box and the black box would provide a diagnosis that you have not expected. <…> You would demand an explanation. I think, you should always divide products into those you can experiment with without thinking so much about mathematics, and those whose functioning must be transparent — Elena Bunina, General Director, Director of Organizational Development and HR Management, Yandex Russia .

Banks have different tasks. They have technological problems that are very similar to those that tech companies try to solve. Those include computer vision, speech recognition or speech synthesis. For us, bankers, it is not really important how it works, the most important thing is the result — Artem Yamanov, Senior Vice President, Business Development Director, Tinkoff Bank.

The material was prepared by the Russian news agency TASS