Research in the field of artificial intelligence today is at the forefront of science and occupies one of the important places in the work of scientists at the National Research Nuclear University MEPhI (NRNU MEPhI). What developments are underway in this area today? What is behind the desire of scientists to teach the machine how to create? Professor of the Institute of Intelligent Cybernetic Systems (NRNU MEPhI) Aleksey Samsonovich told the correspondent of the Social Navigator project of the Russia Today MIA during the XX International Scientific and Technical Conference "Neuroinformatics-2018".
Alexey Vladimirovich, some time ago you talked about the creation of the composer's virtual assistant. Are you attracted to modeling creative thinking?
- In general, I am attracted by the task of creating artificial intelligence of a human level, in particular, emotional intelligence. This is necessary in order to make future virtual agents compatible with humans. People do not perceive something as alive if it is unable to understand emotions and express emotions.
For artificial intelligence to be successful in general, it needs to be compatible with humans. They should see him as an equal partner and not as a means to achieve goals. This is possible only on the basis of mutual understanding.
And the fact that at the moment I am focused on creativity is, rather, an intermediate step. The ultimate goal for me is not creativity or artificial intelligence doing art (although this is certainly an important and interesting task). This is just a link in a chain of steps towards human-level artificial intelligence.
Qrio robots dance Argentine tango in Tokyo, Japan. 2005 year.
First, we needed to demonstrate the indistinguishability of artificial intelligence from humans in a limited environment, which we did one and a half to two years ago. Then it was necessary to move on, still remaining within the "toy" tasks. We set a goal not just to pass the Turing test (a test to determine the "intelligence" of a machine - ed.), But to create a kind of creative product - for example, a work of art - jointly by humans and artificial intelligence.
A lot has already been done in this direction in the world today, and it would be difficult to surpass these achievements. But we are not trying to do this, because our approach is completely different, it is based on cognitive architecture, and not on neural networks, like most other researchers. There, the goals are achieved virtually without any understanding of the mechanisms of the human creative process, which does not allow the created artificial intelligence to move beyond what it was created for. And our approach is based on the fact that solving a specific problem, we find out the general principles and develop an understanding that allows us to solve the following problems.
Promotional video:
- Do you plan to create a virtual artist?
- I gave several students such a task. Our idea was simple. At the beginning of the 20th century, the famous artist, one of the founders of abstract art, Piet Mondrian, worked. Many of his highly acclaimed paintings are rectangles, broken up into many smaller rectangles of different colors.
I proposed the following: create an artificial intelligence that will re-color reproductions of Mondrian's paintings, "not knowing" anything about the coloring of the originals. And offer the subjects two sets of pictures so that they do not know where what is, and compare which is better. Who knows, maybe our artificial intelligence will win? We have previously created a semantic color map to distinguish pleasant from unpleasant colors.
- When you teach artificial intelligence everything, and it enters into emotional and intellectual contact with a person, how can it be used?
- I think it will be used everywhere. In three years, or even earlier, all intellectual means, even the operating system of a computer or smartphone, will have elements of emotional intelligence. Smartphones already now allow us to convey emotions in messages, and even recognize them on a person's face. I don't know how useful it is, but maybe people like it.
I am somewhat pleased that everyone is obsessed with deep neural networks, because if they, like me, were carried away by cognitive architectures, I would have to do something else.
- Are you not afraid that an emotionally trained machine at some point will begin to show the wrong emotions that are expected of it, and say: “I am not in the mood to work”?
- Believe it or not, but in the West, researchers set just such a goal - to create intelligent agents who are able to rebel against their master. This is believed to be beneficial because otherwise they will not be able to refuse stupid tasks.
When I first heard about this from my boss four years ago, I asked, "How is this different from an elevator that doesn't move when you press a button?" He laughed and said, "I hope this does not happen in our building."
- Is there any danger here?
- No, at least not related to artificial intelligence. The danger lies elsewhere - all these giants like Google and Microsoft will force us to think the way their neural networks think, and we will lose the ability to think like humans. But it will be the handiwork of humans, not artificial intelligence.
- Do you expect any pleasant discoveries for humanity in this area in the coming years?
- Yes of course. I think many are now waiting for a "big break" in artificial intelligence. In my opinion, for this, first of all, a person's attitude towards artificial intelligence must change. Recently, at a lecture, I was asked: "Why does artificial intelligence need emotions if we only need it to best solve the task set before it?" I believe that it is precisely in this formulation of the question that the mistake lies: artificial intelligence must itself find the problem that needs to be solved, and must set it itself. At the same time, he should be perceived by us as an intellectually equal partner, with whom one can interact as an equal, on the principles of trust and mutual understanding.
While under human control, artificial intelligence still has to operate largely independently. Or it must be a kind of human supplement, his assistant, or an extension of the human brain.
