Specialists in the field of creating human-level artificial intelligence today have to solve not only scientific, but also ethical problems. What are the challenges facing scientists? How to protect humanity from potential threats posed by the creation of a superpowerful mind, and remain human in the new world? Professor of the Institute of Intelligent Cybernetic Systems of the National Research Nuclear University "MEPhI" (IIKS NRNU MEPhI) Alexey Samsonovich told about this to the correspondent of the project "Social Navigator" of MIA "Russia Today" Anna Kurskaya.
- Alexey Vladimirovich, our conversation is taking place within the framework of the International Conference "Biologically Inspired Cognitive Architectures" BICA 2019. What topics of discussion today come to the fore in connection with the creation of artificial intelligence at the human level?- The conference allowed us to once again realize the main challenges in the field of creating next generation artificial intelligence, which should become an equal partner and friend of man. First, the challenge is to create human-level emotional intelligence that can harness emotions in decision-making and behavior generation. In particular, this quality will allow the machine to establish long-term social ties, similar to those between loved ones. It will also allow her to justify her actions and decisions, choose goals and a way of learning. Emotional intelligence is the key to answering other challenges. It is not limited to the ability to recognize and express emotions, which today are occupied by most researchers working in this field. Although both of these processes are necessary,it is much more important and more difficult to understand the connection between one and the other, which is provided by emotional intelligence. Secondly, the intelligent agent must understand the context of what is happening. This means that, on the basis of the information available to him, he must independently generate and continuously update the global scenario of the past and the future, determine his role in it, and then, on this basis, set specific goals for himself and develop plans to achieve them. In the scientific literature, this ability is usually called “narrative intelligence” (it should not be confused with programs that work according to a script or narrative previously written by a person).that he must independently generate and continuously update the global scenario of the past and the future on the basis of the information available to him, determine his role in it, and then, on this basis, set specific goals and develop plans for achieving them. In the scientific literature, this ability is usually called “narrative intelligence” (it should not be confused with programs that work according to a script or narrative previously written by a person).that he must independently generate and continuously update the global scenario of the past and the future on the basis of the information available to him, determine his role in it, and then, on this basis, set specific goals and develop plans for achieving them. In the scientific literature, this ability is usually called “narrative intelligence” (it should not be confused with programs that work according to a script or narrative previously written by a person).or narrative).or narrative). - Is there a third challenge?
- It will be the implementation of active learning of artificial intelligence, which, like a person, develops a system of values that motivates the learning process, itself forms goals and a roadmap for learning, and chooses means (for example, experiments to test hypotheses, search for available resources or questions to a person), and, driven by his own initiative, uses these means to achieve the goal. At the same time, the machine must be able to independently develop from the level of a child to the level of an adult and further, to superintelligence.
This third area could be called human-level learning intelligence to distinguish it from deep learning and other types of machine learning that are popular today. In my understanding, human-level learning intelligence will be possible through emotional and narrative intelligence.
What research are you doing now? How is your work on the creation of virtual human assistants and emotional intelligence modeling progressing?
- We are engaged in modeling human emotional intelligence with the aim of developing socially compatible virtual human assistants based on our models.
We consider a wide range of specific tasks and research areas. For example, this is the creation of a virtual partner in a video game; virtual coordinator or moderator; virtual creative assistant to the composer, choreographer, designer, assistant to the insight problem solver; a virtual listener (in the future - a virtual interlocutor); as well as a virtual pet, comparable to a real one in terms of its social and emotional characteristics.
The results are difficult to present in detail here, but they can be found in our open publications. For all the breadth of our attempts and successes, we need a breakthrough in one direction: in proving the practical value of the general model of human emotional intelligence, embodied in an intellectual agent.
Recently, our BICA Lab at the Department of Cybernetics at NRNU MEPhI has been enriched with new research methods, including an experimental platform based on virtual and mixed reality (with tracking the direction of the subject's eyes), as well as electromyography and automatic analysis of facial expressions, allowing us to register emotions of subjects immersed in virtual environments. Most of our work is now done using these technologies. EEG and fMRI are also used in some experiments.
Promotional video:
Our immediate overarching goal is to refine and empirically validate the Ebica Socio-Emotional Cognitive Architecture (eBICA) and create effective demonstrations based on it. This work is being carried out at NRNU MEPhI with financial support from the Russian Science Foundation grant No. 18-11-00336.
Recently, Elon Musk's startup Neuralink announced developments that, in the future, could make a person a cyborg, capable of controlling a computer with the power of thought. How promising is this line of research to you?
- In my understanding, artificial intelligence should become, on the one hand, an equal partner and friend of man, and on the other, an extension of normal or replacement of lost human capabilities, including means based on a direct interface with the human brain.
In this case, a person could literally use a computer or microchip as part of their brain. Both are aimed at the good and should be entirely controlled by its creator. I wish Elon Musk every success.
How big is the need to develop means of countering artificial intelligence?
- I think that in our era, artificial intelligence tools will remain under the control of a person, if only because of their dependence on their service by a person and providing them with everything they need. If anyone can turn them against a person, then only the person himself. Then the issue of confronting those people should be considered.
It's no secret that the military needs artificial intelligence first of all. Today, as before, new means of attack and defense are being created. But the peculiarity of our time is that they both include artificial intelligence. In this sense, the need to develop means of confrontation is great and for some countries it is a priority from a financial point of view.
What awaits us after the creation of artificial intelligence, superior to human? Will we coexist with machines or will we unite into some kind of a single whole?
- Artificial intelligence will become, and is already becoming, an integral part of human civilization. At the same time, two ways are still possible: the inheritance of our values, aspirations and ideals by artificial intelligence, or their replacement with "machine values", which extends not only to machines, but also to people. Which of these two paths is realized is up to us.
What does it mean to remain human in the era of artificial intelligence?
- To remain human in the era of artificial intelligence means to implement the first path.