350 rub
Journal Neurocomputers №6 for 2016 г.
Article in number:
Artificial intelligence and the problem of multi-agent systems control
Authors:
M.A. Shestakova - Ph.D. (Philos.), Associate professor, Department of Philosophy and Methodology of Science, Faculty of Philosophy, Lomonosov Moscow State University. E-mail: m.a.shestakova @yandex.ru E.I. Shestakov - Post-Graduate Student, Department of Management Problems, Moscow State Institute of Radio Engineering, Electronics and Automation (MIREA). E-mail: shestakov.ei@gmail.com
Abstract:
The problem of artificial intelligence has many aspects, so there exist many approaches to its solution. One of them may be called functional. From this standpoint, artificial intelligence is not regarded as an \"artificial brain\", but rather as a set of au-tonomous functions whose integration into a single system is not compulsory. Let us clarify this position by using an analogy with an artificial body. Today, the task of creating an artificial body is being solved at the level of individual organs: works are being held to develop artificial hearts, kidneys, skin, etc. The purpose is not so much to reproduce the organ itself, but to reliably replace certain functions of the human body. The artificial heart may not look like a natural one, but it must perform well its function. At the same time this artificial organ can perform its function better than the natural one. For example, artificial skin can have a number of properties which are weakly expressed or absent in human skin. Similarly, the artificial intelligence can be represented as a set of individual intellectual functions, especially of those which can be performed better by artificial systems than by the natural human mind. Work in this direction is underway for a long time, for example, computing function has been being automated. At the same time the computer\' processing capabilities are many times greater than those of an individual human intellect. Another intellectual function the individual mind cannot quite cope with is, oddly enough, the social management. Management decisions, as a rule, require a collective effort in order to collect and analyze information, to explain decisions to the executors, etc. In this regard, the task to automate exactly this function of the human intellect becomes the agenda. The solution of this problem makes sense if the artificial system presents a clear advantage, similar to the computational machines. In this respect, attention should be paid to the works on multi-agent robotic systems (MARS). Theoretical problems arising in the development of MARS are similar to those in social systems management. MARS can be regarded as a prototype of artificial intelligence interpreted functionally, namely in its the social management function. MARS is an organization of intelligent autonomous robots pursuing the execution of a common task. The attractiveness of using MARS for the development of a functional model of artificial intelligence lies in the type of tasks and algorithms for solving them. MARS are oriented towards the execution of applications that require the combined efforts of several agents. In this case, we are talking mainly about the problems that a single agent cannot solve, thus requiring the collective efforts of several robots. In other words, MARS are focused on tasks that have an emergent effect. Typologically, these are the same tasks as in the field of social management systems. The following are considered as the main advantages of MARS: ability to distribute execution of processes among several agents in order to reduce those or other costs; possibility to substitute complex and expensive systems for a group of more simple and cheap robots. In this case, we can say that MARS use the simplest benefits of social systems: distribution of responsibilities between individual executors. In this case, we can say that Mars used the simplest benefits of social systems - distribution of responsibilities between individual performers. The main theoretical problem of MARS is the development of effective methods and algorithms of multi-robot control. The following possibilities are considered: central (hierarchical, leader-followers) control and decentralized (collective, swarm) control. Here it is worth paying attention to the fact that the development of MARS control models may be based on historical experience of social systems management. The most common are centralized (hierarchical) models of social management. Social experience reveals not only their advantages but also their serious disadvantages. Therefore, the most promising direction is to develop mixed and decentralized control models. It is exactly in these areas that automated systems should strengthen the natural human mind\'s capacities. Improving mixed and decentralized models of MARS control is a strategic direction in the creation of artificial intelligence, in terms of functionality. The following theoretical problems of decentralized control of MARS are of great relevance in today\'s research: 1). Improving the quality of decision-making (decentralized control systems need less time to make decisions, but also display poorer quality of decision-making) [1]. 2). The optimal synthesis of decentralized and centralized approaches in the combined control strategies. The success of a combined strategy depends on the decision conditions and peculiarities of the problem. This approach presents the problem of the optimal synthesis of decentralized and centralized approaches depending on the conditions of the problem and the number of executors, and its recombination in case of conditions and/or participants change [2]. In conclusion, we note that the problem of artificial intelligence requires a flexible approach in determining artificial in-telligence. This condition is met by the functional approach presented above. Development of control functions based on ap-proaches and results in the field of MARS present good perspectives in the framework of this approach.
Pages: 50-52
References

 

  1. Kaljaev I.A., Gajjduk A.R., Kapustjan S.G. Modeli i algoritmy kollektivnogo upravlenija v gruppakh robotov. M.: Fizmatlit. 2009. 278 s.
  2. Makarov I.M., Lokhin V.M., Manko S.V., Romanov M.P., Aleksandrova R.I. Smeshannye strategii gruppovogo upravlenija v mnogoagentnykh robototekhnicheskikh sistemakh // Izvestija JUzhnogo federalnogo universiteta. Ser. Tekhnicheskie nauki. 2012. № 3. C. 8-13.