Multi-agent system communication is an innovative technology, which provides means for interacting intelligent agents to establish communication with each other and with elements of the external environment. Multiple agent systems are essential in solving problems that are hard for solving by individual agents. Multiple agent communication technology can be utilized in the organization and management of cloud computing and can act as a distributed, universal operating system.

[A team of developers from Moscow's Institute of Physics and Technology has just published a paper ](https://arxiv.org/ftp/arxiv/papers/1702/1702.08529.pdf) that presented a novel technology that combines decentralized P2P BOINC general purpose computing tasks distribution, multiple agents communication protocol and a rewarding system based on Ethereum's smart contracts. The proposed system can be utilized to act as a decentralized P2P market for computing power that is shielded from manipulation by central authorities. This newly proposed decentralized market can be updated to determine the most efficient combination of software and hardware. When the system can optimize hardware/software efficiency, it can then be updated to general purpose decentralized intelligence, which would act as combination of single purpose Artificial Intelligence (AI). 

## What are multi-agent systems?
Solving a problem by a single agent, via knowledge engineering, represents a point of view of classic AI. The agent, e.g. the intelligent system, that has an overall vision of the problem, possesses all the needed knowledge, skills and resources to present a solution to the problem. Oppositely, when building multi-agent systems (MAS),it is assumed that a single agent usually has only a partial understanding of the task in question and can solve only a portion of its subtasks. Accordingly, in order to solve any given complex problem, typically, the interaction of agents is mandatory, which is an important characteristic of the framework of MAS. Within the context of MAS, tasks are distributed among the agents; each of these is considered a member of the collective group of agents. Task distribution includes assignments of roles for each and every agent and definition of means of measurement of its specific responsibility and experience requirements. 

## BOINC : A system of decentralized artificial intelligence:

BOINC, or Berkeley Open Infrastructure for Network Computing, is an open source platform and a non-profit organization for decentralized computing. BOINC is used for volunteer organization of computing. As shown on the below figure, the architecture of BOINC is designed on the basis of the concept of decentralized AI. Its server is comprised of a group of individual agents, or subsystems, each of these handles well defined tasks, e.g. performing calculations, file transfer....etc. Each one of these agents checks the status of its predefined subtask(s), produces certain actions and modifies the state of subtasks. Generally speaking, the system is comprised of BOINC server, a large number of clients, which perform the tasks of the server and additional elements in the form of GRID affiliated networks. 

https://i.imgsafe.org/d0754ba1e0.png
 
## An overview of technical implementation of the system:

An agent within the AI system is usually referred to as a bot, softbot, chat bot, tobot...etc, and is capable of performing actions in the external environment and most importantly, is capable of communicating with other bots and alter their behavior. Multi-agent systems are comprised of thee agents, which are sometimes not artificial. The authors of the paper found out that the main goals of developing multi-agent systems are:

•	Swarm intelligence
•	Decentralized intelligence
•	Distributed intelligence

Note that there is a difference between decentralized and distributed AI. The concept of distributed task solving postulates demarcation between resources and knowledge among the agents, and to a lesser extent, distribution of power and management. As a rule, it proposes the existence of a governing body that is responsible for decision making whenever there is a critical situation or a conflict. 

Swarm intelligence is comprised of a group of simple reactive agents that can only perform very simple functions. On the other hand, as a system, swarm intelligence is capable of solving a limited number of tasks, even though it is decentralized. 

Nevertheless, the distributed intelligence, within the BOINC framework, is capable of solving a variety of tasks and it is presently used in different areas, but at the same time, it is typically centralized and strictly limited by the design pattern.

Decentralized intelligence SONM uses BOINC for integration and distribution of the tasks, as well as smart contracts, via Ethereum's platform, for decentralization.