Microservice-Oriented Architecture in Distributed Artificial Intelligence Systems and the Language of AI in Bio-Neural Systems

Introduction

When it comes to solving of a complex problem, it is often necessary to break it down into subtasks that are assigned to individual agents. Several agents can consider the problem from different points of view and then combine the results. In particular, the functional distribution of application programs makes it possible to overcome a number of shortcomings of classical expert systems. In them, the centralization of knowledge in a single knowledge base gives rise to problems of completeness and consistency. Moreover, the addition of new knowledge often leads to violations of the coherence of knowledge. On the contrary, an agent in distributed artificial intelligence can be considered without taking into account the characteristics of other agents, and the problem of the consistency of knowledge gives way to the tasks of ensuring cooperation and communication of agents. In many cases, the physical distribution of the task is also required, for example, in the case of using a group of robots.

A typical scheme of distributed solution of tasks by several agents includes the following steps (Figure 1):

Figure 1.

Flowchart

• 1.

  The subordinate agent (manager, central authority) decomposes the original problem into separate tasks.

• 2.

  These tasks are distributed among executing agents.

• 3.

  Each executing agent solves its task, sometimes also dividing it into subtasks.

• 4.

  To obtain a general result, composition is made, integration of particular results corresponding to the identified tasks. The integrator agent is responsible for the overall result (most often, this is the same subordinate agent).

The two most important aspects of distributed artificial intelligence are the distribution of tasks between agents and the pooling of results. So at the stage of decomposition, a single agent can split the task into sub-tasks, but is not able to find their solutions due to limitations on experience and resources. There is a situation of distribution of tasks. After obtaining private results, the problem arises of their coordination and integration.

Here, the main criteria for the effectiveness of a distributed solution to a problem are the time of the solution and the compliance of the subtask with the capabilities of a particular agent. If there is some inconsistency, the executing agent can further dissect the task, seek help from other implementing agents, etc.

In the case of a distributed solution to the problem, the agent-subordinate can resort to two opposite strategies:

• •

  selection of executing agents most suitable for solving specific subtasks (selection of an agent for solving subtasks);

• •

  selection of the most suitable subtask for the executing agent (selection of the subtask for the agent).

Thus, the fundamental features of a social group in distributed artificial intelligence, i.e., a group composed of artificial agents collaborating to achieve a common goal, are the social structure and distribution of roles between agents. In this case, the social structure is formed as a result of the appointment of roles. When an agent gets his role, restrictions such as “permission” and “responsibility” are imposed on his activities. In accordance with this approach, an organization is created and developed “from within” the system. The global structure is formed by integrating elements directly related to agents. The whole community with its social structure functions because agents play certain social roles, which leads to the achievement of social goals.