Towards an Army of Automata?

Do you think it’s still the distant future the day in which an army be consisted of just automata, each of them knowing what to do in every moment and what its final target is? Or a fire brigade of robots –not fearing the fire, which can perfectly move inside a building in flames in order to rescue the likely survivors? Or a squad of automata divers who can dive up to a human-impossible depth in order to set telecommunications submarine cables, to fix oil rigs to the seabed, or to rescue and refloat a ship in a completely autonomous way?

Please watch this video:

What is shown in this video are intelligent agents of a multi-agent system: a branch of the artificial intelligence (AI) which is called swarm intelligence –term used for the first time by Gerardo Beni and Wang Jing in 1989.

The intelligent agents of a multi-agent system are autonomous and heterogeneous components, which are not subordinated to any central control and able to dynamically adapt themselves to the local changes and self-organize.

The inspiration for the development of this branch of the AI comes from observing the behavior of birds, ants, termites… Their common methods of working, defending, building, feeding…

By the end of the seventies the first works about Distributed Artificial Intelligence (DAI) were published. Their target: studying models and techniques to solve problems in which the distribution –physical or functional- was inherent. The DAI systems owns an architecture consisted of intelligent and modular components which interact in a coordinated way.

From the perspective of the distributed problem solution the intelligent agents must have the following features according to the specialists Durfee and Rosenschein:

• Benevolence: the agents cooperate with each other whenever feasible. They cannot lie or hide information.
• Shared targets: all the agents gauge the result of the group activity with the same scale and they wish to help in order to maximize its quality.
• Central design: all the agents are designed to be included in an intelligent system, which is able to solve a problem. The designer must guarantee every single agent plays a role which helps achieve the final goal.

The working procedure of a group of intelligent agents is usually the following:

1. Task division: a task is divided in other smaller less complex tasks.
2. Allocation of tasks and resources amongst agents: it’s defined what agents will have to perform a certain task and what kind of resources are available.
3. Subproblem solution: each agent solves the problems which have been allocated to it.
4. Solution integration: in order to achieve a global solution to the initial task.

For the specialists in multi-agent systems Woodridge and Jennings, the main features of an intelligent agent are the following:

• Ability to solve non-trivial problems: a intelligent agent can reason about the surroundings –a skill which enables it to perform a group of tasks.
• Limited rationality: the agents are endowed with a group of objectives and they launch actions to achieve them. They choose their actions according to the rationality principle; i.e., they prefer the most promising action for their targets.
• Limited autonomy: the agents have their own motivations from which they develop autonomously their targets.
• Reactivity and proactivity: the agents analyze the surroundings and provide an answer to the changes taking place in it.
• Sociability: an agent bears in mind the existence of other agents and interact with them through some kind of communication protocol.

There are still many aspects in this field of the AI to be studied and researched, but every single day we are closer to the authentic etymological meaning of the expression “creating automata”.