Information Technology Reference
In-Depth Information
Agents are autonomous entities [7], as their execution is not subject to a
global flow of control. Indeed, the execution of an agent in a multiagent system
may proceed asynchronously, and the agent's state transition occur according
to local internal timings. This is actually what happens in DCA, because of the
adopted time-driven dynamics. Moreover, agents are situated entities that live
dipped in an environment, whether a computational one, e.g., a Web site, or a
physical one, e.g., a room or a manufacturing unit to be controlled. The agent
is typically influenced in its execution (i.e., in its state transitions)by what it
senses in the environment. In this sense, agents and multi-agent systems are
”open systems”: the global evolution of a multi-agent system may be influenced
by the environment in which it lives. And, in most of the cases, the environment
possesses a dynamics which is not controllable or foreseeable. For instance, com-
putational resources, data, services, as well as the other agents to be found on a
given Web site cannot be predicted and they are likely to change in time. This
sort of openness is the same that we can find in DCA, where the perturbation
of the environment, changing the internal state of a cell, can make us consider
the cell as situated in an environment whose characteristics dynamically change
in an unpredictable way.
Given the above similarities, we argue that similar sort of macro-level behav-
iors are likely to make their appearance also in such systems, raising the need
for models, methodologies, and tools, explicitly taking into account the auton-
omy and environmental dynamics and exploiting them either constructively, to
achieve globally coordinated behaviors, or defensively, to control the behavior
of the system. On the one hand, one could think at exploiting the environmen-
tal dynamics to control and influence a multi-agent system from ”outside the
loop” [15], that is, without intervening on the system itself. In a world of con-
tinuous computations, where decentralized software systems are always running
and cannot be stopped (this is already the case for Internet services and for em-
bedded sensors)changing, maintaining and updating systems by stopping and
re-installing them is not the best solution, and it could not be always feasible.
On the other hand, the reported experiments open up the possibility that a soft-
ware system immersed in a dynamic environment may exhibit behaviors very
different from the ones it was programmed for. Obviously, this is not desirable
and may cause highly damaging effects.
Of course, we are not the first discussing the possibility of emergence of
complex self-organizing behaviors in multi-agent systems. However, most of the
studies (apart from a few exceptions [12])have focused on ”closed” agent sys-
tems, in which the internal dynamics of the systems totally drive its behavior.
Instead, we have shown, via a very simple and ”minimal” multi-agent system,
as a DCA can be considered, that complex non-local behaviors can emerge due
to the influence of the environmental dynamics. The impact of this observation
in the modeling, engineering, and maintaining of distributed agent systems may
be dramatic [8,21].
