Information Technology Reference
In-Depth Information
Research of DAI can be generally categorized into two domains: Distributed
Problem Solving (DPS) and Multi-Agent System (MAS), both sharing the same
research paradigm yet adopting different problem solving means. The goal of
DPS is to establish large-granularity cooperative clusters to accomplish the
common problem solving objectives. In a pure DPS system, problems are
resolved into sub tasks, specific task executors are designed to solve the
corresponding sub tasks, and all interaction strategies are incorporated as an
integral part of the system. Such systems feature top-down design, since the
whole system is established to solve the predefined objectives at the top end.
On the opposite side, a pure MAS system generally comprises pre-existing
autonomous and heterogeneous agents without a common objective. Research on
MAS involves coordinations and cooperations in knowledge, plan and behavior
among groups of autonomous intelligent agents, so that they can jointly take
actions or solve problems. Though the agent here is also a task executor, it is
“open” to other peer agents, and can deal with both single objective and
multiple objectives.
Nowadays, applications of computers are becoming more and more extensive,
and problems to be solved are becoming more and more complex, which makes
centralized control of the problem solving process and centralized processing of
data, information and knowledge more and more difficult. Such distributed and
concurrent processing of data and knowledge hails great potentials along with
many pending difficulties to the development of AI. The spatial distribution,
temporal concurrency and logical dependant relationships of multiple agents
make the problem solving more complex in multi-agent systems than in
single-agent systems.
Despite such difficulties, research on DAI is feasible, desirable and important
for the following reasons:
(1) Technical foundations — Advances in technologies such as hardware
architecture of the processors and communication between the processors make it
possible to interconnect great amount of asynchronous processors. Such
connection might be tightly coupled systems based on shared or distributed
memory, or loosely coupled systems based on local networks, or even very
loosely connected systems based on geographically distributed communication
networks.
(2) Distributed problem solving — Many AI applications are distributed in
nature. They might be spatially distributed, such as the explanation and
