The term “intelligent systems” has come to mean many different things in many different contexts and, like mostthings related to complex systems, it is hard to nail down a specific definition that is both rigorous enough to discriminate out those things whichshould not be included, but is loose enough to include those that are. As in defining terms like “agents” or “robots,” one is ableto find overly inclusive definitions, such as “an autonomously acting entity” where a thermostat in the latter case, or hard diskcontroller in the former, would meet the definition. On the other hand, tighten up the definition and telerobotics or Google's search bots no longer fit,despite being clearly related technologically. In the case of intelligent systems, too tight a definition of intelligence removes, say, the behaviorswe see in a dog, which can seek out prey or be trained to beg for its dinner, but loosen the definition and we find ourselves talking about systemswith the intelligence of a clam.

In this chapter, we are holding primarily to a tighter definition and starting to look at some of the kinds of behaviors that take us intoareas traditionally associated with human intelligence. While several of the sections of this book deal with areas associated with, loosely defined,intelligent behaviors and others look specifically at aspects of Artificial Intelligence tied closely to data or a model, in this short section wepick up some of the missing pieces that help us complete the puzzle. The small number of papers in this section should not cause one to believe that thereis little relation between intelligent systems and complexity, but rather that other sections of this encyclopedia necessarily included aspects ofintelligence defined at some level – controlling against complexity demands it. The reader looking for other articles on ArtificialIntelligence and complex systems will find them in many sections of this volume, particularly including those on:

Agent Based Modeling and Simulation,

Complexity and Non‐linearity in Autonomous Robotics, Introduction to,

Data-Mining and Knowledge Discovery, Introduction to

Data-Mining and Knowledge Discovery, Neural Networks in

Data-Mining and Knowledge Discovery: Case-Based Reasoning, Nearest Neighbor and Rough Sets

Soft Computing, Introduction to, and

Nonsmooth Analysis in Systems and Control Theory

Chronological Calculus in Systems and Control Theory

However, despite the strong ties between these many subareas, there were some aspects of intelligent systems left out of theseother topic areas, and this section is provided to cover these.

One such example is that of the use of mobile agents as the basis for intelligent cooperation among systems(see Mobile Agents). The primary need for mobility is in bandwidth‐limited communications, and withthe growth of modern computer networks, the area has gotten less attention of late. However, with the growing need for, and use of, sensornetworks, wireless networks in noisy environments, deep-sea or space robotics, and other bandwidth‐limited systems, agent-based modelingand simulation techniques can be used to model the networks, but not run on the networks themselves. Thus, an adaptive network that needs to doagent-to-agent communication for self‐tuning will be seriously impacted ina bandwidth‐limited environment. Mobility can be used to help in such situations, moving small amounts of code to where the data itmanipulates is stored, rather than moving the large amounts of data to where the computation could be run. Using such mobile agents bring bothnew capabilities, and new challenges, and as we try to increase the intelligence of systems operating in networks with disconnection, lowbandwidth or high latency, such as many of the networks deployed in warfighting situations, agent mobility is regaining attention as animportant area of research.

Another area that needs to be covered is that of the role of intelligent systems in the modeling and simulation area(see Artificial Intelligence in Modeling and Simulation). While agent-based modeling, as discussed elsewhere in this encyclopedia, is an important area,there are aspects of modeling it does not fully cover. One of these is the use of symbolic reasoning for use in validating models and simulation. Anotheris the actual modeling of the reasoning of other agents. For example, a baseball player recovering a ball that dropped in the outfield must beable to reason about what the base runners, who are trying to move up, and his team-mates, who are trying to hold them back, will most likely do. Theplayer's skill in guessing the behaviors of these other agents, both competitive and cooperative, could be the difference between whether the game is wonor lost.

As well as improving our capabilities in modeling and simulating systems, it is important to look at how, with the use of intelligent systems, wemight better control the complex systems being modeled. Instead of trying to model a complex plant directly, in this section we look at work thattakes a different approach. Instead we consider the knowledge‐based control that results from observing, studying and understanding thebehavior of a plant and/or the behavior of a human controlling it (see Intelligent Control). This area includes looking at soft‐computing approaches to create an“approximate reasoning” solution that can be used for mimicking the control decisions that would be made by a human monitoringa plant, rather than for modeling the plant itself. Fuzzy logic, a particular branch of soft control has been successfully applied to thecontrol of many complex systems, a number of which are described here. (The reader with a sense of humor may see a certain irony in manymanufacturer's use of fuzzy logic to improve the picture on their camcorders and television sets, but let us leave that unexplored). Of particular import,this article outlines various ways in which the mathematical operations used in such control can be combined, allowing for the control of complex,non‐linear systems that defy simpler control regimes.

Looking at very complex situations, a human operator, or at least a program simulating one, may want to look beyond soft computing anddeal with the world at a higher level. One of the key abilities which separates the human from other primates and animals is our ability to learnover time to abstract away many details of the complex world in which we live, and to make plans for how to control it over time. Where planning itselfcan be a complex problem‐solving task, learning how to abstract key aspects of situations and to apply plans to these is a critical needfor dealing with complexity (see Learning and Planning (Intelligent Systems)). Exploring how we learn to plan is an area which has been gaining importance in the intelligentsystems area as approaches which do not learn, but which apply brute force problem solving to larger and larger problems, are reaching the limits of theircapabilities against the increasingly complex domains in which we wish to deploy our computational systems.

A recurring theme that arises in all of these attempts to provide intelligent behavior in evermore‐complex systems environments is thatof using a level of abstraction to reason not about some data or system itself, but about the meaning of the behaviors beingobserved. A critical aspect of performing such abstraction is the ability to represent a model of a system to the computer ina machine‐readable way. The term “ontology” is used to describe this computer‐based representation of the domain in whicha system is trying to function.

Although ontologies have been around for a long time in AI, they have recently come back into their own in trying to help computer systemsinteract with one of the most complex human constructions in history – the millions of billions of dynamically changing bits of informationthat comprise the World Wide Web. While the Web has changed a great many aspects of our society, understanding its dynamics remains a majorchallenge (See Berners‐Lee et al., Creating a Science of the World Wide Web, Science, 313(5788), August2006, pp 769–771.) The use of ontologies, and other AI techniques, to help computers better process this massive information space is the raisond'etre of the “Semantic Web,” an overview of which is also presented in this section (see Semantic Web).

In short, this section, despite its brevity, picks up a number of themes that arise throughout this encyclopedia. These five articles will helpthe reader understand how many of the themes above are connected together via the use of technologies developed in artificial intelligence labs, allowingthe creation of intelligent systems that provide a key tool in our arsenal for dealing with the complexity of the natural world and/or the complexhuman society that has evolved to let us live in it.