Privacy has been a concern for humans long before the explosive growth of the Internet. The advances in information technologies have further increased these concerns. This is because the increasing power and sophistication of computer applications offers both tremendous opportunities for individuals, but also significant threats to personal privacy. Autonomous agents and multi-agent systems are examples of the level of sophistication of computer applications. Autonomous agents usually encapsulate personal information describing their principals, and therefore they play a crucial role in preserving privacy. Moreover, autonomous agents themselves can be used to increase the privacy of computer applications by taking advantage of the intrinsic features they provide, such as artificial intelligence, pro-activeness, autonomy, and the like. This article introduces the problem of preserving privacy in computer applications and its relation to autonomous agents and multi-agent systems. It also surveys privacy-related studies in the field of multi-agent systems and identifies open challenges to be addressed by future research.

A primary research stream that contributed to the birth of case-based reasoning (CBR) was Artificial Intelligence and Law. Since law is largely about cases, it is a particularly interesting domain for CBR researchers. This article surveys some of the historically significant systems and developments in this field.

Considering the great number of agent-oriented methodologies that can be found in the literature, and the fact that each one defines its own concepts and system structure, one of the main challenges in agent-oriented software engineering (AOSE) research is how to make these methodologies interoperable. By defining concepts used in a specific domain in a non-ambiguous way, meta-modelling may represent a step towards such interoperability. Consequently the main objective of the AOSE TFG (Technical Forum Group) is to establish a strategy for identifying a common meta-model that could be widely adopted by the AOSE community. This paper sums up the approach used by this TFG which consists of (i) studying and comparing the meta-models related to some existing methodologies (ADELFE, Gaia, INGENIAS, PASSI, RICA and Tropos) in order to find commonalities and (ii) giving a clear and basic definition for the core concepts used in multi-agent systems for relating and positioning them in a unified MAS meta-model. The first proposal, set up by the working group, for this unified meta-model then concludes this paper.

Data mining has become a well-established discipline within the domain of artificial intelligence (AI) and knowledge engineering (KE). It has its roots in machine learning and statistics, but encompasses other areas of computer science. It has received much interest over the last decade as advances in computer hardware have provided the processing power to enable large-scale data mining to be conducted. Unlike other innovations in AI and KE, data mining can be argued to be an application rather then a technology and thus can be expected to remain topical for the foreseeable future. This paper presents a brief review of the history of data mining, up to the present day, and some insights into future directions.

Explanation of reasoning is one of the most important abilities an expert system should provide in order to be widely accepted. In fact, since MYCIN, many expert systems have tried to include some explanation capability. This paper reviews the methods developed to date for explanation in heuristic expert systems.

When building a Bayesian belief network, usually a large number of probabilities have to be assessed by experts in the domain of application. Experience shows that experts are often reluctant to assess all probabilities required, feeling that they are unable to give assessments with a high level of accuracy. We argue that the elicitation of probabilities from experts can be supported to a large extent by iteratively performing sensitivity analyses of the belief network in the making, starting with rough, initial assessments. Since it gives insight into which probabilities require a high level of accuracy and which do not, performing a sensitivity analysis allows for focusing further elicitation efforts. We propose an elicitation procedure in which, alternately, sensitivity analyses are performed and probability assessments refined, until satisfactory behaviour of the belief network is obtained, until the costs of further elicitation outweigh the benefits of higher accuracy or until higher accuracy can no longer be attained due to lack of knowledge.

In the last decades, most works in the literature have been devoted to study argumentation formalisms that focus on a defeat relation among arguments. Recently, the study of a support relation between arguments regained attention among researchers; the bulk of the research has been centered on the study of support within the context of abstract argumentation by considering support as an explicit interaction between arguments. However, there exist other approaches that take support into account in a different setting. This article surveys several interpretations of the notion of support as proposed in the literature, such as deductive support, necessary support, evidential support, subargument, and backing, among others. The aim is to provide a comprehensive study where similarities and differences among these interpretations are highlighted, as well as discuss how they are addressed by different argumentation formalisms.

Human understanding in design evolves through a process of critiquing existing knowledge and consequently expanding the store of design knowledge. Critiquing is a dialogue in which the interjection of a reasoned opinion about a product or action triggers further reflection on or changes to the artifact being designed. Our work has focused on applying this successful human critiquing paradigm to human-computer interaction. We argue that computer-based critiquing systems are most effective when they are embedded in domain-oriented design environments, which are knowledge-based computer systems that support designers in specifying a problem and constructing a solution. Embedded critics play a number of important roles in such design environments: (1) they increase the designer's understanding of design situations by pointing out problematic situations early in the design process; (2) they support the integration of problem framing and problem solving by providing a linkage between the design specification and the design construction; and (3) they help designers access relevant information in the large information spaces provided by the design environment. Three embedded critiquing mechanisms—generic, specific, and interpretive critics—are presented, and their complementary roles within the design environment architecture are described.

This commentary briefly reviews work on the application of case-based reasoning (CBR) to the design and construction of educational approaches and computer-based teaching systems. The CBR cognitive model is at the core of constructivist learning approaches such as Goal-based Scenarios and Learning by Design. Case libraries can play roles as intelligent resources while learning and frameworks for articulating one's understanding. More recently, CBR techniques have been applied to design and construction of simulation-based learning systems and serious games. The main ideas of CBR are explained and pointers to relevant references are provided, both for finished work and on-going research.

A probabilistic network is a graphical model that encodes probabilistic relationships between variables of interest. Such a model records qualitative influences between variables in addition to the numerical parameters of the probability distribution. As such it provides an ideal form for combining prior knowledge, which might be limited solely to experience of the influences between some of the variables of interest, and data. In this paper, we first show how data can be used to revise initial estimates of the parameters of a model. We then progress to showing how the structure of the model can be revised as data is obtained. Techniques for learning with incomplete data are also covered. In order to make the paper as self contained as possible, we start with an introduction to probability theory and probabilistic graphical models. The paper concludes with a short discussion on how these techniques can be applied to the problem of learning causal relationships between variables in a domain of interest.

The problem of formulating knowledge bases containing action schema is a central concern in knowledge engineering for artificial intelligence (AI) planning. This paper describes Learning Object-Centred Models (LOCM), a system that carries out the automated generation of a planning domain model from example training plans. The novelty of LOCM is that it can induce action schema without being provided with any information about predicates or initial, goal or intermediate state descriptions for the example action sequences. Each plan is assumed to be a sound sequence of actions; each action in a plan is stated as a name and a list of objects that the action refers to. LOCM exploits assumptions about the kinds of domain model it has to generate, rather than handcrafted clues or planner-oriented knowledge. It assumes that actions change the state of objects, and require objects to be in a certain state before they can be executed. In this paper, we describe the implemented LOCM algorithm, the assumptions that it is based on, and an evaluation using plans generated through goal-directed solutions, through random walk, and through logging human-generated plans for the game of freecell. We analyze the performance of LOCM by its application to the induction of domain models from five domains.

Many optimisation problems involve combinatorial constraints on continuous variables. An example of a combinatorial constraint is that at most one variable in a group of nonnegative variables may be positive. Traditionally, in the mathematical programming community, such problems have been modeled as mixed-integer programs by introducing auxiliary binary variables and additional constraints. Because the number of variables and constraints becomes larger and the combinatorial structure is not used to advantage, these mixed-integer programming models may not be solved satisfactorily, except for small instances. Traditionally, constraint programming approaches to such problems keep and use the combinatorial structure, but do not use linear programming bounds in the search for an optimal solution. Here we present a branch-and-cut approach that considers the combinatorial constraints without the introduction of binary variables. We review the development of this approach and show how strong constraints can be derived using ideas from polyhedral combinatorics. To illustrate the ideas, we present a production scheduling model that arises in the manufacture of fibre optic cables.

First-generation expert systems have significant limitations, often attributed to their not being sufficiently deep. However, a generally accepted answer to “What is a deep expert system?” is still to be given. To answer this question one needs to answer “Why do first-generation systems exhibit the limitations they do?” thus identifying what is missing from first-generation systems and therefore setting the design objectives for second-generation (i.e. deep) systems. Several second-generation architectures have been proposed; inherent in each of these architectures is a definition of deepness. Some of the proposed architectures have been designed with the objective of alleviating a subset, rather than the whole set, of the first-generation limitations. Such approaches are prone to local, non-robust solutions. In this paper we analyze the limitations (under the categories: human-computer interaction, problem-solving flexibility, and extensibility) of the first-generation expert systems thus setting design goals for second-generation systems. On the basis of this analysis proposed second-generation architectures are reviewed and compared. The paper concludes by presenting requirements for a generic second-generation architecture.

Pervasive computing aims at providing services for human beings that interact with their environment (encompassing objects and people who reside in it). Pervasive computing applications must be able to take into account the context in which users evolve, for example, physical location, social or hierarchical position, current tasks as well as related information. These applications have to deal with the dynamic integration in the environment of new, and sometimes unexpected, elements (users or devices). In turn, the environment has to provide context information to newly designed applications. This requires a framework which is open, dynamic and minimal. We describe an architecture in which context information is distributed in the environment and context managers use semantic Web technologies in order to identify and characterize available resources. The components in the environment maintain their own context expressed in RDF (Resource Description Framework) and described through OWL ontologies. They may communicate this information to other components, obeying a simple protocol for identifying them and determining the information they can provide. We show how this architecture allows introducing new devices and new applications without interrupting what is working. In particular, the openness of ontology description languages makes possible the extension of context descriptions and ontology matching helps dealing with independently developed ontologies.

The majority of multi-agent reinforcement learning (MARL) implementations aim to optimize systems with respect to a single objective, despite the fact that many real-world problems are inherently multi-objective in nature. Research into multi-objective MARL is still in its infancy, and few studies to date have dealt with the issue of credit assignment. Reward shaping has been proposed as a means to address the credit assignment problem in single-objective MARL, however it has been shown to alter the intended goals of a domain if misused, leading to unintended behaviour. Two popular shaping methods are potential-based reward shaping and difference rewards, and both have been repeatedly shown to improve learning speed and the quality of joint policies learned by agents in single-objective MARL domains. This work discusses the theoretical implications of applying these shaping approaches to cooperative multi-objective MARL problems, and evaluates their efficacy using two benchmark domains. Our results constitute the first empirical evidence that agents using these shaping methodologies can sample true Pareto optimal solutions in cooperative multi-objective stochastic games.

Agent software technologies are currently still in an early stage of market development, where, arguably, the majority of users adopting the technology are visionaries who have recognized the long-term potential of agent systems. Some current adopters also see short-term net commercial benefits from the technology, and more potential users will need to perceive such benefits if agent technologies are to become widely used. One way to assist potential adopters to assess the costs and benefits of agent technologies is through the sharing of actual deployment histories of these technologies. Working in collaboration with several companies and organizations in Europe and North America, we have studied deployed applications of agent technologies, and we present these case studies in detail in this paper. We also review the lessons learnt, and the key issues arising from the deployments, to guide decision-making in research, in development and in implementation of agent software technologies.

A considerable amount of research in case-based reasoning (CBR) has recently focused on conversational CBR as a means of providing more effective support for interactive problem solving. We review progress made to date and identify challenges that remain to be addressed.

A basic observation is that case-based reasoning has roots in different disciplines: cognitive science, knowledge representation and processing, machine learning and mathematics. As a consequence, there are foundational aspects from each of these areas. We briefly discuss them and comment on the relations between these types of foundations.

The objective of this paper is twofold. In its first part, we survey the state of the art in research on agent architectures. The architecture of an agent describes its modules and capabilities, and how these operate together. We structure the field by investigating three important research threads, i.e. architectures for reactive agents, deliberative agents and interacting agents. Then we describe various hybrid approaches that reconcile these three threads, aiming at a combination of different features like reactivity, deliberation and the ability to interact with other agents. These approaches are contrasted with architectural issues of recent agent-based work, including software agents, softbots, believable agents, as well as commercial agent-based systems. The second part of the paper addresses software engineers and system designers who are interested in applying agent technology to their problem domains. The objective of this part is to assist these readers in deciding which agent architecture to choose for a specific application. We characterise the most important domains to which the different approaches described in the first part have been applied, propose an application-related taxonomy of agents, and give a set of guidelines to select the right agent (architecture) for a given application.

There seems to be general agreement amongst those involved in KBS research that in order to be useful, a system must be able to explain its reasoning to a user. This paper reviews the development of explanation facilities in knowledge-based systems. It differentiates between explanation as a problem-solving process, and that which explains a reasoning process. This review concentrates on the latter, identifying and giving examples of three categories of reasoning explanation.

We then look at user requirements for explanation. What makes an explanation useful depends on the expectations of a user, which in turn depends on such issues as user background and system context. Several techniques are examined that have been applied to the problem of producing explanations that are appropriately structured and conveyed.

Finally, we discuss some of the work that has been done in describing theories of human discourse and explanation, and some issues that will become increasingly important for future explanation systems.

An extensive annotated bibliography is provided.