Ebook: Agents and Ambient Intelligence

Throughout the history of mankind, human beings have always had a strong drive to develop tools that could support them in their daily activities. Examples of such artefacts are countless and vary from early hand-axes made out of stone to various types of 21st century computing devices.

There is little doubt that the development of human tools has been considerably accelerated by events like the industrial and the digital revolution. However, another perhaps less notable, but nevertheless important innovation during the history of human tool development is characterised by the introduction of increasingly autonomous artefacts, or agents.

Driven by the growing potential of digital computers in the mid-20th century, the idea that computational systems could be capable of performing certain tasks autonomously (‘on behalf of their human users’) gradually gained space in the decades that followed. Eventually, the concept of an intelligent agent became widely adopted in the early 1990s. Although many definitions were used, the term roughly referred to an autonomous entity that observes its environment via sensors and acts upon it via actuators, and is able to communicate with other agents. Depending on ones viewpoint, several additional properties were ascribed to agents, such as goal directedness and the ability to learn. At the time, the term was mostly applied to software agents (such as shopping bots), although the notion of hardware agents (robots) was used as well.

Partly in parallel with the intelligent agent paradigm, another concept emerged in the late 1990s, namely the perspective of Ambient Intelligence. As the readers of this book series probably know, Ambient Intelligence (AmI) was put forward as a vision on the near future of computing. As such, it referred to a world in which human beings are surrounded by intelligent electronic systems that are unobtrusively incorporated in their environment, monitor their behaviour using sensors, and support them in their daily activities.

Now that the concepts of intelligent agents and AmI have become more mature, it seems to be an appropriate moment to reflect on the research area that integrates the two. When comparing the definitions of both concepts, a number of overlapping elements can be identified (which in turn may be interpreted as arguments why it makes sense to combine both areas). First of all, the intelligent agent paradigm may serve as a useful conceptual framework to talk about AmI systems. Since AmI systems themselves are often (partly) autonomous systems, which observe (the humans in) their environment and act upon them, it takes a relatively small step to state that any AmI system is in fact an intelligent agent. Consequently, all the properties that were traditionally ascribed to (varying classes of) agents, such as goal directedness and adaptivity, could in principle be re-used for AmI systems as well.

A second reason why it may be relevant to study the intersection of intelligent agent technology and AmI is related to the notion of multi-agent systems. An interesting characteristic of intelligent agents, which I have not mentioned so far, is the fact that in practice there are often multiple of them. After all, an important asset of intelligent agents is that they are social entities with the ability to communicate amongst each other, enabling them to potentially solve complex problems in a distributed manner. Since AmI systems are typically active in complex and dynamic real world environments, and possibly have to interact with humans via different modalities and at different locations, the capacity to solve tasks in a distributed manner seems to be highly beneficial for such systems. Therefore, instead of treating an AmI system as one individual agent, the notion of a multi-agent system may sometimes be a suitable metaphor as well.

Inspired by these similarities between agent technology and Ambient Intelligence, the main purpose of the current book is to provide an overview about the state-of-the-art of the scientific area that integrates the two. And indeed, this entails that some chapters focus more on AmI systems that can be seen as one single intelligent agent, whereas others address AmI systems that are composed of multiple agents.

The chapters included in this book address a wide variety of topics related to agents and AmI, including theoretical as well as practical issues, design-related as well as implementation-related issues, ethical as well as philosophical issues, and so on. For obvious reasons, there is no straightforward order in which the different chapters should be presented. Nevertheless, in an attempt to create some structure, the 12 chapters have been clustered into four parts of three chapters, which can loosely be described as follows:

1) ethical and philosophical issues of agent-based AmI systems

2) methods for development of agent-based AmI systems

3) towards more intelligent and adaptive agent-based AmI systems

4) applications of agent-based AmI systems

In a way, the order in which these four parts are presented can be seen as going from general to specific, of from abstract to concrete.

More in particular, the first part contains three chapters that discuss ethical and philosophical issues behind the general principle of agent-based AmI systems, such as the ambitions behind the AmI vision (Heylen), the possibilities to define a conceptual framework to describe agent-based AmI systems, inspired by notions such as Mirror Worlds and stigmergy (Castelfranchi et al.), and the responsibilities for users and designers of these systems (Detweiler et al.).

The second part comprises three chapters that present and explore various approaches that can be used to develop agent-based AmI systems. These approaches include the use of verification tools to support the design of correct systems (Augusto et al.), an agent-based simulation approach to test the behaviour of AmI systems (specifically for the domain of location-based services) before their actual implementation (Martinez et al.), and a middleware-based approach used to embed agents within AmI systems (O'Hare et al.).

The third part consists of three chapters that share the goal to endow AmI systems with useful properties like intelligence and adaptivity. In particular, these chapters introduce (and evaluate) a novel approach to learn timing of prompts in agent-based smart homes (Das et al.), an approach to personalise the level of autonomy of agent-based AmI systems (Ball et al.), and an overview of logic-based approaches for intention recognition in these systems (Sadri).

The fourth and last part of the book is composed of three chapters that present concrete applications of agent-based AmI systems. These applications include a system for user identity verification in the security domain (Dovgan et al.), a system for group emotion support (Duell et al.), and an approach to equip conversational systems with humour (Dybala et al.).

The combination of chapters included in this volume provides more insight in recent achievements as well as future challenges in the intersection of agent technology and Ambient Intelligence. It presents enlightening examples of how agent technology can be used to develop more intelligent, flexible, effective and user-friendly AmI systems, but also poses critical questions concerning the future of the AmI perspective, and the role of the agent paradigm therein.

To conclude, I wish to express my gratitude to all who contributed to the publication of this book, including all authors for sharing their interesting research with us, all reviewers for guarding the quality of the chapters, and last but not least, to the editor-in-chief and the publisher of the AISE book series for offering me the opportunity to publish this volume.

Tibor Bosse

VU University Amsterdam, Agent Systems Research group, De Boelelaan 1081, 1081 HV Amsterdam, The Netherlands

Amsterdam, March 15, 2012

This chapter presents an analysis of the ambitions that lie behind the concept of Ambient Intelligence as it is presented by the advocates and researchers working in the field. In particular it looks at the ideas regarding the forms of natural and intuitive forms of interaction that are envisaged – including agents and robots – from a philosophical perspective. The views on interaction are analysed and framed in the conceptual framework of Don Idhe's phenomenological analysis that describes how we as humans relate to the world existentially, mediated through the technologies that we construct. We compare and contrast the ambient intelligent vision on technology with some other ones.

In this chapter we introduce a vision of agent-oriented AmI systems that is extended to integrate ideas inspired by Mirror Worlds as introduced by Gelernter at the beginning of the eighties. In this view, AmI systems are actually a digital world mirroring but also augmenting the physical world with capabilities, services and functionalities. We then discuss the value of stigmergy as background reference conceptual framework to define and understand interactions occurring between the physical environments and its digital agent-based extension. The digital world augments the physical world so that traces left by humans acting in the physical world are represented in the digital one in order to be perceived by software agents living there and, viceversa, actions taken by software agents in the mirror can have an effect on the connected physical counterpart.

Ambient Intelligence Systems are complex, knowledge providing systems. The group of intended users typically comprises both various kinds of professionals and of laymen. In this chapter we address responsibility issues associated with the use and design of AmISs. We explain the connection between knowledge and responsibility, and use that connection to show that the usage of Ambient Intelligence Systems leads to moral responsibilities for all types of users (professionals and otherwise). It follows from the same reasoning scheme applied to the role of the AmIS developer, that she is responsible to design the system for responsible use. We give some initial criteria for such design for responsibility of AmISs.

We explore the availability of tools which can support the development of more correct and reliable Multi-Agent Systems (MAS) immersed in Ambient Intelligence (AmI) systems. We explain how state of the art software engineering methods and tools can be used to guide the development of such systems and increase our confidence on them. The explanation assumes little technical background on behalf of the reader. We use this exercise as an opportunity to highlight the advantages of available tools as well as to illustrate why the AmI area should build a new generation of development frameworks.

The main concern of this book chapter is about the software engineering process of developing ubiquitous computing services. The kind of services addressed are those that must operate in complex software environments and in complex scenarios involving many users and different situations in a proactive, intelligent and nonintrusive way. More specifically, this chapter focuses on indoor location services and how to start building them in the lab, as a first phase to a real deployment. The procedure shown is based on agent based social simulation of the environment in which the service works. Simulations are used to test the service, its reliability and its accuracy in a prior phase to work in a real set up. The specific location service used is based on a Bluetooth wireless infrastructure. The proposed location service was tested in a simulated environment showing that this procedure is able to test the service before a final deployment avoiding expensive and annoying tests.

This chapter reflects upon the challenges that confront the deployment of Ambient Intelligence (AmI) applications. Ambient Intelligence demands that everyday artefacts be imbued with intelligent reasoning capabilities together with the capacity for collaborative intelligent behaviour. Traditional ambient devices do not provide the requisite computational platform to support such requirements. With the ongoing developments of ubiquitous devices, however, the situation is changing. This chapter discusses a software stack, which supports the needs of ambient applications that incorporate embedded intelligence.

Over the last decade there has been a significant growth of research endeavors in the area of ambient intelligence. An anticipated increase in the older adult population around the world and increasing health care expenditures have increased the demand of smart health assistance systems. Delivering in-home activity interventions to residents for timely reminders ensuring successful completion of daily activities, is receiving significant attention in the community. In this chapter, the problem of automated in-home activity interventions is described and prospective solutions are compared. The approaches and challenges are based on a prototypic model of an automated prompting system, namely PUCK, which is an on-going project at the Center for Advanced Studies in Adaptive Systems at Washington State University. The previous study done on this project investigated the application of machine learning techniques to identify appropriate timing of prompts based on data provided by off-the-shelf sensors. The fundamental problem in learning timing of prompts is caused due to the under representation of prompt situations in the training examples as compared to no-prompt situations. While a method was originally proposed to deal with this problem, popularly known as learning from imbalanced class distributions, in this chapter a novel Cluster-Based Under-sampling (CBU) approach is proposed that shows promising results.

There are many arguments for and against the use of autonomous-agents in ambient intelligence and intelligent environments. Some researchers maintain that it is vital to restrict autonomy of agents so that users have complete control over the system; whereas, many others maintain that there is a greater benefit to be gained by employing autonomous-agents to take some of the work load off the user and increase user convenience. Both of these approaches have their distinct advantages but they are not suitable for all since people's opinions and concerns regarding autonomy are highly individual and can differ greatly from person to person. This work explores how it is possible to make intelligent environments more dynamic and personalisable by equipping them with adjustable autonomy, which allows the user to increase or decrease agent autonomy in order to find a comfortable sweet-spot between relinquishing/m gaining/losing convenience. This chapter discusses how adjustable autonomy can be achieved in intelligent environments, reports on a recent online survey conducted to gauge people's opinions of different levels of in intelligent environments, and discusses a user study for which an experimental adjustable autonomy enabled intelligent environment was developed. This work aims to raise awareness of the issues with using static (and extreme) levels of autonomy amongst researchers of intelligent environments and ambient intelligent environments.

In this chapter we discuss the contribution of intention recognition in agents in ambient intelligence, and the role of logic in intention recognition. We consider the relationship between causal theories used for planning and the knowledge representation and reasoning used for intention recognition. We look at the challenges and the issues, and we explore several case studies.

This chapter presents a high-security access-control system. The system consists of a set of intelligent agents that are organized in several layers and store the global knowledge in a global ontology. They learn user behavior at the access points and recognize significant behavior deviations produced by intruders or due to other unwanted situations. This chapter also presents three prototypes of the system, each specialized for a subset of security problems. The experiments show that the system is efficient in recognizing stolen-identification entries.

This chapter introduces an agent-based support model for group emotion, used by ambient systems that support teams in their emotion dynamics. Using model-based reasoning, an ambient agent analyses the team's emotion level for present and future time points. In case the team's emotion level is found to become deficient, the ambient agent provides support to the team, for example, by proposing the team leader to give a pep talk to certain team members. The support model has been formally designed and simulation experiments have been performed within a dedicated software environment.

Humor processing is quite an innovative area in intelligent agent technologies. In this chapter we present the results of our work on a multiagent joking conversational system, which is able to use humor during non-constrained conversations with humans, in reaction to their particular emotional states. Thus, this work combines the areas of humor processing, emotiveness analysis and dialogue agent technologies, being the first one to do so. We first describe the background of our study, including state of the art of research in the field of humor-equipped agents. We describe our multiagent system's general construction and algorithms of each constituent agent, i.e. a chatterbot, a humor generator and emotion detector. Two evaluation experiments: user-oriented and automatic, showed that the humor-equipped system was evaluated as more friendly, making users feel better and was generally more likable than a similar system without humor. These results are discussed and some ideas for the future are given.