Socially Intelligent Agents – Creating Relationships with Computers and Robots

Every so often a new computing technology comes along which catches the imagination of people well outside its immediate circle. Expert systems in the 1980s and virtual reality in the 1990s were good examples, and agent technology is yet another. However, a characteristic of such high‐impact areas is that those not actually working within the associated research community are sometimes uncertain about its exact contribution and about the meaning of the new terms and vocabulary it introduces. In reviewing a book called Socially Intelligent Agents it therefore seems a good idea to start by clarifying the key term “agent”.

Agents are often defined as computational systems that are capable of flexible, autonomous action, in dynamic, unpredictable environments. More broadly, an agent is a system situated within part of an environment that senses that environment and acts on it, over time, in pursuit of its own agenda and so to affect what it senses in the future. This second definition of course goes way beyond software and hardware to biological systems, from ants to humans, since they also have these characteristics.

So from one point of view an agent is a modern take on the original aims of Artificial Intelligence. Agent‐based systems in practice cover a wide range – autonomous robots, synthetic characters (2D or 3D graphical characters), expert assistants, and purely software agents, sometimes referred to as knowbots, or softbots, and widely used for internet retrieval and e‐commerce. Like most expanding research communities, agent research has diversified into sub‐communities with somewhat different interests and agendas, with, for example, some research concentrating on agents as software engineering, working at a higher level of abstraction from object‐oriented programming, and others concentrating on embodied agents and their interaction with human users.

Interaction between agents and humans is very much the focus of the book under review. So why “socially intelligent”? The thesis here – confirmed by research – is that agents, with their emphasis on autonomy and proactivity, are effectively seen in some sense as “people” by human users, especially when they are embodied and these bodies become expressive and thus part of the overall interaction process. This well‐known human tendency to anthropomorphism cannot be ignored – since users will inevitably treat such agents as part of the social context, agents have to be equipped with the social knowledge and behaviour that will allow them to meet user expectations. Just what this is must of course be researched, but it certainly includes topics such as personality, affective systems, facial and gesture‐based expressiveness and criteria for believability. Concerns that might only have worried animators and other artists in the past have now entered a technical research community.

Alongside this research impetus for socially intelligent agents, one might also add that more functional technology and better computer and engineering platforms are also taking such agent systems out into the big wide world to a much greater extent in the past. Consider the Sony Aibo robots for example – marketed as computer pets, bought as consumer goods and not just as research tools. Or consider virtual news readers, one of which, Ananova, changed hands from the Press Association to Orange in July 200 for GB£95 million. Most people have heard of human‐computer interaction (HCI) but this is now being supplemented by the specific area of or human‐robot interaction (HRI).

The genesis of Socially Intelligent Agents lies in one of the symposia organised in 2000 by the American Association for Artificial Intelligence (AAAI) as part of their annual Fall Symposia event. Specifically entitled “Socially intelligent agents – the human in the loop”, the event targeted the human interaction aspect just introduced. The book has a particular aim, which is to introduce the whole range of activity in this new but expanding research area, and to meet this aim it includes 32 relatively short (eight‐page) papers rather than longer but fewer chapters carrying more detail. The virtue of this approach is that it contains something that will interest most readers – from robots, through virtual dramas including synthetic characters, to interface agents on the Web, social simulation and discussions of the theory underlining basic concepts.

Of course there is also a downside, which is that the overall feel may be a little “bitty”, making the book more useful as a resource into which to dip than something to be read cover to cover. Conscious of this problem, the four editors discuss themes and structures in their introduction in order to give the reader a bit of a compass. They explain that first the book is divided into two: an initial section of 12 papers concerned with theory, concepts and technology, and a second section concerned with applications. The first section then divides into three themes – relationships between agents and humans, agents and emotion/personality, and communities of social agents.

The second section divides into five application areas, including robots, education and training and games and entertainment. A figure later in this introduction also lists chapters against themes in order to cope with the overlaps, though it might have been easier to grasp as a table. There is a risk here too that the reader who looks at the table of contents – which just lists the chapters one after the other without subdivision – and does not read the introduction, will miss this thematic organisation and find it hard to read in a sensible order. Having said this, the separate papers, though short, are full of fascinating work and ideas, and almost anywhere the reader starts or continues, they are likely to find interest and stimulation. This makes it a shame that the publishers have chosen to charge the sort of price (GB£88.00) which means that only libraries are likely to afford it.

With such a variety of work, it seems a little invidious to pick out particular papers. However if you thought robots were those big strong manipulators in factories, the set of papers on robot applications here will change your mind. The well‐reported work on the MIT Kismet robot by Brezeal shows how including facial expression driven by an underlying emotional model induces humans to interact with Kismet as if it were a baby, even though it looks nothing like one. To show that this need not require heavy‐duty engineering, Canamero demonstrates similar effects with a simple Lego robot, Feelix. Meanwhile, both Dautenhahn et al. and Michaud and Theberge‐Turmel have been using robots for therapeutic purposes with autistic children. Blocher and Picard have combined computer animation with a set of soft toys to try to teach the recognition of emotional expressions to children affected by autism.

The therapeutic theme is continued – but using graphical characters – by Marsella, who aims to teach the mothers of child oncology patients a cognitive strategy to help deal with their problems, which of course involve a substantial emotional component. Graphical synthetic characters also appear in Machado and Paiva, where virtual actors support children in story authoring, using a fairy‐tale approach, while Montemayer et al. also support story‐telling, but this time via collaborative robot design and construction in which children play and equal role with adult experts.

In summary, this book provides a vivid and multi‐faceted introduction to a new research area that is still exploring its terrain, and bringing together a variety of disciplines and groups. Many papers may leave you wanting to know more – and this is indeed the intention of the editors. The approach would also very much suit many late undergraduate or graduate courses where the aim is to provide leads to the state of the art and start students off on a process of independent exploration. There is nothing else so far which covers as much ground in this area, and the work can thus be regarded as a unique contribution to this field.