Foundations of Distributed Artificial Intelligence

Citation

Andrew, A.M. (1999), "Foundations of Distributed Artificial Intelligence", Kybernetes, Vol. 28 No. 3, pp. 310-311. https://doi.org/10.1108/k.1999.28.3.310.1

The Preface starts by defining distributed artificial intelligence (DAI) as a subfield of AI concerned with distributing and coordinating knowledge and actions in multiple agent environments. There are aspects of human intelligence that only become apparent when people interact. Also, as computers are programmed to behave intelligently they come to be seen as intelligent assistants rather than as tools in the usual sense of the word, and the study of multi‐agent intelligent systems has implications for man‐machine interaction.

It is also pointed out that some application areas are inherently distributed in character. This is clearly so in certain robotics applications. The example is also quoted of experts with different specialities collaborating to solve a problem; this seems less clear since if the experts are implemented in similar hardware they could be seen as one parallel machine. There may, however, be advantages in the “distributed” paradigm, encouraging sparseness of communication between the experts except at a high level. Something of the sort is well‐known in the “blackboard” schemes used for complex perceptual processes such as speech recognition. Blackboard schemes are mentioned at several points in the book, and may be used to relate, not only partial problem solutions as in applications to perception, but also goals or queries.

This book reviews the field with impressive thoroughness, using contributions from an international team of 29 authors. It is divided into four parts. The first has four chapters dealing with general aspects, including a lengthy initial chapter giving an overview of the field and including nine and a half pages of references. It is also mentioned that one of the references (Bond and Gasser, 1992) contains a subject‐indexed bibliography of more than 570 items ‐ the literature is extensive and growing. Apart from separate reference lists following the other chapters, the book ends with a list of information sources including books, journals, conferences and Internet resources.

Approaches to the subject can roughly be divided into those that focus on the agents, with attention to the characteristics that allow them to interact with their fellows (who may or may not be similar to themselves), and approaches that focus on groups. Parts two and three of the book roughly correspond to this division of interest, with part two headed: “Cooperation, coordination and agency”, and part three: “DAI frameworks and their applications”. Part four then treats a number of related matters, particularly the impingement of the topic on philosophy and sociology.

Within the first chapter, a schematic diagram shows the “General architecture of a social agent” with no fewer than 29 interconnected blocks, of which ten represent processes such as “Perception” and “Plan activation” and 19 represent categories of stored data. This general architecture is not claimed to be unique or all‐embracing but it provides a framework for discussion. Chapter two is on “Logical foundations of DAI” and invokes such deep issues as temporal modal logic and possible‐world semantics. In the later chapters quite a number of existing schemes, represented by acronyms of the kind familiar in AI, are treated in detail. These include, in chapter 17, an agent factory, or software automating the design of multi‐agent systems, reminiscent of the compiler devised to aid the implementation of programming languages.

In the descriptions, a great many terms are used that have an emotive content, including belief, desire, intention, negotiation, commitment, convention and interest. Chapter four, on industrial applications of DAI, indicates very practical incentives to use the technique. Much industrial planning is arguably better performed by a distributed system than by a central planner. If everything goes smoothly, the overall effect may be much the same since the central planner probably uses a heuristic process of successive adjustment that is similar to negotiation between distributed agents. However, the system using distributed agents is likely to be more robust and flexible when there are unexpected disturbances, representing either difficulties or opportunities. Also, the distributed system is likely to lend itself better to participation of humans.

In the final part, related disciplines are discussed, mainly with reference to social systems. An aspect that is not treated, but would seem to be relevant, is comparison with other biological systems with regard to their distributed nature, especially the nervous systems of animals. Simple reflexes, for example, are relatively autonomous agents that have been attributed to self‐contained arcs within the spinal cord, and yet they do not operate in isolation, as remarked by Melzack and Wall (1982). The suggestion that useful paradigms can be found in neurophysiology has been strongly advanced by Stafford Beer (1962). Apart from this particular aspect, ideas on distributed intelligent systems are very thoroughly covered in the new book.