Book Review from Kybernetes 32, no. 7/8, pp. 1189-91, 2003
Imitation in Animals and Artifacts
edited by Kerstin Dautenhahn and Chrystopher L. Nehaniv
MIT Press (Bradford Book), Cambridge, Mass., 2002, xv + 607 pp., ISBN 0-262-04203-7, hardback, £44.95
This substantial volume is based on a symposium with the same title, held in Edinburgh in April 1999, under the auspices of the Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB). It has 22 chapters by different authors from a variety of countries. The papers are all "meaty" and well prepared and since the average length is about 26 pages there is a great deal of material here. In addition, in an Appendix it is mentioned that the symposium also prompted a special double issue of the journal Cybernetics and Systems (vol. 32, nos. 1-2, 2001) on "Imitation in Natural and Artificial Systems", with another 10 relevant papers. A list of presentations and speakers at the symposium is on the web at: <http://homepages.feis.herts.ac.uk/~nehaniv/aisb.html> (which is not exactly the address given in the book itself, which doesn't work!).
One aspect that is not covered is the genetically-determined imitative marking shown particularly by insects and fish. The interest if the AISB is in behavioural imitation, which gives plenty to be going on with. A rather pleasant feature is that imitation is invariably viewed as something positive; the terms "deception", "intimidation", "warning" and even "caricature" do not appear in the subject index. A feature of the high standard of presentation is that there is a good overall subject index.
In the Preface and the first two chapters the editors claim that the symposium broke new ground by bringing together researchers from a broad range of disciplines ranging from animal behaviour to artificial intelligence and from computer science, software engineering and robotics to experimental and comparative psychology, neuroscience, primatology and linguistics. They also show that the importance and complexity of the topic are often underestimated. Imitation has tended to be dismissed as a rather low-grade activity having, almost by definition, no innovative content, but this is wrong. For one thing, even apparently simple imitation of a physical action by another individual involves complicated processing to go from the observation, probably visual, to the corresponding muscular activity. In many cases, especially when a skill is transferred by example, the learned behaviour is a subtle mixture of imitation with adaptive fine-tuning by the learner and incorporation of the learner's prior capabilities for balance, etc., and modifications necessitated by the learner's different situation.
Despite the necessary complexity of its mechanism, the human ability to imitate seems to be innate and to be displayed at an early age, as shown by the readiness of very young children to play imitation games, some involving facial expressions. In papers in the volume, variations in the character of the imitation are studied, for instance the degree to which the precise "style" of the exemplar is copied. Where a sequence of actions achieves a goal, it may be copied in a goal-centred way, without regard to style. A difference between autistic and other children in this respect is noted by one author, with the autistic subjects more goal-directed, and there are several references to autism throughout the volume.
Various studies of imitation in animals are reported. One paper is on imitation by dolphins, and on the dust cover there is a very sweet series of illustrations showing a dolphin rather exactly copying postures assumed by a human. Primates are extensively studied, and the title of one paper draws attention to the fact that the readiness of monkeys to imitate is enshrined in the verb "to ape". A study of imitation by parrots is reported showing that their production of human language has much more structure and relevance to tasks and goals than is generally supposed, and is not at all what is generally referred to as parrot-like repetition.
It has been mentioned that the imitation of human movements by another human involves complex processing. The nature of this can be studied by building robots able to imitate humans, and there is a chapter in the book describing work to this end, in the Artificial Intelligence Laboratory of MIT. As well as interest in imitation for its own sake and for its social function, there is the aim of finding convenient ways of training robots in new tasks. The experimental anthropomorphic robots Cog and Kismet are referred to and shown in illustrations.
Another chapter reports a very interesting example of machine learning, with the title "Learning to fly". It is by a group in the Turing Institute in Glasgow that includes Professor Donald Michie, and the experiments are designed to train an artificial system to become an effective autopilot, and particularly an autolander for an aircraft. The training has been done using a flight simulator. It is claimed that existing autopilots are inadequate in some respects, particularly when responding to unexpected gusts of wind close to a landing. Since human pilots capable of better performance are available as models, there is the possibility of developing better autopilots by imitation of the human performance, and experiments towards this end are described.
An extremely interesting and significant study is reported in a chapter by Michael Arbib, the longest in the book with 52 pages. The chapter has the title: "The Mirror System, Imitation, and the Evolution of Language". Observations of the brain reactions of chimpanzees have shown that, in a certain brain area, the neural response is the same when an action is observed and when it is actually performed by the animal. This seems to be consistent with the observation that humans have an innate capacity for imitation, and a neural structure showing the effect is termed a "mirror system". An additional suggestive observation is that the mirror system in chimpanzees corresponds to an area associated with language in humans.
These observations have led to a theory of the origin of language in which it is suggested that language developed from the mechanism allowing imitation of actions, initially not constituting intentional communication, but later evolving into it and finally into speech. Arbib also points out that the analysis of a complex action into constituents has some correspondence to the parsing of a linguistic communication. The theory is developed in detail and this is an important contribution both because it relates the observations to the neurophysiology and because if correct it clarifies the puzzling phenomenon of the emergence of language.
The 22 chapters inevitably contain much important material that has not been mentioned here, including several attempts to devise formalisms for the systematic study of imitation and findings from a great deal of experimentation. I think this account of a selection of topics will leave no doubt that the book is an extremely valuable and well presented and up-to-date source of information on the many facets of its topic.
Alex M. Andrew
Back to Reviews