Browse Books

Reviewer: James Clinton Spohrer

This book explains the situated action model of purposeful human activity and illustrates its importance in the design of interactive artifacts—intelligent machines that interact with people. The situated action model and related theories of purposeful human activity have only recently been gaining widespread attention in the artificial intelligence (AI) and cognitive science communities (see [1]). The situated action model questions the standard AI view, which emphasizes that purposeful actions are largely guided by plans (internal mental models of behaviors used to achieve goals). Instead, the situated action model tries to show that purposeful actions emerge out of essentially ad hoc situated actions (actions taken in the context of particular, concrete circumstances). In the planning model, a plan is the principal basis for actions, but in the situated action model a circumstance is the main resource giving rise to actions. For instance, when a mother is teaching her son to cook a particular dish, compare the relative importance of the recipe (plan) and the moment-to-moment kitchen context (situation) in determining what actions are being performed. Neither model completely denies the importance of the other (preconditions, opportunism, and execution monitoring are context-sensitive aspects of the planning model, while plans and hypotheses about other agents' internal mental states are resources in the situated action model), but where the emphasis is placed is diametrically opposed in the two models. The book consists of the following eight chapters: (1)Introduction (2)Interactive Artifacts (3)Plans (4)Situated Actions (5)Communicative Resources (6)Case and Methods (7)Human-machine Communication (8)Conclusion Before providing a detailed explanation of the alternative notions of plans and situated actions, Suchman illustrates the practical importance of theories of human action in the context of designing interactive artifacts. As machines become more sophisticated, the need for machines to be able to explain themselves to potential users, to monitor whether users are successfully accomplishing their tasks, and to provide interactive help to users who have problems becomes more and more important. A survey of relevant research on human-human communications in chapter 5 sets the stage for chapters 6 and 7, in which human-machine interactions are discussed. Human-machine interactions are discussed in the context of designing and evaluating an expert help system: a computer-based help system for a large, complex photocopier. Instead of providing the users with lists of instructions (plans) as is commonly done with machines today, an alternative approach was undertaken. In the new approach, the machine was equipped with a video display and sensors of its own internal state, thereby allowing it to become an active, responsive participant in accomplishing a photocopying task. This book is important for the theoretical perspective it presents as well as for the practical and methodological recommendations it offers for designing and evaluating interactive artifacts. This book should appeal to a wide range of readers, including artificial intelligence researchers, cognitive scientists, linguists, philosophers, and social scientists. System designers grappling with the issues of designing intelligent machines have the most to gain from this work. I was least convinced by the author's arguments about the limitations “in principle” of the existing AI approach in chapters 2 and 3. I most enjoyed her detailed and insightful analysis of the user-machine protocol data in chapter 7.