Plans and Situated Actions

In the AI Story

The book's target was the dominant AI paradigm of the 1980s: expert systems and goal-based planning, which assumed intelligence consists of representing goals, forming plans, and executing them. This paradigm organized billions of dollars of research funding at Xerox PARC, MIT, Carnegie Mellon, and the defense-funded AI laboratories. Suchman's intervention was unusual in arriving from anthropology rather than computer science, and in grounding its critique in empirical study of actual users rather than in philosophical argument about the nature of mind. The combination proved devastating.

The central empirical material was a corpus of video recordings of pairs of users attempting to make double-sided copies on a Xerox photocopier equipped with an expert help system. Users consistently misread the system's prompts, formed hypotheses the system's designers had not anticipated, and proceeded by improvisation rather than by executing the plan the system was trying to support. Suchman's analysis — informed by ethnomethodology and conversation analysis — revealed that the users' activity was genuinely intelligent, but its intelligence lay in their responsive, adaptive engagement with the specific situation, not in any internal plan.

The book triggered the formal exchange between Suchman and Herbert Simon in Cognitive Science in 1993, in which Simon and Aran Vera argued that situated action could be reabsorbed into a sophisticated planning framework. Suchman's response sharpened the distinction: the question was not whether situated action could be formally described, but whether the description captured what was actually happening when a competent agent acted intelligently. The debate remains unresolved and now organizes much of the critical discourse around large language models.

The book's influence extended far beyond AI. It became foundational to human-computer interaction, workplace studies, participatory design, and STS. Its method — close ethnographic attention to how specific people actually use specific technologies in specific settings — provided an alternative to the abstracted modeling that dominated both computer science and cognitive psychology. Suchman's 2007 expansion Human-Machine Reconfigurations updated the framework for an era of ubiquitous computing and autonomous systems.

Origin

Suchman's dissertation research was conducted during her early years at Xerox PARC, where she had been hired in 1979 as an anthropologist to help the laboratory understand why users struggled with machines the engineers considered well-designed. The methodological traditions she brought — ethnomethodology from Harold Garfinkel, conversation analysis from Harvey Sacks, interactional sociology from Erving Goffman — were alien to the computer science environment. The book's power comes partly from this cross-disciplinary collision.

The decision to publish the analysis as a book rather than as internal reports reflected Suchman's growing conviction that the argument had implications beyond product design. The book's audience ultimately included AI researchers, philosophers of mind, sociologists of science, and the emerging community of scholars attempting to articulate alternatives to the computational theory of mind.

Key Ideas

Plans are representations of action. They precede or follow action but do not constitute it. The plan describes what should happen; the action is what actually occurs when an agent meets specific circumstances.

Intelligence is responsive, not prescriptive. Competent action arises from the practitioner's engagement with what she actually encounters, not from the execution of rules formed in advance.

The photocopier was not arbitrary. Suchman chose a mundane technology deliberately — to show that even simple task domains reveal the improvisational character of human action that AI's exotic domains (chess, theorem proving) systematically obscured.

Communication is situated too. What a machine's message means to a user depends on the user's interpretive work, which draws on the full apparatus of human social intelligence — a resource the machine cannot reciprocate.

The asymmetry problem. Human users bring rich interpretive capacities to machine interactions; machines bring procedural responses. The interaction looks like conversation but is structurally asymmetric, and the asymmetry deepens rather than closes as AI sophistication grows.

Debates & Critiques

The book generated lasting controversy. Computational cognitive scientists argued Suchman had overstated the gap between planning and situated action and that sufficiently sophisticated planning models could accommodate her observations. Others worried that her framework provided no positive theory of intelligence, only a negative argument against the reigning paradigm. Suchman's response — that theory-building should follow close attention to actual practice rather than precede it — has become a methodological touchstone in STS and critical AI studies.