Action is situated: the ethnographic challenge to the model of the user

An interface is a plan made material. A menu orders the designer's guess about the sequence a user will want. A wizard partitions one task into steps the designer believes belong together. None of those choices is neutral, and once built none is visible as an assumption.

This unit argues that the model of the user an interface encodes does not explain what people do at it, and that the dispute over it stays unresolved for a reason that is not goodwill. The situated-action critique attacks that assumption at its base. It does not claim that interfaces are imperfect. It claims that the model they encode pictures a person who holds a plan and executes it. Lucy Suchman's study of a photocopier help system is the founding case. The two sides accept different kinds of evidence as decisive, and no study here produces evidence both accept. The next unit takes up what counts as evidence.

The plan is an assumption inside the interface

Suchman (1987) watched people use a photocopier whose help system was built to diagnose faults and instruct the user. The system assumed a user in difficulty who holds a goal, follows the instruction, and arrives at the goal. What she found was different. Users read each instruction against the state of the machine in front of them and against what they were trying to do at that moment. The instruction is text, and the reader who has little of its language, or for whom reading is effortful, fits the same sentence to the setting from less. The same sentence produced different actions in different circumstances. Her formulation is exacting: "plans as such neither determine the actual course of situated action nor adequately reconstruct it" (Suchman, 1987, p. 3, quoted in Vera and Simon, 1993).

The claim is stronger than the observation that users improvise: a plan is an artefact of reasoning about action. People use it to account for what they did and to project what they will do, and it is not the mechanism that produced the action. If that is right, an interface built from a plan encodes a description its users would give, not a cause of their behaviour.

Order is an accomplishment, not a given

The intellectual source is Garfinkel (1967). His ethnomethodology holds that the people in a situation produce its ordinary sense, rather than receiving it. In his breaching experiments, members confronted with a small violation worked to restore a normal reading, showing that the normal reading was their ongoing achievement rather than a property of the scene. Suchman's users do that work at the photocopier: they make the instruction sensible in a setting the instruction cannot see.

A machine has access to a keystroke and a state variable. A user has access to what the keystroke meant, the goal behind it, and the queue of work around it. That asymmetry between what the machine can observe and what the person uses is, for Suchman (1993), the central problem of human-machine communication.

The exchange that fixed the dispute

In 1993 Cognitive Science published Vera and Simon's attack, Suchman's reply, and a further reply from the authors. The exchange is the clearest statement of the disagreement. Vera and Simon (1993) did not deny the photocopier observations. They denied that the observations require the situated conclusion. A symbolic system, they argued, can carry productions indexed by features of the situation. Reconstructing Suchman's example in those terms, they concluded that the situated programme needs the symbolic framework rather than displacing it.

We propose that the goals set forth by the proponents of SA can be attained only within the framework of symbolic systems. (Vera and Simon, 1993)

Suchman's reply concedes nothing and gives ground at once. She accepts that she never claimed planning is irrelevant to everyday activity. She restates her position: plans are "conceptual and rhetorical devices" that are deeply consequential for those who organise action in their terms (Suchman, 1993). The disagreement is therefore not about whether plans matter, but about what role a plan plays and what would show one side wrong.

Competence is setting-specific

The photocopier case is one system, and a critic can call it idiosyncratic. Lave (1988) removes that escape. In the Adult Math Project she followed the same people through grocery shopping and then tested them on closely matched arithmetic problems in a school-like format. In her report of the project, best-buy calculations made in the supermarket came out right about 98 per cent of the time. The same people averaged about 59 per cent on the school-like tests. In the aisle they decomposed prices from the largest unit down and used the shelf as a calculating device, while on paper they used written algorithms. Her conclusion, that cognition is distributed across people, tools and settings, makes Suchman's move on a task with a right answer.

Lave and Wenger (1991) carry it to learning: a newcomer becomes competent by moving towards fuller participation in a community of practice, not by acquiring stored knowledge. They develop the mechanism, legitimate peripheral participation, from apprenticeships including midwifery and tailoring. Read against the interface, an instruction is a bid to be recognised as a competent participant, and its meaning depends on whether the reader already belongs.

The unit of analysis becomes the system

Hutchins (1995) supplied the empirical case that the situated claim is more than interpretation. Studying navigation aboard ships, he showed that the crew, the instruments and the procedures perform the computational work of fixing a position together. The properties of that system cannot be inferred from those of the individuals in it, however well each is known.

His shorter study makes the point at a scale a seminar can hold. In a commercial cockpit, Hutchins (1995b) showed that the memory for speed settings is distributed across instruments, dials and procedures rather than held by the pilots. The aircraft's configuration acts as a memory that can be read back. If a display can hold a memory, the boundary between mind and interface is a design decision rather than a fact.

The critique asks for systems that support work, not a model of it

A finding that action is situated is not yet a design. Dourish (2001) draws on phenomenology and on tangible and social computing to argue that interaction is embodied and social practice. On his account meaning arises in the coupling of people and technology, not in the designer's intention.

Button and Dourish (1996) face the difficulty squarely. If ethnomethodological findings describe how a setting is organised rather than specify a system in advance, a design cannot be derived from them. Their proposal, technomethodology, builds systems that support the accountable, situated organisation of work rather than encode a model of it. Anderson (1994) separated the two uses of ethnography, as data collection and as analysis. Blomberg and colleagues (1993) show what field methods yield in projects, including practices participants had not reported.

A task model omits the articulation work, at a recorded cost

Schmidt and Bannon (1992) supply the class of work that plan-based models systematically omit. Cooperative work depends on articulation work, the effort of arranging who does what, in which order. That effort is continuous, partly invisible, and not captured by a task model of the work it makes possible. A system built to represent the task alone therefore automates the visible part and burdens the part that keeps it running.

The cost of encoding a plan that a setting does not follow has been recorded at scale. The inquiry into the London Ambulance Service (1993) examined a computer-aided despatch system that failed on 26 and 27 October 1992 and again on 4 November. Its findings read as the critique applied. The system assumed near-perfect information about vehicle location and status. The control room was rearranged on the day, so staff worked in unfamiliar positions without paper backup. Management believed that installing the system would by itself change working practices. The design carried a model of the work, and the inquiry records what the work did to the model.

Contested question: is the disagreement about evidence or about mechanisms?

Three positions are on the table. Vera and Simon (1993) hold the symbolic position. The situated phenomena are real, and a symbolic system whose productions are indexed by situational features captures them. On that account the critique is absorbed rather than victorious. Suchman (1993), with Button and Dourish (1996), holds the situated position. The account describes the setting and the machine's blindness to it, and it need not predict. Hutchins (1995) occupies neither camp cleanly, measuring properties of the distributed system in a way that can be checked without deciding whether the individual computations are symbolic. The evidence differs in weight. Vera and Simon's case is a reconstruction of one example in a notation they chose, with no new data from users. Lave's Adult Math Project is independent of both, and Hutchins offers a detailed observation of one ship and one cockpit.

The disagreement turns on what a good account must do. For the cognitive side a model earns its name by performing the task and by making a prediction that can fail. Vera and Simon's evidence is that symbolic systems already do this in demanding environments. For the situated side such a system is indexed to features a real setting does not present, and Suchman (1993) asks what it could mean to specify a situation in full. Button and Dourish (1996) draw the consequence: a design that claims to be derived from a description of work has smuggled in a model. If what people do is partly carried by the artefact, as Hutchins shows, then a model written at the level of the system may explain behaviour that a model of the individual cannot.

Two pieces of evidence would settle the disagreement, and neither is present here. The first is a setting in which each account is written down before the system is built. The prediction concerns a new task that the two accounts make differently, and it is tested on the same users. The second is a consistent unit of analysis chosen in advance, individual or system, and held to across the study. That is what Hutchins (1995) does, and why his result is the hardest to dismiss.

While it is open, build to what both accounts allow. The technomethodology proposal is defensible under either reading. So is one that names where its model is a guess and records where the setting diverges from it.

Where the argument is heading

Suchman (2007) reconsiders the argument for systems that act in their own right. She argues that how human and machine agency is configured is a design and political question rather than a technical one. Button and Dourish (1996) name the methodological stake: what a designer may legitimately claim to have learned from a setting.

The sources this unit rests on

• Suchman, L. A. (1987). Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge University Press. https://archive.org/details/planssituatedact0000such Found help-system users reading instructions against the local setting, so one instruction produced different actions. Plans account for action rather than cause it.
• Suchman, L. (1993). Response to Vera and Simon's situated action: a symbolic interpretation. Cognitive Science 17(1), 71-75. doi:10.1207/s15516709cog1701_5 Answers Vera and Simon, holding that plans are conceptual and rhetorical devices.
• Vera, A. H. and Simon, H. A. (1993). Situated action: a symbolic interpretation. Cognitive Science 17(1), 7-48. doi:10.1207/s15516709cog1701_2 Argue Suchman's phenomena are captured by a symbolic system whose productions are indexed by features of the situation.
• Suchman, L. (2007). Human-Machine Reconfigurations: Plans and Situated Actions, 2nd edition. Cambridge University Press. doi:10.1017/CBO9780511808418 Reconsiders the 1987 argument for systems that act in their own right. Argues that configuring human and machine agency is a design and political question.
• Lave, J. (1988). Cognition in Practice: Mind, Mathematics and Culture in Everyday Life. Cambridge University Press. https://archive.org/details/cognitioninpract0000lave Found the same people solving supermarket arithmetic differently from school arithmetic, doing better in the shop. Argues cognition is distributed across people, tools and settings.
• Lave, J. and Wenger, E. (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge University Press. doi:10.1017/CBO9780511815355 Argue that learning is a change of participation in a community of practice, not the acquisition of stored knowledge. Introduce legitimate peripheral participation.
• Hutchins, E. (1995). Cognition in the Wild. MIT Press. doi:10.7551/mitpress/1881.001.0001 Showed ship navigation performed by crew, instruments and procedures together, so the unit of analysis is the system.
• Hutchins, E. (1995b). How a cockpit remembers its speeds. Cognitive Science 19(3), 265-288. doi:10.1207/s15516709cog1903_1 Showed cockpit speed memory distributed across instruments, dials and procedures, with the aircraft's configuration acting as memory. Makes display design a cognitive matter.
• Dourish, P. (2001). Where the Action Is: The Foundations of Embodied Interaction. MIT Press. doi:10.7551/mitpress/7221.001.0001 Argues interaction is embodied and social practice, with meaning arising in the coupling of person and technology.
• Button, G. and Dourish, P. (1996). Technomethodology: paradoxes and possibilities. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '96), 19-26. doi:10.1145/238386.238394 Show that ethnomethodological findings describe a setting rather than specify a system. Propose technomethodology, building systems that support the accountable organisation of work.
• Anderson, R. J. (1994). Representations and requirements: the value of ethnography in system design. Human-Computer Interaction 9(2), 151-182. doi:10.1207/s15327051hci0902_1 Argues ethnography informs the representational scheme a system imposes, not requirements data, and sets the conditions under which findings transfer.
• Blomberg, J., Giacomi, J., Mosher, A. and Swenton-Wall, P. (1993). Ethnographic field methods and their relation to design. Participatory Design: Principles and Practices (Schuler and Namioka, eds.), Lawrence Erlbaum. https://www.taylorfrancis.com/chapters/edit/10.1201/9780203746349-7/ethnographic-field-methods-relation-design-blomberg-giacomi-mosher-swenton-wall Show what field methods yield in design projects, including practices participants had not reported.
• Schmidt, K. and Bannon, L. (1992). Taking CSCW seriously: supporting articulation work. Computer Supported Cooperative Work 1(1-2), 7-40. doi:10.1007/BF00752449 Argue cooperative work depends on articulation work, arranging joint activity, which task models omit.
• Garfinkel, H. (1967). Studies in Ethnomethodology. Prentice-Hall. https://archive.org/details/studiesinethnome0000garf Set out ethnomethodology, showing through breaching experiments that the sense of a situation is an ongoing accomplishment.
• London Ambulance Service Inquiry. (1993). Report of the Inquiry into the London Ambulance Service. South West Thames Regional Health Authority. http://www0.cs.ucl.ac.uk/staff/A.Finkelstein/las/lascase0.9.pdf Found the computer-aided despatch system failed in October and November 1992, with near-perfect information assumed about vehicles, a rearranged control room, incomplete training and management belief that installation alone would change practice.