Transcript suchman-9
Study & Conclusions Perspectives on Face-to-face Interaction • Success at anticipating the actions of the other – Implies need for • Model of user that supports prediction of actions • Recognition of actions predicted • Prescription of appropriate response • Success at responding to unanticipated actions of the other – Implies need for • Maximizing sensitivity to actions taken by minimizing predetermined sequences of machine behavior Expert Help System • Engineering an appropriate response – Limitations • Designers’ ability to predict user’s actions • System’s access to and ability to make sense of user actions • Goal: more situated than documentation, but reusable Plans and Predictable Action • Backward chaining used for locating step within a plan – Allows for skipping steps whose outcomes are already detected • Problems with intermediate steps in plans – Human and machine have interpretations of current state • No history of actions – could have used one The User’s Resource: Situated Inquiry • The procedural assumption of always providing next step does not allow for repair and abandonment • Context of “help” requests ambiguous – What level of plan should be considered? • Designer cannot predict all methods of failure – Thus, cannot predict all types of help needed • Instructions and physical objects mutually reinforce one-another’s interpretation Conditional Relevance of Response • Conversational paradigm – New response indicates successful action – No response indicates incomplete action – Repeated response is ambiguous • Could be iteration or trouble to be repaired • Slow response can be interpreted as no response Iteration vs. Trouble • In human-human interaction loops/problems are identified/escaped from via – Exploratory questions (“Hello. Hello. … Are you there?”) – New explanations of the same idea – Assertions of completion • How can the system know the difference? • How can the user access the plan? Communication Breakdowns • False alarm – Response to correct action interpreted as failure – May occur • When there are conflicting indicators • Due to different interpretations of plan/task • Garden Path – Incorrect action not discovered until later • Losing context of original problem – Machine can interpret action as correct for some alternative path • Users can assume they know the process without machine • Trivial breaches of understanding can become “fatal” Lessons from Theory/Studies • Situation is not something to be avoided by systems determining action • Mutual intelligibility of actions relies on situation • Communication practices maximize use of context • Face-to-face communication includes resources for recognizing/correcting trouble Considerations for Interface Design • Asymmetry of human-machine communication – Increase machine’s access to user actions/context – Make clear the limits of the machine – Compensate for lack of access with computation • User models & coaching – Differential modeling and deviation from ideal • Depends on match and boundaries of domain – Diagnostic inconsistencies • Machine/design must consider potential for own mistakes – Local vs. global interpretations • Separate to support short-term/long-term goals – Constructive use of trouble • Can be used as an opportunity to “teach” On Communication • “Communication … is not a symbolic process that happens to go on in real-world settings, but a real-world activity in which we make use of language to delineate the collective relevance of our shared environment.” On Plans • “While plans can be elaborated indefinitely, they elaborate actions just to the level that elaboration is useful; they are vague with respect to the details of action precisely at the level at which it makes sense to forego abstract representation, and rely on the availability of a particular, embodied response.”