I am working on errors in interaction, in their most comprehensive definition: any situation in which the system’s response does not correspond to the user’s wishes. I study how people perceive errors and their associated costs, pointing out the limits of existing methods for designing interactive interfaces, and suggesting new approaches. I have begun to create and evaluate interactive solutions to usability problems, so that error can occur in the use of an interface while minimizing its consequences.

Interaction interferences

My PhD’s first project focuses on interaction interference, a type of error that is little-studied but happens when the interface changes during the motor process of an action by the user, who can no longer interrupt or modify his gesture [2]. This distorts the system’s interpretation of the user’s intention. In today’s scenario, for the machine, an input will always be interpreted with the same level of reliability, regardless of whether the interface has changed in the immediate past.
However, characterizing the issue does not solve it: detecting interference, preventing it or correcting its consequences proves very difficult in practice, due to its immediacy. Current systems are not at all designed to allow this: it requires studying the precise chronology of perceptible interface changes, not its execution logic.

We began by developing a design space for solutions to the problem. These fall into 3 main categories: « prevent », « avoid » and « correct », from which we defined four generic behaviors designed to avoid interference.

I then conducted a controlled experiment with 20 users to evaluate these behaviors. In this study, they had to click on targets to place them in the center of the screen. Pop-up windows popped up to disrupt this primary task; they are a recurring example of a cause of interference. These pop-ups were meant to cause interferences, in order to gather feedback from participants on the event itself and on the solutions tested. The experiment was thus divided into several parts, each of which presented the participant with a different popup behavior, and a reference condition with no particular behavior. Experimental measurements reinforce the characterization of the phenomenon; interferences seemed to prevent participants from reacting to the appearance of popups. The results also showed a preference for new popups behaviors over the default [1].

[1] A. Loizeau, S. Malacria, M. Nancel. GUI Behaviors to Minimize Pointing-based Interaction Interferences. ACM Transactions on Computer-Human Interaction, In press. hal-04460441
[2] P. Schmid, S. Malacria, A. Cockburn, and M. Nancel. 2020. Interaction Interferences: Implications of Last-Instant System State Changes. ACM UIST 516–528.

Error as a concept

Through my work on the literature, I realized that the notion of error was widely used in HCI, yet little discussed theoretically. A lexical comparison enabled me to discover that the concepts covered by this notion in my field of research are not necessarily identical to those covered by its everyday use. In HCI, error is linked to ideas of control, controllability, precision, reliability and commands. Likewise, it is associated with questions of agency, or the feeling of being at the origin of and in control of one’s action. Also, since HCI has an experimental component, it is strongly linked to the notion of performance – the « error rate » is an almost systematic metric in studies.
Despite their long history, fields such as ergonomics and safety sciences are still debating the concept of error and how it should be taken into account in design processes [3]: is error a disruptive element that can be attributed to the operator, or is it an integral component of a system’s use that should be integrated into its design [1]? On the HCI side, error is generally approached as a fault, in a system analysis approach: the operator is a service provider, manipulating the input devices (keyboard, mouse, joysticks…) that enable the system to function, and can cause errors as such [2].
In my opinion, the way we think about errors conditions the way we handle them. My research will provide tools or methodological guidelines for HCI research, in order to integrate an awareness of error as it is experienced on the other side of the screen.

[1] M.D. Cooper. 2022. The Emperor has no clothes: A critique of Safety-II. Safety Science 152.
[2] P. Palanque, A. Cockburn, and C. Gutwin. 2020. A Classification of Faults Covering the Human-Computer Interaction Loop. In Computer Safety, Reliability, and Security. Springer 434–448.
[3] G. J. M. Read, S. Shorrock, G. H. Walker, and P. M. Salmon. 2021. State of science: evolving perspectives on ‘human error’. Ergonomics 1091–1114.

Discussing Errors

My thesis aims to go beyond and widen existing definitions of errors by examining the user’s lived experience. To this end, I have chosen to study the perceived cost of error, covering the emotional, material and temporal costs of error. What is the relationship between this material cost and the perceived cost? Does the perceived cost depend on the type of error? Does it depend on the person’s characteristics (profession, gender, age, etc.)? Some people adopt slower but safer strategies when the cost of the error is high. How does the possibility of an error condition action strategy?
I first explored these questions through four workshops with different populations: HCI researchers, computer scientists and regular users. Over 2 hours, each group was presented with a categorization of errors, based on a taxonomy I had developed. I am currently carrying out a thematic analysis [1] on the transcripts of these sessions, to identify common anticipations and reactions to errors. This has guided the design of an online questionnaire on the cost of error.
As this study is still in progress, I do not currently have all the results. However, the workshops have already produced two additional categories of error: problems of consent and of accessibility. This confirms the value of looking at errors from the user’s point of view, for whom these issues are important, while they would not be detected from the system’s point of view.