After the UX Graz talk, I was left with a useful tension. The research had developed into a broad field of connected questions, but the next step needed to be more concrete. My professor encouraged me to experiment rather than force the work into a final thesis topic too early. This semester is a chance to build something, test one direction and learn from what does not work.
What I want to experiment with is neuroadaptive interaction.
This does not mean that neuroadaptive interfaces have become my final thesis direction. Adaptive interaction design is still the larger area I am exploring. My interests include interruption, recovery, cognitive workload and the changing relationship between people and increasingly proactive systems. Neuroadaptive interaction gives me one way to bring these interests together in a form that can be experienced rather than only described.
I have been especially drawn to experimental HCI research from the 1990s and early 2000s. Before current discussions about AI agents and intelligent assistants, researchers were already asking how systems might respond to changing human capacity. Work on mixed-initiative interfaces, adaptive automation and physiological computing treated interaction as a continuing feedback loop. The system observed something about the user or task, adjusted its behaviour and then observed the consequences.
Pope, Bogart and Bartolome’s (1995) biocybernetic research is a good example. Their system used an EEG-derived engagement index to change the balance between manual and automated work in a flight-deck task. The experiment belongs to a specialised context and its measurements should not be treated as a direct model for everyday interfaces. Still, the interaction question remains valuable. If a system changes its level of support according to an estimate of the user, how should that change be communicated and how can the user remain in control?
Horvitz’s (1999) work on mixed-initiative interfaces approaches a similar problem without making physiological sensing the centre. A system may be able to suggest an action or offer help, but that does not mean every moment is suitable for intervention. It has to decide when to act, when to ask and when to remain quiet. These choices distribute initiative between the person and the system.
The historical work interests me because it is experimental in a direct way. It does not only propose that interfaces could become more responsive. It constructs situations where the relationship between user state, task demand and system behaviour can be observed. This is the kind of step I want to take now.
The experiment is not about proving that a computer can read somebody’s mind. Neuroadaptive systems work with indicators and inferences. EEG, gaze, behaviour and task performance can suggest that something may be changing, but they do not provide uncomplicated access to attention or mental workload. Fairclough (2009) describes a biocybernetic loop as a process of sensing, inference, adaptation and feedback. The inference stage is where uncertainty enters. One physiological change can have several possible causes, and the same cognitive state can appear differently across people.
Treacy Solovey et al. (2015) make the design gap explicit when they write that “the interaction techniques and design decisions for their effective use are not well defined” (p. 1). Their work suggests that physiological signals may be more appropriate as implicit context than as direct commands. Adaptations should be modest, confidence should matter and changes should remain reversible.
This shifts my attention from the accuracy of a single sensor toward the behaviour of the interface. What should a system do when it estimates that workload may be rising? Should it reduce information, delay a low-priority interruption or increase automation? How much should it explain? What should happen when its estimate is wrong?
To examine these questions, I need a setting where task demand and adaptation can both become visible. A simple interface would not create enough pressure to make the experiment meaningful. If a person is only watching one value or completing one predictable action, there is little reason for the system to adapt. The context needs several competing demands without becoming impossible to understand.
This is why I became interested in complex interfaces as an experimental environment. Control rooms, monitoring systems and operational dashboards require people to maintain awareness across several sources of information. Some events are urgent while others can wait. The interface may need to guide attention without hiding information or taking control away from the operator.
Complex interfaces also make cognitive load a design issue rather than an abstract score. Information density, competing priorities and time pressure can all affect what the user is able to process. An adaptive system could respond by foregrounding the most relevant event, reducing noncritical detail or temporarily taking over a routine action. Each of these adaptations can help, but each can also create new problems. Removing information may reduce visual load while weakening situation awareness. More automation may lower immediate demand while making it harder for the user to understand what the system has done.
The experiment therefore needs to be about control negotiation, not automatic optimisation. The user should be able to inspect why the system changed, reject the change or disable adaptation. An uncertain estimate should never give the system unquestionable authority. If the prototype adapts, it should do so in a way that keeps its reasoning and limitations visible.
Interruption and recovery remain part of this experiment. In a complex interface, the system has to decide whether a new event deserves immediate attention. Adamczyk and Bailey (2004) showed that interruption timing affects performance and emotional response. Chen et al. (2025) demonstrate that recovery cues can support users after an interruption by directing them toward previous or upcoming task information. These findings suggest several interactions that can be tested without pretending to solve cognitive workload as a whole.
The first version can begin with simulated signals. This allows me to test what the interface does before depending on hardware. Later, a Tobii eye tracker and a Muse headset may provide live inputs, but they should not define the project. A real sensor cannot rescue an adaptation that is confusing, intrusive or impossible to challenge.
For a concrete implementation, I am using a drone-triage scenario. The user monitors several semi-autonomous drones while managing interruptions and changing priorities. This setting provides a complex interface where workload, timing, assistance and recovery can be represented clearly. The system can defer a lower-priority alert, foreground a risk or provide a cue after the operator returns to the main task.
The drone scenario is not the research decision. It is a practical container for the experiment. Another operational context might eventually work better. What matters is that the interface gives me a way to test neuroadaptive behaviour under visible task pressure while preserving explanation and human override.The aim is to make one part of the research concrete enough to question, observe and revise.
References
Adamczyk, P. D., & Bailey, B. P. (2004). If not now, when?: The effects of interruption at different moments within task execution. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 271-278). ACM. https://doi.org/10.1145/985692.985727
Chen, Y., Zhang, C., Fang, W., & Ma, J. (2025). The effects of cues on task interruption recovery in a concurrent multitasking environment. Scientific Reports, 15, 25992. https://doi.org/10.1038/s41598-025-09358-4
Fairclough, S. H. (2009). Fundamentals of physiological computing. Interacting with Computers, 21(1-2), 133-145. https://doi.org/10.1016/j.intcom.2008.10.011
Horvitz, E. (1999). Principles of mixed-initiative user interfaces. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 159-166). ACM. https://doi.org/10.1145/302979.303030
Pope, A. T., Bogart, E. H., & Bartolome, D. S. (1995). Biocybernetic system evaluates indices of operator engagement in automated task. Biological Psychology, 40(1-2), 187-195. https://doi.org/10.1016/0301-0511(95)05116-3
Treacy Solovey, E., Afergan, D., Peck, E. M., Hincks, S. W., & Jacob, R. J. K. (2015). Designing implicit interfaces for physiological computing: Guidelines and lessons learned using fNIRS. ACM Transactions on Computer-Human Interaction, 21(6), Article 35, 1-27. https://doi.org/10.1145/2687926
