Adaptive Interaction Design

This work is the record of a research direction developing over time, not a case study built around one prototype. It began with a broad interest in attention and flow, expanded through several connected areas of HCI, and gradually became a larger question about adaptive interaction: how should systems respond when a person's capacity, context and need for control are constantly changing?

Where the research began

My starting point was flow. I first encountered Csikszentmihalyi while studying game design and returned to the theory from an interaction-design perspective. What interested me was not the promise of a permanently frictionless experience. It was the relationship between clear goals, feedback, concentration, challenge and a sense of control. That led to a more practical question: what happens when interfaces repeatedly break those conditions?

Interruption became the first visible problem. Notifications were an accessible example, but the research quickly moved beyond notification design. An interruption can be a message, a recommendation, a request for approval, a shift in automation or any system action that competes with an existing goal. Its cost is not limited to the moment it appears. People may need to reconstruct what they were doing, recover information held in memory and regain confidence in the next step.

What I researched

The first phase covered flow, attention, interruption timing, fragmented work, cognitive load, emotional design, memory and task resumption. Each area changed the framing. Cognitive-load research showed that people are not equally interruptible at every moment. Emotional-design research showed that interruptions communicate urgency, pressure and respect, not only information. Memory research showed that reopening the same screen is not the same as recovering the same thought.

The research then expanded into interaction taxonomies, intelligent notification systems, mixed-initiative interfaces, physiological computing, workload-aware systems, proactive assistants and human oversight of AI agents. Spatial and wearable interfaces remain possible application contexts rather than the identity of the work. Across these fields, the recurring issue is the distribution of initiative: when should a system act, ask, wait, assist or hand control back?

How the direction evolved

The early work was intentionally broad. Fifteen research posts and later reflections allowed me to examine one part of the problem at a time before deciding what connected them. The common thread was not notifications, EEG or one type of interface. It was the mismatch between dynamic human capacity and systems that behave as though every moment and every user state were equivalent.

That common thread became adaptive interaction design. I use the term broadly: an adaptive system may respond to task state, behaviour, context, preference, AI inference or physiological signals. The central research problem is what the interface should do with incomplete evidence, how much authority it should have and how the person can understand or challenge its response.

How I worked

The research archive combines close reading, literature mapping, reflective writing and research-through-design. The local Zotero collection contains more than one hundred records and a large full-text library, but the dossier distinguishes between sources read closely, sources annotated for synthesis and sources kept as part of the wider landscape. This matters because the size of a bibliography is not the same thing as the depth of an argument.

Writing was part of the method rather than a final reporting step. Successive essays helped expose contradictions, connect older HCI questions with contemporary systems and record changes of direction. Visual research boards and a public presentation tested whether the argument could make sense outside my own notes. Feedback from that process showed that the field was coherent but too broad to communicate through abstractions alone.

Prototype as one research probe

The neuroadaptive UAV supervision console was built to make one adaptive loop concrete. It combined a Muse EEG signal, a webcam-derived attention proxy and task demand, then changed assistance when estimated workload rose. The operational setting created competing priorities and made timing, automation and override visible. It was a useful experimental container, not the subject of the entire research programme.

The pilot was small and the software changed between sessions, so it cannot establish that the system detected attention accurately or improved performance. Its more valuable contribution was exposing the fragility of adaptation. Different participants produced very different signal patterns; early versions rarely triggered; later versions sometimes ran too hot; faster responses came with lower accuracy. The prototype turned uncertainty from an abstract caveat into an interaction-design problem.

What I learned

A system can be technically responsive and still create a poor relationship with the user. If it acts on uncertain evidence, the authority of its response should be proportional to that uncertainty. Adaptation should remain legible, negotiable and reversible. Manual control must be practically usable, and returning after system action should not require the person to reconstruct an invisible history.

This shifts evaluation away from one accuracy score. A future study should ask whether people understand what changed, whether they trust the intervention appropriately, whether they can correct it and whether the system preserves enough continuity for them to resume their work.

What the dossier contains

The full dossier documents the research trajectory rather than presenting only a final result. It includes the origin of the inquiry, a thematic synthesis of the literature, the evolution across the post sequence, the role of public explanation, the prototype and pilot as one probe, explicit boundaries around what the evidence can support, and a research agenda for the next phase.

The current direction asks how adaptive interfaces can respond to uncertain estimates of human capacity while keeping intervention understandable, open to negotiation, reversible and subject to meaningful override. It is a working research position, designed to become more precise through future studies rather than a finished thesis claim.