ISLS 2026 · ICLS Paper
SmartFlash is a working AI flashcard prototype: learners bring their own material, an optional check-in targets what they don’t yet know, the AI drafts editable cards, and a short follow-up assessment confirms what stuck. The result is a small assess, generate, study, and reassess loop. Through six think-aloud sessions, the paper asks when this automation helps learners start studying, and when it quietly takes over the verification, editing, and choice that self-regulated learning depends on.
The project starts from a practical frustration. Learners know flashcards and retrieval practice work, but making usable study materials can consume the very energy that should go into learning from them. One medical student in our study spent three of his four exam-prep months just building flashcards, and never finished reviewing them.
SmartFlash is an AI flashcard tool built around learner control. Learners bring their own material, whether pasted text, an uploaded file, or a link, and the system drafts flashcards from it. General-purpose AI study tools such as Google’s NotebookLM already summarize documents and answer questions across them; SmartFlash deliberately narrows to one specific, repeated study behavior, flashcards for retrieval practice, and uses it as a focused probe for the automation and agency mechanism in self-regulated learning. Three commitments are designed in from the start: AI output is transparent, editable, and configurable.
Two lightweight assessments bracket the session, one shaping what gets generated, one verifying what was learned, making review a small, self-correcting loop rather than a one-way handoff.
The demo foregrounds that loop: bring materials into the system, convert them into reviewable cards, receive guidance about what to study next, and move into practice. The interface gives the paper a concrete object for discussing where AI should help and where learners still need control.
The paper treats SmartFlash as a formative design artifact. The workflow combined AI-seeded design hypotheses, researcher cognitive walkthroughs, and student think-aloud sessions using participants' own study materials. The analysis then compared what AI predicted with what learners actually needed, feared, and tried to control.
Reflexive thematic analysis surfaced four themes spanning the cognitive, metacognitive, agentive, and affective dimensions of studying with an AI tool.
Organizing materials and drafting cards drains the energy meant for practice. Automating it drew immediate relief; one participant called it a “life-saver.”
“I spend all my energy preparing instead of learning.” P5, Biology
Uncertainty about study direction turned into anxiety and procrastination. Metacognitive scaffolding like the Smart Study Path was strongly endorsed when it clarified without dictating.
“Without knowing what to do next, I get overwhelmed and avoid starting.” P5, Biology
Students wanted AI cards as proposals to check, revise, and discard, not final answers. Non-editable content was the single most frequent complaint. Editing is how learners take cognitive ownership.
“I wanted to edit, but it doesn’t allow it. This makes me anxious.” P1, Mathematics
Leaderboards energized some learners and shut others down; one asked for streaks over social comparison. The same feature can motivate and demotivate, so it must be adjustable.
“Higher score… makes me very anxious. It demotivates me.” P1, Mathematics
The study used a novel AI-seeded workflow: we prompted a large language model to generate design hypotheses, then validated them against researcher walkthroughs and student sessions. The LLM reliably named the functional problems, time-consuming preparation and difficulty prioritizing, but consistently missed the affective and agentive ones: the anxiety beneath the overwhelm, the need to edit and own AI output, and the way motivation diverges across learners. That gap is itself a finding: LLMs surface features, but lived experience still needs human-centered analysis.
AI should reduce the blank-page labor of creating cards while keeping learners responsible for checking and shaping them.
Recommendations should explain what they are based on, where they may be incomplete, and how learners can override them.
Practice games, study paths, and decks should be adjustable so the system can support both high-energy review and low-friction maintenance.
The useful question is not how much can be automated, but which parts of self-regulation should remain visible and practiced.
The talk follows the paper's central question: how can AI make studying easier to start while preserving the judgment, verification, and choice that self-regulated learning depends on?
Ongoing SmartFlash work continues this direction in a more formal system. This page focuses on the tested demo and the ISLS paper's central evidence: learners welcomed AI help with preparation, but still wanted verification, revision, and choice to remain in their hands.
