Shared research case · Planned
From AI assistance to independent explanation
- Duration
- 3 weeks
- Dates
- To be announced
- Research lead
- Esther Bawo Tsotso
Central research questionAfter equal allocated study time, what differences in unaided explanation and delayed transfer are observed following Flowst Airs and source-based self-study, and which Airs interactions appear to support or obstruct independent reasoning and learner control?
Study approach
One case, seven connected lenses.
Methods note
Week 1 is an internal protocol and system audit. In Week 2, each researcher guides one or two adult external participants through counterbalanced Airs and source-based self-study conditions. Week 3 compares immediate and 48-hour unaided responses, negative cases, learner accounts, system behaviour, and independent human ratings.
Safeguards
Team dry runs are not treated as learner-effect evidence. The public journal contains de-identified methods, observations, limitations, and findings only. Participant identities, contact details, consent records, raw transcripts, raw audio, and identifiable research data do not enter the publishing system.
Connected investigations
Distinct contributions to a shared question.
Esther Bawo Tsotso · Adaptive learning systems
Turning learning principles into executable teaching decisions
When should Airs question, hint, explain, correct, revisit the source, or step back?
Evidence and method
Evidence sought. The relationship between intended teaching decisions, actual conversations, support withdrawal, and learner revisions.
Method. Conversation review and decision mapping across planned learning moments.
Adebayo Chike · UX and human-centred research
Designing voice study for learners unfamiliar with AI
Can learners understand the interaction, recover from confusion, and retain control?
Evidence and method
Evidence sought. Observed confusion, interruption and recovery patterns, comfort, and clarity about next steps.
Method. Observation, short interviews, and interaction walkthroughs.
E.U. Eniang · AI evaluation and data governance
Evaluating instructional reliability beyond fluent responses
Does Amina elicit an independent attempt, and are Kai’s claims supported, qualified, and faithful to the approved evidence?
Evidence and method
Evidence sought. Source grounding, unsupported claims, missed misconceptions, premature completion, transcription errors, latency, and instruction adherence.
Method. Structured audits of Amina and Kai against approved source material, saved evidence, and the objective ledger.
Gbolabo Onasanya · Educational practice and implementation
What makes AI-generated learning evidence useful to teachers?
Weekly record
3 weeks, published as the work develops.
Only published updates appear publicly. An absent entry is not presented as evidence.
The public weekly record begins when the first update is published.