Research
We study how evidence from real learning can guide AI's next decision. Our work starts in education, where we can try an approach, measure what happened, and improve the system each time.
Interaction produces evidence. Evidence refines the model. The model changes what happens next.
Our approach
Performance, not preference
What people say about how they learn is useful context. What they actually do is better evidence. We measure the second.
Estimates, not labels
We never assign a fixed type. The system holds an estimate about specific teaching decisions, in specific contexts, and keeps it open to revision.
Test, then decide
An approach earns its place by improving measured outcomes like recall, application and transfer, not by sounding right.
Built on established methods
Retrieval practice, spacing, worked examples and feedback have strong research behind them. Our work is learning when and how to use them for each person.
Questions we're working on
These are open questions. We'll publish what we learn, including what doesn't work.
Getting an answer right once is weak evidence. Recalling it days later, or applying it to a new problem, is stronger. We're studying which combinations of signals are most reliable, and how to collect them without making learning feel like a test.
A model that needs months of data is not useful. One that decides after two sessions is not trustworthy. We're working on how to act early while being honest about uncertainty.
Some difficulty is productive. Making everything easier can feel better while teaching less. We want adaptation that improves long-term results, not only comfort in the moment.
People should have a say in how they learn. We're exploring how a system can respect stated preferences while still testing whether other approaches work better.
Engagement is easy to measure and easy to mistake for learning. We focus on outcomes that are harder to fake: delayed recall, application and transfer.
What we won't claim
- That Waysflow knows exactly how you learn.
- That anyone is a visual, auditory or kinaesthetic learner.
- That we can detect personality, intelligence or any medical condition.
- Results we haven't measured.
When we have results, we'll publish them with the method behind them.
Research notes
Working on learning science or AI?
We'd like to hear from researchers and educators who care about measuring learning properly.
Get in touch