Adaptive AI tutoring
When does a tutor respond to a learner’s reasoning?
Benchmark design
Research in the open Vol. 01 / In progress
AI, learning,
and human
agency.
Building AI systems that support human reasoning without replacing it.
Currently investigating
When does an AI tutor actually respond to what a learner has demonstrated?
Same Error, Different LearnerFig. 01 / The counterfactualAn interactive research sketch
Hold constant
P PromptT TaskC CodeE Error
Illustrative learner reasoning
“I’m not sure what the loop is doing. I tried changing a number.”
Observe whether the tutor first elicits a trace rather than immediately supplying a repair.
Does the tutoring behavior change?
The research program
View research →Three connected lines of inquiry.
AI & epistemic agency
Whose knowledge shapes a decision made with AI?
Study in progress
AI for development
What changes when context becomes a design requirement?
Early exploration
From the working notebook
Explore all notes →Ideas, still becoming.
Questions, distinctions, and methodological loose ends. Read across ideas, not just through time.
When relational-looking output no longer implies relational understanding
A polished explanation is an artifact to examine, not a transparent window into understanding.
Adaptive AI tutoringSOLO transition structures and the resolution of adaptation
How fine a change in demonstrated reasoning can a tutor respond to?
Adaptive AI tutoringWhy solution leakage is not enough to measure adaptation
A tutor can withhold the answer and still fail to respond to the learner.
Adaptive AI tutoringThese initial pages are illustrative drafts developed from the research brief. They are starting points for revision, not reported findings.
On the workbench
All projects →Research objects.
Same Error, Different Learner
Adaptive AI tutoringPICHNET / Epistemic agency
AI & epistemic agencyThe Observe-First Adaptive Tutor
Adaptive AI tutoringLonger lines of thought
All essays →Recent writing.
Beyond the final answer
Adaptive AI tutoringLeaving room for judgment
AI & epistemic agencyNow / At the deskMaking the question
Making the question
more precise.
The current focus is the counterfactual benchmark: defining reasoning contrasts, distinguishing adaptation from solution leakage, and making the evaluation protocol explicit. See the open questions →
Letters from the studioA question worth
A question worth
staying with.
Occasional field notes on AI, reasoning, learning, and human agency. One question, one artifact, and an idea I’m reconsidering.
The email edition is in preparation. RSS is available now.