Research

Most of what people do that’s worth understanding, they do with others and with tools. That makes it hard to study — the record of what happened is scattered, the interesting structure is in how actions relate to one another rather than in any one of them, and the finished product hides most of it.

I build the apparatus for getting at that structure — software for capturing activity, schemes for coding it, models for finding pattern in it, and tests for knowing when to believe the results. Mostly I apply it to writing and collaborative problem-solving, and lately to what happens when one of the participants is a machine.

Capturing activity

Studying process needs a record of it, and that record has usually been expensive to get. Trace records writing sessions as they happen — what people read, type, ask and paste — either in a self-contained web platform or through a browser extension that follows people into the tools they already use. It has carried studies ranging from single sessions with a few hundred students to six-week longitudinal work, and it’s increasingly taken up beyond my own projects.

Coding it

A record isn’t useful until something decides which parts of it matter and what they mean, and that decision is where the theory lives. I build coding schemes that work from both ends — the categories a literature hands you and the categories the data insists on. Lately that work has moved to language models as coders, which raises a problem earlier methods didn’t have. Their classifications shift with the prompt and with the model version, so agreement demonstrated once doesn’t stay demonstrated, and much of what I’m doing now is working out what has to be re-checked and how often.

Modelling interaction

Counting coded actions one at a time throws away how they relate to each other; treating a whole session as a single unit throws away the differences between the people in it. Epistemic network analysis and ordered network analysis model those relationships instead, and I’m a co-author of the software most people use to run them (rENA, ncodeR) as well as of the mathematics underneath. More recent work joins them to graph neural networks — more representational power, and the harder problem of keeping the results readable.

Testing the methods

The models themselves are new enough that we don’t fully know how they behave. I test them by simulation — generating interaction data where the answer is already known, running it through the pipeline, and seeing what comes back. It’s a way of establishing what a method can and can’t support before anyone rests a claim on it.

What comes of it

My work suggests that writing with AI is neither good nor bad in itself; what matters is how. What separates stronger work from weaker isn’t whether someone used a chatbot but the shape of the exchange they had with it, which the finished text doesn’t show. The next questions are scale — whether any of this holds across disciplines and institutions rather than the courses I happen to teach — and agents, which change the interaction enough that most of the apparatus needs rethinking.

Funded projects

2026–2029

Authentic Assessments for the Age of Generative Artificial Intelligence Australian Research Council, Discovery Early Career Researcher Award · Primary Chief Investigator (PI equivalent) · AUD $720,677

2024–2027

Assessments for Writing with Generative Artificial Intelligence Australian Research Council, Discovery Projects · Primary Chief Investigator (PI equivalent) of a multi-institutional investigator team · AUD $742,212

2022–2025

Large-Scale Multimodal Knowledge Management: From Organization and User Modeling to Fast Contextual Presentation DARPA, Knowledge Management at Scale and Speed program · Chief Investigator (co-PI equivalent), first-named after the lead · AUD $5,556,372

2021–2026

CELLA 2 CERES: Connecting the Center for Learning and Living with Artificial Intelligence to CERES Jacobs Foundation · Associate Investigator · AUD $435,609

2020–2023

Leading Virtual Learning Teacher Course Department of Education (Victoria) · Chief Investigator · AUD $450,391

Earlier work as a graduate researcher in the Epistemic Analytics Lab at UW–Madison (PI: David Williamson Shaffer) was supported by the National Science Foundation and the Army Research Laboratory.