Meanings of Open Artificial Intelligence in Education

“Open AI” can mean many different things – open weights, open training data, open licences, open governance, open access to AI tools, or transparency in how they’re used in decision making. The differences matter a great deal when the context is education. In this project, we set out to map the range of meanings of “openness” so that conversations about open AI in education can be better informed about what is actually being claimed.

I led this collaboration as Principal Investigator since April 2024, working with colleagues in Charles Sturt University in Australia, and The Open University in the UK.

We conducted phenomenographic interviews with 24 experts in artificial intelligence and open education, exploring how they understand openness in this context and what they see as the benefits, risks, harms, and next steps for open AI in education.

Understandings turned out to vary widely. Some participants see AI as fundamentally incompatible with openness; some think it possible in principle but not yet achieved; others point to concrete examples they consider open. Most held complex views. We identify a framework of increasing complexity, running from a binary “open or closed” view, to a linear scale of “how open”, to a multi-dimensional view distinguishing what is open – data, code, model – and to what extent, and finally to a contextual understanding that incorporates whom those aspects are open to and for what purposes. A recurring distinction is between technical, “inside the system” openness such as open-source code, and practical, “outside the system” openness such as open educational practices.

Our team shared preliminary findings in a panel discussion at Open Education Global 2024 and the final findings at Open Educational Resources 2026. A research article written with my co-authors Vi Truong (Charles Sturt University and University of Melbourne) and Robert Farrow (The Open University) is currently under review.