Abstract

We start with an open statistics textbook, and a genuine situation we’ve been in: There's a textbook and one educator wants to add their own assessments, while another wants to add a different set. We now have two books.

Now, say that one wants their book adapted from R to Python. We would have three books.

The other likes that, and now they want the same. We repeat the work, and have four books.

Then come requests for French, examples situated in Kenya, agricultural contexts, a non-statistical audience, alternative software, different pedagogical approaches, new interactive exercises, different notation, and more.

Each request is reasonable, and each resulting resource is useful. But if every combination becomes a new book, the number of resources quickly becomes unmanageable. And when an error is corrected, an explanation improved, or new material added to the original textbook, what happens to all of those other books?

The problem started right at the beginning: We never wanted two textbooks with two different sets of assessments. We wanted one textbook with two variants.

This talk presents the technical infrastructure we are building to make that possible. A textbook can have variants for all sorts - different assessments, software, languages, notation, contexts, datasets, examples, pedagogical approaches - while those variants stay connected to the original resource. Variants can also build on other variants. So the French version can use Python, the Python version can use agricultural examples, and a community can add its own assessments on top of both.

In this talk, I will show how this works in practice, and how changes can move through this growing family of resources. Doing this at scale creates a lot of work, so I will introduce (some of) our of AI agents that help with tasks such as translating content, changing software, adapting examples, and applying changes across variants - with humans remaining firmly in the loop. Authors and educators decide what they want, subject experts review the results, and humans remain responsible for the educational materials being produced.

Ultimately, we want more people to adapt open educational resources, not fewer. If somebody needs a French, Python-based version of a statistics textbook with agricultural examples from Mali and their own assessments, they should be able to make one without making a new book completely.