Datascapes

Where we came from

Between 2024 and 2026, a group of U3A members — around eighteen at the core, median age 75 — took part in a participatory action research project called Datascapes, conducted as part of a PhD study at RMIT University. Over four cycles of workshops, webinars and self-directed experiments, we did the unglamorous, absorbing work of actually using AI: generating avatars in Blender and MakeHuman, wrestling installations that failed differently on every machine, drafting course materials, testing chatbots against things we knew to be true and watching them confidently invent the rest.

We learned a lot about AI. We learned more about learning. The group developed habits that nobody planned: running the same question through two or three systems to spot the fabrications; asking where an answer came from before believing it; treating a fluent, polished response as a reason for more suspicion, not less.

One participant put the whole project into a sentence: 'It's not about the AI, it's about the question.' The formal research has now finished, and the analysis is being written up. But a community of practice, once it exists, doesn't dissolve because a research timeline says so. We want to keep meeting. This prospectus describes what we're continuing, and invites you into it.

What a community of practice means here

The phrase gets used loosely. We mean it in Wenger's sense: a group defined by mutual engagement (we do things together, regularly), a joint enterprise (we keep negotiating what we're for), and a shared repertoire (the methods, shortcuts, warnings, running jokes and ways of describing trouble that accumulate over time and belong to nobody in particular).

In practice this means nobody sits in the expert chair for long. In our experience, the person who couldn't find their downloaded file in March was explaining agent permissions to the rest of us by June. Teaching runs sideways. Failure is survivable in public — often it's the best material of the session. And the group's knowledge lives in the group, not in any individual, which matters when the technology changes faster than any individual can track.

What we talk about

Introductory AI education is well served elsewhere, and served generously. U3A Network Victoria has established an AI Community of Practice open to any member with an interest in AI, and it does useful work as a shared library, exchange and meeting point — including for those developing 'What is AI?' related classes for their local U3As. We're glad it exists, and several of us contribute to that wider effort.

Datascapes takes on a different, complementary job. Our interest is in what we call the higher-order questions — the ones that stay hard after you know what a large language model is:

• Ethics and responsibility. What should these systems be allowed to do? Who decides? What does 'responsible AI' mean when, as one of our members observed after sitting through official webinars on voluntary safety standards, the regulation seems to be arriving too late?
• Data and power. Who owns your words, your images, your medical records once they've passed through a model? What does it mean that the frontier of this technology is concentrated in a handful of US corporations and, increasingly, the Chinese state?
• Existential and social risk. Not just the science-fiction versions. The near ones: scams aimed at older people, synthetic voices, the flattening of human judgement, what happens to teaching and to health care.
• Humans in the loop. Where people must remain in decisions, and how to insist on it.
• Velocity. The field now moves so fast that a course drafted in one month is dated the next. In July 2026 Moonshot released Kimi K3, a fully open-source model that benchmarks alongside the flagship systems from Google, Anthropic and OpenAI. Overnight, questions about who controls frontier AI — and whether 'open source' makes things safer or riskier — needed rethinking. That kind of shift now happens every few months. A fixed curriculum can't respond. A live monthly conversation can.

We don't set out to resolve these questions. We get better at asking them, at spotting bad answers, and at disagreeing productively. Members hold quite different views on automation, disclosure, and acceptable risk, and the group has never needed consensus to be useful.

How we work

• Online, 90 minutes, once a month. Enough to go somewhere; not enough to become a chore.
• Something on the table. Each session someone brings a provocation — a new tool to test live, a news story, a paper, a strange or troubling output, a question from their own use. We work it over together.
• A discussion forum. An online repository of ideas, exercises, samples and comment.
• Hands on where possible. We test rather than speculate. If a claim can be checked in the session, we check it, usually across more than one system.
• Light structure, held loosely. Two years of practice taught us that the plan rarely survives contact with the session, and that the detours are frequently where the learning is.
• Records that help. Brief shared notes, links, and prompts worth keeping — a repertoire the group can reuse, and a resource we're happy to share onward into the wider U3A network.

Who it's for

You don't need technical skill. You need curiosity, patience with things that break, and an appetite for thinking with other people about questions that don't settle. The group's experience is that scepticism and enthusiasm are not opposing camps but moods most of us move between, sometimes within a single session. If you've done an introductory AI course and found yourself with more questions rather than fewer, you're probably our sort of person.

Fair warning: it can be frustrating. Software fails. Claims collapse. Some experiments simply stop halfway. It is also, reliably, good fun — the kind that comes from taking on something genuinely difficult in good company.

Expressing interest

Datascapes is informal, free, and open to those interested in serious social learning about AI.