A practical guide to designing an Action Research cycle for ethical AI:

1. Choose a bounded, specific question
Action Research works best when focused. “Ethical AI” is too broad. “How do people who are denied loans by our AI scoring system understand and experience that decision, and what does this reveal about the system’s fairness and transparency?” is actionable.

2. Identify your collaborators
Who needs to be part of this inquiry? Whose knowledge is essential? Whose participation is ethically required? Build your research team to include diverse perspectives:   particularly those most affected.

3. Plan your first cycle conservatively
Start small. A first cycle is often exploratory:   designed to learn enough to design a better second cycle. Resist the temptation to resolve everything in one pass.

4. Build in genuine reflection time
Reflection is the most frequently sacrificed part of Action Research, because it is the least visible. Protect time for it. Schedule it. Treat it as non-negotiable.

5. Document as you go
Action Research produces learning that is easily lost if not documented. Keep a record of decisions made, observations noted, and reflections reached at each stage.

6. Share findings with participants before acting on them
This is both methodologically sound and ethically required. Participants may correct your interpretation, add nuance, or identify implications you had not considered.

7. Plan the next cycle before declaring the work done
Action Research does not end. Each cycle should conclude with a clear plan for the next.
The question is not “are we finished?” but “what have we learned, and what does that mean we should investigate next?”