The Action Research cycle has four stages: Plan, Act, Observe, Reflect
These stages repeat, with each cycle building on the learning from the previous one.

Plan

The planning stage begins with identifying a specific ethical AI question or challenge. Crucially, this question should be developed collaboratively:   with the teams working on the AI system, with those affected by it, and with anyone with relevant expertise or stake in the outcome.

In ethical AI, powerful Action Research questions might include:
– “How are the people affected by our AI hiring tool experiencing its decisions, and what does that tell us about its fairness?”
– “What happens when our clinical decision-support AI makes a recommendation that the clinician disagrees with, and what does that reveal about accountability?”
– “How do communities in our target market understand and experience our AI recommendation system, and what ethical questions does this raise?”

The planning stage also involves deciding: what data will we collect? How? Who will be involved? What are the boundaries of this cycle of inquiry?

Act

The action stage involves implementing whatever change, investigation, or intervention has been planned. In ethical AI, this might mean: conducting user research with affected communities; piloting a new impact assessment process; running a structured dialogue between AI developers and the people their systems affect; introducing a new transparency mechanism and observing how it is used.

The key is that action in Action Research is deliberate and informed:   not a random experiment, but a carefully designed step that will produce observable outcomes.

Observe

The observation stage involves systematically collecting data about what happened and what it means. This might include: interviews with participants, analysis of system outputs, documentation of decisions made, tracking of unexpected outcomes.

In ethical AI, observation often reveals things that were not anticipated:   reactions that surprise, impacts that were invisible, values in conflict that no one had named. This is not a failure of the research. It is the research working. The complexity of ethical AI situations only reveals itself through careful, attentive observation.

Reflect

Reflection is where meaning is made. What did we learn? What does the evidence tell us about the ethical dimensions of this AI system? What does it tell us about our organisation’s values, practices, and assumptions? What questions does it raise that we had not previously considered?

Reflection in Action Research is not solitary. It happens in dialogue:   with the research team, with participants, with stakeholders. Different perspectives produce richer understanding. A data scientist’s reflection on the same observation as a community advocate will surface different insights, and both will be needed.

The reflection stage ends with a question: what is the next cycle? What will we plan, act, observe, and reflect on next, building on what we have just learned?

Reflection question: Take one ethical AI question you care about. Walk it through one cycle of Plan-Act-Observe-Reflect. What would each stage look like in practice?