Ethical Decision-Making Frameworks and Case Analysis
Scope
This domain tests process: can you structure an ethical decision reproducibly, surface the right stakeholders, and justify a recommendation without collapsing into intuition?
Frameworks to know
- Principlism as a balancing checklist (autonomy, beneficence, non-maleficence, justice) - it does not rank the principles for you.
- Value sensitive design (Friedman): identify stakeholders, values and harms iteratively across conceptual, technical and empirical phases.
- Responsible innovation (Stilgoe, Owen, Macnaghten): anticipation, reflexivity, inclusion, responsiveness.
- Ethical matrix (Mepham): rows for stakeholder groups, columns for consequences, rights and duties - it forces coverage rather than resolving conflict.
- Precautionary principle and proportionality: where harms are uncertain and potentially irreversible, act before certainty; proportionality then tests whether the measure matches the risk.
- Rest's four components: moral sensitivity, judgement, motivation and character - explaining why good people fail, not only why systems do.
- Moral intensity (Jones): magnitude of consequences, social consensus, probability of effect, temporal immediacy, proximity and concentration of effect.
A workable case method
- State the decision to be made and who decides.
- List facts and flag unknowns - do not fill gaps with assumptions.
- Map stakeholders, including those absent from the room: data subjects, downstream users, future cohorts.
- Name the values and duties in tension.
- Generate at least three options, including the status quo and a do-not-deploy option.
- Test each against the frameworks and the law, and check reversibility.
- Decide, document the reasoning and dissent, and set review triggers.
- Build contestability and monitoring in from the start.
Worked example
A city proposes predictive policing on historical arrest data. Facts: arrests correlate with patrol patterns as much as with crime. Stakeholders include residents of over-policed districts, officers, the city council and civil liberties groups. Options: deploy as proposed; deploy with safeguards on patrol allocation but no individual targeting; do not deploy and fund community alternatives. The ethical matrix shows the largest burden falling on the least powerful group, who are also absent from the decision. A defensible recommendation rejects individual-level targeting, requires published evaluation, and creates a standing review with community representation.
Traps
Jumping to a solution before naming the decision; treating a two-option framing as complete; ignoring the least powerful stakeholder because they are not in the room; treating documentation as an afterthought rather than part of the decision; expecting a framework to resolve conflicts mechanically.