Pillar 7: Manage Ethics Always¶

"Ethics is not compliance and it isn't moralizing — it's phronesis."

The Argument¶

Organizations typically manage AI ethics through one of two insufficient approaches. The first is compliance: checklists, policies, review boards, and governance frameworks that reduce ethical reasoning to procedural box-ticking. The second is moralizing: abstract declarations of principles ("we are committed to fairness, transparency, and accountability") that function as reputational insurance but produce no meaningful behavioral change.

Neither approach is adequate because neither engages with the actual nature of ethical challenges in AI adoption. These challenges are not primarily about following rules or affirming values. They are about making good judgments in particular situations where rules underdetermine the right course of action, values conflict with each other, and consequences are uncertain.

Aristotle called this capacity phronesis — practical wisdom. Phronesis is the intellectual virtue of knowing what to do in specific circumstances, not by applying universal rules but by perceiving the morally salient features of a situation, weighing competing considerations, and exercising judgment (Aristotle, Nicomachean Ethics, Book VI). It is cultivated through experience, reflection, and exposure to exemplary practice — not through training modules or policy documents.

AI adoption generates a continuous stream of situations requiring phronesis. Should this AI tool be deployed when it performs well on average but poorly for a specific demographic subgroup? The compliance answer (does it meet the bias threshold?) and the principled answer (fairness requires equal treatment) both underdetermine the decision. The phronetic answer depends on the specific context: who is affected, what alternatives exist, what the consequences of action and inaction are, and what precedent is being set.

Other examples: When should a team override an AI recommendation they believe is wrong? How transparent should an organization be about AI's role in a decision when full transparency might undermine legitimate business interests? When an AI system produces an outcome that is legal and profitable but feels wrong, what weight should that moral intuition carry? These are not compliance questions. They are judgment questions.

Managing ethics always — the "always" is deliberate — means building phronesis as an organizational capability, not outsourcing ethical reasoning to a policy document or ethics board. It means creating the conditions in which employees at every level can recognize ethical dimensions of AI-related decisions, reason about them with sophistication, and act on their judgments without fear.

This requires several structural commitments. First, ethical reasoning must be embedded in operational workflows, not siloed in a separate review process. Ethics is not a gate; it is a lens applied continuously. Second, organizations must cultivate moral perception — the ability to notice when an AI decision has ethical stakes. Most ethical failures are not failures of reasoning but failures of recognition: no one noticed there was an ethical question. Third, organizations must create psychological safety around ethical dissent. If raising an ethical concern is career-risky, phronesis is irrelevant — it will not be exercised.

The relationship between ethics and trust (Pillar 3) is reciprocal. Ethical behavior builds trust; trust creates the safety required for ethical dissent. An organization that manages ethics always is simultaneously building the trust reservoir that enables adoption.

In Practice¶

A technology company replaced its AI ethics review board — which had become a bottleneck and a rubber stamp — with embedded "ethics partners": experienced practitioners assigned to product teams whose role was not to approve or block decisions but to raise ethical questions in real time. These partners were trained not in compliance but in ethical reasoning: how to identify stakeholders, surface hidden assumptions, anticipate second-order consequences, and facilitate moral deliberation within the team. The shift from gate to lens increased both the frequency and the quality of ethical engagement.

A financial institution developed an "ethical pre-mortem" practice for AI deployments. Before launch, the team conducts a structured exercise: imagine this system has been deployed for 12 months and has caused significant harm. What went wrong? Who was harmed? What did we fail to anticipate? The pre-mortem format — borrowed from project management (Klein, 2007) — makes it psychologically safe to raise concerns because participants are describing a hypothetical scenario, not criticizing a colleague's work. The practice has surfaced issues that neither compliance reviews nor bias audits detected.

A public sector organization created an "ethics case library" — a curated collection of real situations in which AI deployment raised ethical questions, documenting how they were handled, what was learned, and what the organization would do differently. This library serves the same function as case law in legal reasoning or case studies in medical education: it develops phronesis through exposure to exemplary (and cautionary) practice, building the organization's collective moral reasoning capacity over time.

The 4×1 Matrix¶

Diagnostic Questions¶

1. Ethics location: Where does ethical reasoning happen in your AI adoption process? If the answer is "at the review board" or "in the policy," ethics is siloed rather than embedded. Phronesis requires continuous exercise, not periodic review.
2. Moral perception: Can you cite a recent example where someone in the organization noticed an ethical dimension of an AI decision that was not flagged by existing compliance processes? If not, your ethical sensing capability may be atrophied.
3. Dissent safety: If a junior employee raised an ethical concern about a senior-sponsored AI initiative, what would happen? If the honest answer involves career risk, your organization lacks the safety conditions for phronesis.
4. Beyond compliance: Does your organization distinguish between "this AI application is compliant" and "this AI application is ethical"? Compliance is a floor, not a ceiling. Phronesis operates above the floor.

← Back to Change Agility Overview