Technology Ethics & Governance Specialist

A team debates whether to launch a feature with a known failure rate affecting a small group of users. One person argues the aggregate benefit is large. Another argues the affected group did not agree to bear that cost. A third says the team has always shipped early and fixed forward. All three are making ethical arguments, but only the first two know it — and the third one usually wins, because it does not sound like an argument at all.

Three framings, and what each is good at

Consequentialist reasoning judges an action by its outcomes: the right choice produces the best overall result. It handles trade-offs naturally and scales to large populations, which is why it dominates product thinking. Its weakness is that aggregate benefit can conceal severe harm to a minority, since the arithmetic permits it.

Deontological reasoning judges actions against duties and rights independent of outcome: some things are wrong even when they produce good results. It protects individuals from being averaged away, and it underlies most privacy and consent law. Its weakness is rigidity when duties conflict.

Virtue-based reasoning asks what a person of good character would do, and shifts attention from decisions to dispositions. It captures something the other two miss — that ethics is mostly a pattern of habits rather than a sequence of dilemmas — but gives less traction on a specific choice.

Ship fast and iterate is a position, not the absence of one

This is the point that changes how a team argues. "We will launch and fix problems as they emerge" is a consequentialist bet with three embedded claims: that harms during the iteration period are recoverable, that the people harmed are the same people who benefit, and that speed produces enough value to outweigh the interim cost.

Sometimes all three hold. For a collaboration tool with reversible failures, iterating is genuinely the responsible choice. For a system making decisions about credit, employment, or safety, none of the three holds — harm is not recoverable, it falls on people who capture none of the upside, and the iteration is conducted on them without their knowledge.

Treating this as a neutral default rather than a claim is what prevents the argument from happening. The useful move in a design review is to name it: "we are making a consequentialist bet that the harm is recoverable — is it?" That question has an answer, and asking it turns an unexamined habit into a decision.

Failure modes

Invisible default — a shipping norm treated as the absence of an ethical position. Signal: ethics discussed only when someone raises an objection.

Aggregate concealment — net-positive metrics hiding concentrated harm. Signal: strong overall numbers alongside a specific group's repeated complaints.

Framework shopping — selecting the framing that endorses the preferred conclusion. Signal: reasoning style that changes with the desired answer.

Ethics as final gate — review after design is fixed. Signal: findings that are noted rather than acted on, because changing course is now expensive.

Explaining this to stakeholders

On why this is not philosophy class: every product decision already contains an ethical position. The only question is whether we said it out loud, because unstated positions cannot be examined.

On iterating: shipping early and fixing later is exactly right when mistakes are cheap and reversible, and exactly wrong when they are not. The judgement is about which situation we are in, not about how fast we like to move.

On aggregate benefit: a large average gain can hide a serious loss for a small group. If those are not the same people, we are not making a trade-off, we are transferring a cost onto someone who did not agree to it.