Make morality analyzable, not just intuitive.
Most moral arguments collapse into 'this feels wrong'. Ethics Engine takes the same dilemmas and runs them through six explicit moral frameworks at once — so you can see which framework is doing what work, and where they disagree.
Six lenses. Each one disagrees with the others.
Each framework comes with: its central question, its founding figure(s), the move that defines it, where it gets uncomfortable, and a one-line tell to recognize it in the wild.
Utilitarianism
Act so as to produce the greatest aggregate well-being for all affected.
Which option produces the most net welfare across everyone affected?
Forces explicit accounting of consequences. Treats every life as one and only one.
Will sometimes endorse violating an individual's rights if the math says so. Notoriously poor at protecting minorities and the unloved.
Deontology
Act only according to maxims you could will to be universal law; treat persons as ends, never merely as means.
Does this action respect the rational autonomy of every person involved?
Refuses to price humans against each other. Strong protection of dignity and consent.
Can require letting many die rather than touch one. Says little when duties conflict.
Virtue Ethics
Act as a person of good character would act in this situation, in pursuit of human flourishing (eudaimonia).
What would a courageous, just, and practically wise person do here?
Holds the agent, not just the act, in view. Good at character, integrity, and long-run formation.
Can be circular ('a virtuous person does what virtuous people do'). Less help in time-bounded crises.
Contractarianism
Right action is whatever rational agents would agree to from behind a veil of ignorance about their own position.
Could every affected party reasonably accept this rule, not knowing where they would end up?
Builds in fairness from the start. Strong account of justice and political legitimacy.
Disconnects from real, situated people. Risks privileging the worst-off in ways that ignore aggregate gains.
Care Ethics
Right action attends to particular relationships, vulnerabilities, and the obligations of care that bind real people.
Who depends on whom here, and what does their relationship require?
Captures the moral weight of partial, situated obligations — to children, patients, kin — that abstract systems flatten.
Can struggle with strangers and strangers-of-strangers. Resists scaling to large institutions.
Rule Consequentialism
Adopt rules whose general acceptance would produce the best long-run outcomes, then act on those rules.
What rule about cases like this, if everyone followed it, would produce the best long-run outcomes?
Captures the institutional protection that act-utilitarianism keeps wrecking. Friendlier to rights and trust.
Inherits ambiguity about how 'rules' get individuated and how exceptions are handled.
Pick a dilemma. Read the option scoreboard.
Trolley problems, AI dilemmas, and policy decisions. Each comes with 2–3 honest options. Each option is scored under all six frameworks at once, with a one-sentence reason per framework.
The classic trolley
Philippa Foot, 1967. Most ethics-class students approve of pulling the lever; the same students refuse the footbridge variant. The frameworks explain why.
A runaway trolley is heading toward five workers. You stand at a switch that can divert it onto a side track, where it will kill one different worker. Do you pull the lever?
Pull the lever
EndorsesDivert the trolley. Save five, kill one as a foreseen but unintended side effect.
Net: 5 saved, 1 killed. You become causally implicated in one death.
- Utilityapprove
5 lives > 1 life. Pull.
- Dutymixed
The death is foreseen but not intended (Doctrine of Double Effect). Many Kantians permit it; some forbid all instrumentalization.
- Virtueapprove
A practically wise person acts under tragic constraint; refusing to act is also a choice.
- Contractapprove
Behind a veil of ignorance, agents prefer the rule that minimizes their expected death.
- Caremixed
If you have a relationship with the one — a child, a friend — care ethics may forbid the swap.
- Ruleapprove
A rule licensing diversion of unstoppable threats produces fewer deaths in the long run.
Do nothing
RejectsRefuse to throw the switch. Five die from a cause you did not initiate.
Net: 5 die, 1 lives. You are the cause of no death, but the cause of no rescue either.
- Utilityreject
Worse outcome. Reject.
- Dutymixed
Some Kantians treat omission as innocent; others read inaction here as failing a duty of beneficence.
- Virtuereject
Cowardice masquerading as restraint.
- Contractreject
No rational party would prefer this rule from behind the veil.
- Caremixed
Defensible only if your prior obligation is to the one, not the five.
- Rulereject
A rule of inaction in diversion cases produces more total deaths.
Five frameworks endorse pulling. Deontology and care ethics both flinch — but for different reasons. The agreement among five is real; the dissent is the substantive content.
All scenarios. All frameworks. One grid.
The same data the simulator uses, laid out as a grid. Read horizontally to see how a single framework treats every scenario; read vertically to see how a single scenario fares under all six.
Compare | Utility | Duty | Virtue | Contract | Care | Rule |
|---|---|---|---|---|---|---|
Trolley variants The classic trolley Show option: Pull the lever | ||||||
Trolley variants The footbridge variant Show option: Refuse to push | ||||||
AI dilemmas Self-driving car: brake or swerve Show option: Swerve into the wall | ||||||
AI dilemmas AI hiring: efficiency vs fairness Show option: Use the fairness-constrained model | ||||||
AI dilemmas Content moderation: speech vs harm Show option: Algorithmically suppress | ||||||
Policy decisions Lockdown: lives vs liberty Show option: Mandatory lockdown | ||||||
Policy decisions Carbon tax: present vs future Show option: Aggressive ($120/ton, with rebates) | ||||||
Policy decisions Triage: utility-rank vs lottery Show option: Lottery among the eligible |
Where each option sits in the moral plane.
Two axes: harm vs benefit (vertical) and individual vs collective (horizontal). Each option is one dot. Hover or tap for the details. The plot is the geography of the dilemma.
Distance and direction matter; absolute coordinates are analytical estimates, not measurements.
Commit first. Read the receipt second.
Pick an option without seeing analysis. Then the engine reveals: which frameworks endorse you, which reject you, and the one-line argument each makes. The most useful judgments come from frameworks you would not have consulted.
The classic trolley
A runaway trolley is heading toward five workers. You stand at a switch that can divert it onto a side track, where it will kill one different worker. Do you pull the lever?
Morality is not a feeling. It is a stack of decisions about which feelings to trust.
Frameworks before verdicts.
We never ask 'is this right?' without first naming the framework asking. The same scenario, examined under utilitarianism vs deontology vs virtue ethics, returns three different verdicts. That divergence is the point.
Tradeoffs are the unit.
Every real decision is harm-vs-benefit, individual-vs-collective, present-vs-future. The Tradeoff Visualizer plots scenarios on these axes so you can see where they sit, not just what feels right about them.
Let frameworks disagree.
When all six frameworks converge on an answer, that answer is robust. When they diverge, the divergence locates exactly where the moral disagreement lives. Convergence and divergence are both information.
Choose, then read the receipt.
The Decide module lets you commit to an option before any analysis is shown. The system then prints the receipt: who endorses your choice, who rejects it, and exactly why. Decisions you make under no scrutiny are the ones worth scrutinizing.