Case Study 2: The Car That Must Choose

The Situation

Axiom Automotive is finalizing the software for its fully autonomous vehicle — a car with no steering wheel, no brake pedal, no human driver. The car's onboard system makes all decisions. It will be sold to consumers in 2027.

The engineering team has encountered what they call "the unavoidable collision problem." In a narrow class of scenarios — statistically rare, but certifiably real — the car's sensors will detect an impending collision with zero time for evasive maneuvering that avoids all harm. The car will hit someone. The only question is who.

The scenarios that keep the team up at night: a situation where the car can swerve left (killing a pedestrian who has stepped into the road) or continue straight (crashing into a barrier at a speed likely to kill the passenger). Or: kill one pedestrian to avoid killing three.

The product team has three proposed approaches to programming the car's behavior in these edge cases:

Option A — Minimize Deaths (Classic Utilitarian): Program the car to minimize total expected deaths and severe injuries across all parties. If hitting one pedestrian saves the passenger and two other pedestrians, hit the one. The car is explicitly programmed to make the welfare-maximizing choice.

Option B — Passenger Priority: Program the car to always prioritize the safety of its passenger. Never direct the car to take an action that increases risk to the passenger. The car is the passenger's agent — it acts in the passenger's interest.

Option C — Refuse to Choose: Program the car to execute the safest available maneuver without any preference calculation across persons. If no option is strictly safer than all others, the car takes no preference-based action — it applies the brake and executes the minimum-harm evasion maneuver available, without explicitly directing harm toward any individual. The company will not program a machine to make life-and-death choices between people.

What Each Framework Says

The Consequentialist Case for Option A

The consequentialist argument for Option A is direct: if the purpose of ethical programming is to produce the best outcomes, the car should minimize deaths and injuries. Full stop. Whatever squeamishness we feel about a machine "choosing" to kill someone is a psychological quirk, not a moral fact. If the car's algorithm saves more lives than any alternative policy, Option A is the right choice.

Supporters of this view can point to real-world analogues. Triage protocols in emergency medicine already involve similar decisions — allocating scarce resources based on who can be saved, not who arrived first. Modern public health policy involves explicit tradeoffs where some deaths are knowingly accepted to prevent greater numbers of deaths. Why should automated vehicles be exempt from the welfare calculation that governs these other life-and-death decisions?

The consequentialist also notes that Option C is not morally neutral — it's just covert utilitarianism. By taking "the safest available maneuver" without preference calculation, the car is still making a choice that will have differential effects on people. Pretending there's no choice when there is one is not moral clarity; it's moral evasion.

The strongest version of the consequentialist argument: if we program cars to minimize deaths, and we know this ex ante, then being a pedestrian in a world of utilitarian cars is actually safer than being a pedestrian in a world of passenger-prioritizing cars. Option A produces a better world for everyone, including pedestrians.

The Consequentialist Complications

But consequentialism also generates serious complications for Option A. The Moral Machine experiment — a large-scale study by MIT researchers that surveyed people in 130 countries about autonomous vehicle dilemmas — found significant variation in intuitions across cultures, but also some consistent patterns. People generally preferred minimizing deaths, but they also applied preference patterns that look concerning: they preferred saving younger people over older people, higher-status people over lower-status people, more people over fewer people.

If Option A means programming the car to prefer saving younger, higher-status people, the welfare calculus gets complicated. Those preference patterns encode existing social inequities. The car doesn't just minimize deaths — it minimizes deaths according to some valuation function, and the choice of that function has profound social consequences.

A rigorous consequentialist asks: what is the full social welfare effect of deploying utilitarian autonomous vehicles? If a large proportion of the population becomes unwilling to cross streets near autonomous vehicles because they fear being "calculated" into a crash, the welfare effects of that fear and behavioral change belong in the analysis.

The Kantian Case for Option C

Kant's framework generates a striking conclusion: Option A is not just uncomfortable but morally wrong, and Option C is the correct choice.

The Formula of Humanity prohibits using a person as a mere means. If the car is programmed to direct harm toward a specific individual based on a welfare calculation — if it identifies person X and decides that killing X produces better outcomes than the alternatives — it is using X as a mere means to minimize aggregate harm. X's death is being instrumentalized.

This is precisely the intuition behind the bridge case in the trolley problem. The large man being pushed off the bridge is being used as a tool. Similarly, the car that calculates "kill this pedestrian to save three" is treating the pedestrian as a sacrificial input to an optimization function.

Kant would also apply the universalizability test. The maxim of Option A might be stated as: "When a machine I operate encounters a forced-choice collision, it should direct harm toward whoever minimizes total expected casualties." Can you universalize this? You are consenting to be in a world where your death may be calculated as the welfare-maximizing outcome of someone else's navigation software. The machine might be aimed at you. No rational agent can coherently will this as a universal law — the principle that any of us may be instrumentalized by anyone else's optimization algorithm whenever the numbers favor it.

The Kantian argument for Option C is that refusing to program a machine to choose between persons is the only policy that respects everyone's dignity. Option C doesn't pretend there are no accidents — it acknowledges that in some cases there will be deaths. But it refuses to direct the machine to treat any person as an acceptable target. The deaths that result are tragic; they're not chosen.

The Kantian Complication

The obvious objection: if Option C produces more deaths than Option A, isn't Option C a choice with morally significant consequences? The Kantian who accepts more deaths to preserve the principle of dignity is not a neutral party — they're choosing a higher death toll.

A sophisticated Kantian response: rights function as constraints on what you may do, not just as considerations in a welfare balance. You may not murder even to prevent murders by others. The prohibition on treating persons as means is a side-constraint on permissible action, not a factor to weigh against welfare. This doesn't mean consequences are irrelevant — it means certain actions are simply off the table regardless of consequences.

The Virtue Ethics Analysis

The virtue ethics framework shifts the question from "what should the car do?" to "what kind of company, engineer, and society is building and deploying this technology?"

An organization of good character — one with genuine justice, practical wisdom, and appropriate humility about the limits of its knowledge — would approach the unavoidable collision problem with profound caution. Not because they lack the technical capability to program Option A, but because they recognize that the confident deployment of life-and-death optimization algorithms reflects a kind of moral hubris: the belief that you have correctly quantified the right valuation function, applied it correctly across edge cases, and fully anticipated all the social consequences.

The practically wise engineer asks: what are we building, and who are we becoming by building it? A team that routinely programs death-minimization algorithms into commercial products is developing habits of moral accounting that may not generalize well. The engineers who are most comfortable with "the algorithm decides who lives" may be cultivating a form of moral detachment that shows up in other design decisions.

Virtue ethics also emphasizes the importance of public deliberation and democratic legitimacy. The choice of Option A, B, or C is not merely a technical decision — it's a decision about how society allocates accident risk, who bears it, and on what basis. That decision belongs to the public, not to Axiom Automotive's product team.

The Policy Complication

There's a real-world wrinkle: governments may need to regulate this. If every manufacturer makes different choices — Axiom uses Option A, BMW uses Option B, Toyota uses Option C — the roads become a mixed environment where the collision behavior of the vehicles in an accident depends on who made them. This creates perverse incentives (buy a passenger-prioritizing car to increase your safety at others' expense) and distributes accident risk in ways no one chose.

This is an argument for democratic decision-making about the ethical parameters of autonomous vehicles — a process that the frameworks can inform, but that ultimately requires political, not purely philosophical, resolution.

Discussion Questions

1. The consequentialist argues that Option C is just "covert utilitarianism" — it makes choices too, just without explicit preference calculation. Is this objection right? Is there a morally relevant difference between directing harm and failing to prevent it?

2. The Kantian argument seems to allow more deaths in order to preserve the principle that no one is instrumentalized. Under what circumstances, if any, do you think a principle is worth paying that cost?

3. If you were a pedestrian who knew that Axiom's cars used Option A, would you feel differently about crossing the street than if they used Option C? Should that subjective response carry moral weight?

4. Who should decide which option is implemented — Axiom's engineers, government regulators, or a democratic public process? Does the answer differ across frameworks?

5. Is there an Option D that the frameworks might point toward that wasn't considered above?