Should an autonomous car protect people who caused the danger by breaking the rules?
Pure injury minimisation says yes — it counts bodies, not fault — which means programming the car to sacrifice the innocent in order to spare the people who created the hazard.
The scenario: a group of six people walks onto the road without paying attention to traffic, and an autonomous car cannot avoid them without seriously injuring its own occupants or a pedestrian on the pavement. A purely injury-minimising programming would accept that if it is the only way to avoid more severe injuries to a larger number.
Two objections follow, and they are of different kinds:
- Fairness. It seems highly unjust for the costs of one actor's risky misconduct to be artificially imposed on another who did nothing wrong. The car would be explicitly programmed, in an emergency, to sacrifice "innocents" in order to protect the actual causes of the accident from the consequences of their own misbehaviour. Is that morally justifiable?
- Incentives again. Is it really desirable that, for a pedestrian, physical proximity to the largest group promises more safety than complying with the rules? Would it not make more sense to programme autonomous vehicles so that rule-compliant behaviour guarantees the greatest possible safety?
The incentive worry is not hypothetical. Attention has been drawn to the danger of hazardous "games" or bets — and, less maliciously but equally dangerous, groups of drunk people who, knowing their numerical superiority and greater vulnerability, decide they need not wait for a clear road when crossing, since the car will swerve for them anyway.
That is the deep problem with an ethics that counts only outcomes in the moment: it hands a systematic advantage to whoever is willing to create the danger.
Go deeper:
Wikipedia: Jaywalking — how road rules already allocate fault between pedestrian and driver, and how differently across countries.