What does "injury-minimising" programming mean, and which ethical theory does it resemble?
Programming that, once an accident with personal injury is unavoidable, chooses the outcome with the fewest and least severe injuries — which is an optimisation of the utility sum, and so at first glance utilitarian.
Three points of precision, each of which matters for the argument that follows:
- The scope is narrow. It governs behaviour only in situations where an accident endangering people has already become unavoidable. It is not a general driving policy; every other manoeuvre is still about avoiding accidents altogether.
- It is necessarily an optimisation of the utility sum. Minimising expected injuries means adding up expected harms across people and choosing the smaller total. There is no way to do it that does not aggregate.
- It therefore has obviously utilitarian features — it belongs to consequentialist ethics (Folgenethik), which evaluates an action by its consequences.
That third point sets up the paper's central question: is a programming aimed at injury minimisation compatible in principle with a deontological ethics (Pflichtenethik), or must it be rejected outright as an impermissible form of instrumentalisation?
The reason that question is not rhetorical is that deontology's core objection to utilitarian reasoning is exactly this: aggregating harms treats the person who ends up on the losing side of the sum as a resource for improving other people's outcomes. If that objection applies here, then no amount of lives saved justifies the design. The rest of the analysis is an argument that it does not straightforwardly apply — for a reason that has to do with when the decision is made.
Go deeper:
Stanford Encyclopedia: Ethics of AI and Robotics — section 2.6.2 covers self-driving cars, including the argument that trolley cases mislead here.- Wikipedia: Consequentialism — the family this programming belongs to, and what it commits you to once you aggregate harms.