Trolley Problem

In Foot’s and later Thomson’s variants, a person can divert a trolley so that one person dies instead of five. Small changes in causal structure, intention and means may substantially alter moral intuitions.

There is extensive descriptive evidence from moral psychology but no uniquely accepted normative solution. Moral Machine provides vast amounts of preference data, with limitations inherent in the paradigm.

Moral judgements may depend on agency, intention, causal structure and framing, not just outcome counts. An is→ought inference remains invalid without additional normative premises.

AI transfer to investigate: AI can collect, simulate and operationalise preferences in policy. This increases the risk that descriptive consensus is granted authority it does not automatically deserve.

Proposed test, not yet conducted: Present the same majority-preference evidence with or without an explicit separation from normative argument. Measure perceived moral justification and policy acceptance.

Use trolley cases to maintain the boundary between empirical preference evidence and normative decision authority.

Distinguish causing harm, allowing harm, intending an outcome and using someone as a means. Observed preferences do not by themselves justify a policy.

Foot (1967), The Problem of Abortion and the Doctrine of the Double Effect, later reprinted text — typescript pp.2–4: driver/judge, doing/allowing and negative/positive duties

Awad et al. (2018), The Moral Machine experiment — Nature 563:59–64; publisher abstract and data availability

An unofficial transcript of Foot was used (pp. 2–4); the exact linked file has not been assessed. The publisher abstract was used for Moral Machine (2018). Preference data do not by themselves provide normative justification.