Why We Disagree on Ethics: The Science Behind Moral Disagreement

Two thoughtful, well-intentioned people look at the same trolley problem and reach opposite conclusions. One says pull the lever — save five lives. The other says keep your hands clean — you cannot use someone as a means to an end. Both have reasons. Both feel certain. Neither is obviously confused or cruel. So what is actually happening?

Research in moral psychology over the past three decades has assembled a partial answer — and it is more unsettling than "one person has better values." Moral disagreement appears to be built into the architecture of human reasoning, shaped by evolution, culture, and the competing cognitive systems we use to evaluate right and wrong.

Moral foundations are not all the same

Psychologist Jonathan Haidt and colleagues proposed Moral Foundations Theory as one account of why moral disagreement runs so deep. The theory identifies several distinct moral intuitive concerns that all human societies draw on — including care, fairness, loyalty, authority, and purity — but in different proportions. Political liberals tend to weigh care and fairness heavily. Conservatives tend to weigh loyalty, authority, and purity alongside care and fairness.

This does not mean one group is right and the other is wrong. It means that the same dilemma activates different moral modules in different people. When someone says "I cannot betray my group, even to prevent greater harm," they are not being irrational — they are running on a different moral operating system, one that assigns heavy weight to the loyalty foundation. When someone else says "the numbers are the numbers — five lives outweigh one," they are running a care-and-fairness-dominated calculation.

Two cognitive systems, one dilemma

A separate strand of research, associated primarily with psychologist Joshua Greene at Harvard, focuses not on the content of moral foundations but on the process of moral reasoning. Greene's influential dual-process account proposes that moral judgments are produced by two competing systems: a fast, automatic, emotionally-driven System 1, and a slower, deliberative, more consequentialist System 2.

Classic trolley experiments illustrate the tension. Most people are willing to pull a lever that redirects a trolley, killing one to save five. Far fewer are willing to push a large man off a bridge to achieve the same arithmetic result. The math is identical. The physical immediacy is not — and System 1, which evolved to respond strongly to direct physical harm, fires much harder in the footbridge version. Whether that emotional signal is tracking a genuine moral distinction (direct harm really is different from indirect harm) or producing a bias (numbers should be what matter) is still debated.

Cross-cultural patterns — and where they break down

A 2018 study published in Nature — the Moral Machine experiment — collected over 40 million moral decisions from people across 233 countries, asking how self-driving cars should prioritize lives in unavoidable crash scenarios. Some patterns were remarkably stable across cultures: a preference for saving more lives over fewer, a preference for sparing children over the elderly. Others varied sharply: preferences for sparing higher-status individuals, for prioritizing pedestrians over passengers, and for swerving versus staying straight differed substantially by region.

The cross-cultural variation suggests that moral foundations are not a fixed human universal. Culture shapes which concerns get weighted heavily, which scenarios activate moral alarm, and which tradeoffs feel acceptable. The same action that feels like an obvious violation in one moral community may feel like the obvious right answer in another.

The moral dumbfounding problem

Haidt's research on "moral dumbfounding" adds another layer. When people are presented with scenarios that trigger strong moral intuitions but where every harm-based justification has been pre-empted by the story's setup, many continue to maintain that something is wrong — even when they cannot say what the harm is. They know it feels wrong before they know why.

This suggests that moral reasoning is often post-hoc rationalization rather than the cause of moral verdicts. We reach conclusions first, via intuition, and then construct arguments to justify them. If that is accurate, then moral disagreement may persist even after extensive rational discussion — because the argument was never really what was driving the verdict.

What SplitVote data shows

On SplitVote, the same dilemmas that have split laboratory participants split real-world voters in remarkably similar proportions. Trolley-style scenarios that pit aggregate welfare against individual rights tend to produce near-even splits. Scenarios that activate strong emotional alarm — physical immediacy, identifiable victims, violations of bodily autonomy — tend to produce stronger consensus, often against the arithmetically "efficient" outcome.

SplitVote is not a scientific instrument — the sample is self-selected, the framing is fixed, and individual votes carry no experimental controls. But at scale, the patterns are suggestive: moral disagreement is not random noise. It clusters around the fault lines that moral psychology has identified — System 1 vs. System 2 activation, loyalty vs. harm-prevention foundations, cultural variation in authority and purity concerns.

Is moral disagreement resolvable?

Moral psychology does not settle the philosophical question of whether there are objective moral truths that disagreement fails to reach. But it does reframe the question. If two people disagree on a moral dilemma, it may not be because one has thought harder than the other. It may be because they are activating different emotional systems, weighing different foundations, or applying frameworks shaped by different cultural histories.

That is not a counsel of relativism. Some moral positions are still better reasoned than others, more consistent, more attentive to evidence. But it is a reason for epistemic humility: the confidence that feels like clarity may be a fast-responding System 1 alarm, not a carefully derived conclusion. Moral progress — when it happens — usually requires slowing down enough to examine the intuitions rather than simply trusting them.

References to Moral Foundations Theory, dual-process moral cognition, and the Moral Machine study are for contextual and educational purposes. SplitVote is an entertainment and reflection platform, not a scientific study. Results represent community votes — not peer-reviewed data, and not conclusions about individual moral character. Theories described here are contested within moral philosophy and cognitive science.