Beginning in the late 2000s, Eric Schwitzgebel and his colleagues tested a quiet assumption of academic ethics: that people who study morality professionally reason about it better than others. The results were unflattering. Contemporary ethics books were more likely to go missing from university libraries than comparable philosophy titles. In later work with Fiery Cushman, professional philosophers proved just as susceptible as non-philosophers to order effects in trolley-style dilemmas, even on scenarios they had spent careers analysing.

No comparable finding would embarrass physics. Nobody asks whether physicists fall more gracefully than the rest of us, because physical expertise is anchored to instruments, predictions and replication rather than to the expert's own sense of things. Moral inquiry has no such anchor. Its primary data are intuitions, and intuitions are outputs of the same cognitive machinery that experimental work keeps showing to be sensitive to framing, sequence and affect.

This asymmetry is easy to state and hard to interpret. It might show that there is no moral knowledge at all. Or it might show that moral knowledge has a different epistemic architecture from empirical science. Three points of divergence matter here: the unobservability of moral facts, the shape moral expertise would have to take, and whether progress requires convergence. The picture that emerges supports neither debunking nor complacency.

Gilbert Harman's classic formulation in The Nature of Morality (1977) contrasts two observations. A physicist sees a vapour trail in a cloud chamber and concludes that a proton passed through. The best explanation of that observation cites the proton. You round a corner, see children setting a cat on fire, and immediately judge the act wrong. Yet the best explanation of your judgment seems to cite your upbringing and psychology, not an independent fact of wrongness.

This is the problem of explanatory dispensability. Scientific posits earn their place by doing causal work: remove the proton and the vapour trail goes unexplained. Remove the moral fact and nothing observable changes. J. L. Mackie pressed a related worry, his argument from queerness. Moral facts would have to be objective features of the world that are also intrinsically action-guiding. No other entity in our ontology combines those properties.

Naturalist realists have pushed back. Nicholas Sturgeon argued that Hitler's moral depravity genuinely explains his conduct, and that moral properties may be identical to complex natural properties, much as water is identical to H₂O. The difficulty is that such explanations look parasitic. Whatever explanatory work depravity does seems fully carried by non-moral descriptions of Hitler's beliefs, desires and dispositions. The moral vocabulary adds a verdict, not a mechanism.

Moral psychology sharpens the challenge. Joshua Greene's neuroimaging work associates characteristically deontological verdicts with affective responses to personal, up-close harm, responses with plausible evolutionary origins. Sharon Street's Darwinian dilemma generalises the point. Selection pressures shaped our evaluative dispositions to promote reproductive success, not to track mind-independent moral truth. So realists must either posit a lucky coincidence between fitness and truth or admit that our intuitions are poor instruments for detecting moral facts.

The strongest realist reply invokes companions in guilt. Mathematical facts are not observed either, yet we claim mathematical knowledge. The analogy is imperfect. Mathematics is indispensable to our best physical theories, while moral facts are not. David Enoch has argued instead for deliberative indispensability: we cannot decide what to do without presupposing that some answers are better than others. If that argument works, moral facts are justified by the practical standpoint, not by observation.

Takeaway

Moral facts, if they exist, would earn their place by being indispensable to deliberation rather than to explanation. That difference changes what could ever count as evidence for them.

In 1972 Peter Singer argued that moral philosophers could be genuine moral experts. Their training in argument, their familiarity with relevant facts, and the time they can devote to reflection should give them an edge. The empirical record complicates this. Schwitzgebel and Joshua Rust found that ethicists behave no better than other academics on measures such as voting, replying to student emails and returning library books. Ethicists rated eating meat as worse, yet reported eating it at similar rates.

Behaviour, however, is the wrong test. Expertise concerns judgment, not virtue. Here the findings are more troubling. In studies published in 2012 and 2015, Schwitzgebel and Cushman showed that philosophers' verdicts on trolley cases and on the Asian disease framing problem shifted with presentation order. These shifts were as large as those among non-philosophers. Self-reported expertise with the specific scenarios did not reduce them, and neither did prompts encouraging careful reflection.

Daniel Kahneman and Gary Klein identified the conditions under which intuitive expertise develops: an environment with stable regularities, and prolonged practice with rapid, unambiguous feedback. Weather forecasters and chess players meet these conditions; their errors are revealed and corrected. Moral judgment rarely does. No outcome announces that one verdict on the trolley problem was mistaken. Without feedback, practice can entrench a judgment as easily as improve it.

If moral expertise exists, it probably takes a procedural form rather than an oracular one. Its components would include identifying morally relevant features, detecting inconsistency across cases, separating factual from evaluative disagreement, and recognising the conditions under which intuitions are known to be unreliable. Greene's dual-process framework suggests a version of this. The expert knows when to distrust the automatic setting and switch to manual mode. That is a skill about intuitions, not a superior supply of them.

This model fits a curious feature of moral testimony. We readily defer to a geologist on the age of a rock. Yet philosophers such as Alison Hills and Sarah McGrath note that deferring wholesale to someone else's verdict about whether an act is wrong seems defective. The deferrer would hold a correct belief without understanding why it is correct. Moral expertise may be the kind of knowledge that must be reconstructed by the learner rather than simply transmitted.

Takeaway

A moral expert may be less like a laboratory scientist with privileged data and more like a skilled editor. Their value lies in spotting inconsistency and distortion in reasoning everyone already shares.

In science, convergence functions as a rough proxy for truth. C. S. Peirce went so far as to define truth as the opinion inquiry is fated to reach. Ethics looks worse by this standard. The 2020 PhilPapers Survey found professional philosophers divided roughly into thirds among consequentialism, deontology and virtue ethics, after more than two millennia of sustained inquiry. Mackie read persistent disagreement of this kind as evidence that moral beliefs reflect ways of life rather than perceptions of objective values.

The standard realist response is that much moral disagreement is not fundamental. It stems from disputed non-moral facts, self-interest, or unequal information, and it would dissolve under idealised conditions. John Doris and Alexandra Plakias tested this claim against Richard Nisbett and Dov Cohen's research on cultures of honour. Southern US participants responded to insult and endorsed retaliatory violence in ways their Northern counterparts did not. The difference resisted explanation by any obvious factual error, which makes it a candidate for disagreement that would survive idealisation.

Yet moral progress appears real. Chattel slavery has lost all mainstream defenders, judicial torture has largely been abandoned, and moral concern has expanded to include outsiders and nonhuman animals. Allen Buchanan and Russell Powell, among others, treat this expansion as a genuine achievement. The pattern is instructive: convergence has occurred at the level of mid-level judgments, such as the wrongness of slavery, while foundational theory remains contested.

This suggests that moral progress may work largely by subtraction. Victor Kumar and Richmond Campbell argue that consistency reasoning drives much of it, by forcing people to treat like cases alike. Abolitionist arguments did not require consensus on the foundations of ethics. They required showing that the factual premises supporting slavery were false and that the principles slaveholders applied to themselves could not coherently exclude the enslaved. Error elimination can advance without any agreed positive theory.

If this is right, persistent theoretical disagreement does not refute moral knowledge, but it does limit the confidence anyone should place in foundational claims. The peer disagreement literature becomes directly relevant. When equally informed and careful thinkers diverge on whether consequences alone matter, conciliationists hold that each side should lower its credence. A defensible position follows: secure knowledge of particular wrongs, alongside genuine uncertainty about why they are wrong.

Takeaway

Moral knowledge may grow the way a sculpture does, by removing what is clearly wrong. Unlike a scientific theory, it need not be built upward from agreed foundations.

Taken together, these findings place moral epistemology closer to mathematics, law or interpretation than to laboratory science. Moral facts would be justified deliberatively rather than observationally. Expertise would be procedural, not oracular. Progress would be measured by eliminating error rather than converging on a final theory.

The practical stakes are considerable. Bioethics committees and policy advisors should be valued for their skill in exposing inconsistency and clarifying facts, not for authoritative verdicts. In machine ethics, the absence of a moral ground truth means alignment cannot simply be optimised against a labelled dataset. Systems trained on human judgments inherit the order effects, framing sensitivities and parochial intuitions that experimental philosophy has documented.

None of this makes ethics arbitrary. It means that the methods appropriate to moral inquiry must be built for a domain without instruments, which makes the discipline of reasoning carefully more important rather than less.