Research
A research programme on how moral verdicts are formed, by people and by AI systems, and why they may resist revision, with methods for measuring the judgements of AI systems reliably.
Published · Moral Violations Are Made, Not Found: A Theory of Motivated Violation Construction, New Ideas in Psychology, 84, 101313 (2027).
One judgement, three proposed stages
Where in a judgement is the verdict decided?
Each paper places the verdict at a different point in the same judgement. Trait Judgmentalism version 4 treats Motivated Violation Construction as the upstream stage, as a testable commitment, and itself as the downstream revision stage. MVC’s argument does not depend on the wider programme. The ordering is a proposal, not an established architecture.
-
Classification
Is this act a moral violation?
Motivated Violation Construction
Published Journal article Vol. 84, 2027
Proposes that, for ambiguous behaviour, motivation shapes whether an act is classified as a violation at all, before its severity is weighed. Published in New Ideas in Psychology, 84, 101313 (2027).
-
Attribution
Does the act come to define the person?
Moralised Essentialism
Preprint
Proposes that once an act comes to define the person, later excuses can still reduce blame for the act but no longer reach the judgement of the person. Unlike Trait Judgmentalism, the excuse is still weighed, but only for the act, not the person.
-
Revision
Does understood context change the verdict?
Trait Judgmentalism
Preprint Starting point of the programme
Proposes that context is understood and accepted but given negligible weight. A preregistered reanalysis of public blame-updating data supplied no evidence for or against Trait Judgmentalism and set the design requirement for later tests.
Key terms
Judgmentalism: understanding relevant, credible mitigating context and leaving the verdict on the person unchanged (Trait Judgmentalism, December 2025).
- Trait Judgmentalism
- The programme’s first paper proposes judgmentalism as a stable individual difference: an evaluator comprehends relevant, credible mitigation and does not revise the verdict on the person.
- Contextual Moderation Failure
- Its proposed mechanism: a failure of weighting after comprehension. The context is understood and accepted but given negligible weight.
- Evaluative rigidity
- The programme’s term for a general disposition toward verdicts about people that resist revision. The current version of Trait Judgmentalism proposes it as an antecedent of the trait, grounded in moralised essentialism, the reading of conduct as fixed character. It may also operate downstream of MVC, on behaviour already classified as a violation.
How the programme developed
Trait Judgmentalism came first. Motivated Violation Construction and Moralised Essentialism grew out of it, and the language-model work began by applying MVC’s logic to models.
- Trait Judgmentalism, the first paper.
- Motivated Violation Construction and Moralised Essentialism released as preprints. First language-model study. Concept note, Evaluation Without Revision.
- Concept note, Moral Repair Is Not Reversal.
- Trait Judgmentalism version 4, with its preregistered reanalysis.
- The Screen Decides the Verdict accepted as a peer-reviewed poster at the TAE workshop at NeurIPS 2026. The Cell, Not the Call released as a preprint.
- Motivated Violation Construction accepted and published in New Ideas in Psychology, the programme’s first peer-reviewed journal article.
Language models as evaluators
Language models are now used to evaluate human behaviour. These studies ask whether MVC’s classification-before-evaluation logic describes their outputs, and whether their judgements can be measured reliably. They concern language models and do not test MVC in people. They are co-authored with José Daniel Muñoz Arciniegas.
Accepted NeurIPS 2026 workshop
The Screen Decides the Verdict
Revised preprint
Same Prompt, Not the Same Instrument
Preprint Under review