# Computational Assays of Moral Inference

> **NIH NIH R21** · PRINCETON UNIVERSITY · 2020 · $154,303

## Abstract

We choose who to trust or avoid based on inferences about whether they are likely to help or harm us.
Successfully inferring the moral character of others is crucial for initiating and maintaining healthy social
relationships. When moral inference breaks down, we might place our trust in the wrong people, or prematurely
end relationships because we incorrectly infer our partner means us harm. Past work suggests disruptions to
moral inference may underlie interpersonal problems observed in psychiatric disorders, but the neurocognitive
mechanisms that underlie this relationship are poorly understood. Progress has been limited because we lack
precise, quantitative measures of moral inference that can bridge levels of analysis between self-report, behavior
and brain function. To address these limitations, the proposed research will develop parametrically detailed
computational assays of moral inference in humans. We build on work in my lab that has begun to characterize
the computational basis of moral inference in healthy people using a formal Bayesian framework for modeling
inference under uncertainty, the Hierarchical Gaussian Filter (HGF). Our HGF model of moral inference
describes how people infer another person’s tendency to be harmful or helpful based on observing their behavior,
and has several parameters that are potentially relevant to psychopathology. Here, we focus on two: the
uncertainty of beliefs about the harmfulness of others, and prior beliefs about the harmfulness of others. Our
recent work in patients with Borderline Personality Disorder suggests that people who are unable to form
appropriately certain beliefs about social partners may have problems forming healthy attachment relationships.
We further hypothesize that people with pessimistic prior beliefs about the harmfulness of others may be less
motivated to affiliate with and trust others. Crucially, our task and model can quantify individual variance in belief
uncertainty and prior beliefs simultaneously in behavior and self-reports, thus bridging these levels of analysis.
In the proposed study, over a six-month period we will collect three within-subject samples of moral inference
task performance alongside a battery of clinical and transdiagnostic questionnaires measuring social, affective
and cognitive dysregulation. In Aim 1, we will validate our model and task’s psychometric properties (such as
test-retest reliability) and characterize normative variance in model parameters that future studies can compare
against patient behavior. In Aim 2, we will characterize the longitudinal relationship between model parameters
and questionnaire measures, investigating the extent to which our model parameters describe stable individual
traits and track with clinical symptoms over time. Addressing these questions will pave the way for future work
developing more detailed computational assays of moral inference linking behavior, self-report and brain function
in patients. Our findings ...

## Key facts

- **NIH application ID:** 10774537
- **Project number:** 7R21MH124093-02
- **Recipient organization:** PRINCETON UNIVERSITY
- **Principal Investigator:** Molly J Crockett
- **Activity code:** R21 (R01, R21, SBIR, etc.)
- **Funding institute:** NIH
- **Fiscal year:** 2020
- **Award amount:** $154,303
- **Award type:** 7
- **Project period:** 2020-09-01 → 2023-08-31

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/10774537

## Citation

> US National Institutes of Health, RePORTER application 10774537, Computational Assays of Moral Inference (7R21MH124093-02). Retrieved via AI Analytics 2026-05-26 from https://api.ai-analytics.org/grant/nih/10774537. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*