PROUSTMEContribute
    Ideas & Research

    Interpretive Fidelity

    Measuring Whether Artificial Intelligence Understands Human Motive

    By

    Author and Founder & CEO of ProustMe

    Proposed evaluation framework

    Definition

    Interpretive Fidelity refers to the degree to which an artificial intelligence system’s interpretation of a human narrative accurately reflects the intentions, motivations, circumstances, and meanings supported by the available evidence.

    It is a proposed dimension of model evaluation, not a claim that a universally accepted benchmark already exists.

    Central Question

    How do we distinguish a plausible explanation of human behavior from a faithful interpretation of the individual’s experience?

    Statistical Plausibility vs. Interpretive Fidelity

    A model may generate an explanation that sounds reasonable because it resembles patterns encountered during training.

    But a plausible explanation is not necessarily faithful to the actual narrative.

    Interpretive Fidelity investigates whether models can distinguish between explanations that are statistically likely and explanations grounded in a particular individual’s testimony.

    Proposed Evaluation Dimensions

    Motive Recognition: Does the model identify motivations supported by the narrative?

    Contextual Accuracy: Does it account for the circumstances surrounding the decision?

    Contradiction Recognition: Can it recognize competing motives and internal tensions?

    Uncertainty Calibration: Does it acknowledge when the evidence is insufficient?

    Narrative Grounding: Can it connect its interpretation to specific evidence in the account?

    Human Validation: Can the interpretation be evaluated against contributor feedback and appropriately designed independent assessments?

    Proposed Research Approach

    Develop consent-based narrative evaluation datasets.

    Compare model performance with and without relevant first-person narrative context.

    Use clearly defined evaluation criteria, appropriate baselines, and human review.

    Report both successful and unsuccessful results.

    This page presents a proposed research approach, not completed studies, benchmark scores, or demonstrated model improvements.

    Closing Proposition

    “An explanation can be plausible without being faithful to the person whose experience it claims to explain.”

    About the Author

    Mark Bolding is an author and entrepreneur whose work examines human agency, economic sovereignty, narrative identity, and the systems that shape human possibility. He is the Founder & CEO of ProustMe, a Narrative Intelligence company developing infrastructure for originating, structuring, and permissioning first-person human experience for artificial intelligence.

    Mark Bolding

    Related Ideas