Saved in:
| Hovedforfatter: | |
|---|---|
| Format: | Recurso digital |
| Sprog: | engelsk |
| Udgivet: |
Zenodo
2025
|
| Fag: | |
| Online adgang: | https://doi.org/10.5281/zenodo.18076873 |
| Tags: |
Tilføj Tag
Ingen Tags, Vær først til at tagge denne postø!
|
Indholdsfortegnelse:
- <p><span lang="EN-US">Predictive success in computational psychiatry is frequently treated as if it licensed explanatory or interventional authority, yet that transition is rarely justified with an explicit validity argument. This paper introduces the notion of epistemic role drift, the unacknowledged slide from prediction to explanation, or to intervention guidance, and treats it as a structural, not merely rhetorical, failure mode in model use. Drawing on philosophy of scientific modelling and contemporary validity theory, the analysis distinguishes three model roles, predictive, explanatory, and interventional, and specifies the evidential burdens that must be met when a model is repurposed across roles. Formal criteria are introduced to separate accuracy from interpretability, and interpretability from manipulability, with schematic representations used to make inferential steps and their vulnerabilities transparent. The resulting framework does not deny the utility of high performing models, rather, it constrains what can be claimed on their basis, especially when outputs migrate into high stakes decision settings. These constraints are applied to educationally adjacent contexts where developmental heterogeneity and institutional incentives can amplify role drift, sharpening the core question of when, and under what conditions, prediction can legitimately be treated as explanation.<br><br></span></p> <p><strong><span lang="EN-US">Keywords:</span></strong><span lang="EN-US"> computational psychiatry, epistemic role drift, model validity, prediction, explanation, educational decision making</span></p>