Evaluating Safe and Ethical Uses of AI for Psychological Wellbeing in Higher Education through Application of a Novel Framework

Poade, Donna ORCID: https://orcid.org/0009-0001-7405-6790 and Crawford, Russell (2026) Evaluating Safe and Ethical Uses of AI for Psychological Wellbeing in Higher Education through Application of a Novel Framework. In: Positive Human-AI Interaction: Fostering Psychological Well-being in the Age of Artificial Intelligence. Springer, Cham, Switzerland. ISBN not specified (In Press)

Text (Chapter) AI research chapter 230126.pdf - Accepted Version Restricted to Repository staff only until 1 January 2099. Available under License All Rights Reserved. Download (708kB) | Request a copy

Abstract / Summary

The paper introduces a “trickle filter” framework that inserts a “Stop Think Reflect” checkpoint before students act on AI generated wellbeing advice, aiming to reduce safety and ethical risks while promoting self agency. Using a mixed methods experiment with seven student scenario prompts and post task interviews, the authors show that the filter leads to earlier referrals to professional help, shorter decision times, and lower perceived stress, suggesting it can make AI mediated mental health support in higher education safer and more responsible. While the work addresses an important topic of safe AI enabled wellbeing support, there are some methodological limitations that undermine the validity of the results.
Some minor remarks: The paper does not use the suggested format. That leads to several inconsistencies in section heading and table design. There are also numerous typos, subject-verb mismatches and inconsistent hyphenation. In-text citations are inconsistent. Also, table captions are missing the tables are not referred in the text. The formatting and orthographic issues distract the reader and make the content less understandable.
The described research is “ongoing”, nevertheless, crucial information on the research design and methods is missing. A statement of the research question is missing. Methodological details, e.g. sample size, coding scheme, coding reliability, are also missing. The described limitations do not link back to the research method (and the missing research question).
The paper title states “evaluation” and “framework” positioning the paper into the design science research. Nevertheless, there is no positioning of the research design or methods into the design science research process, i.e., the methodological foundation is missing. If being properly positioned into a research framework, the authors would not have missed a literature analysis to point out the research gap their framework is supposed to fill. There is accordingly also no benchmarking of the framework to the existing AI safety and Gating-frameworks that makes the significance of the presented contribution unclear.

View Record (staff only)