Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Responsible Evaluation of AI for Mental Health

Hiba Arnaout, Anmol Goel, H. Andrew Schwartz, Steffen T. Eberhardt, Dana Atzil-Slonim, Gavin Doherty, Brian Schwartz, Wolfgang Lutz, Tim Althoff, Munmun De Choudhury, Hamidreza Jamalabadi, Raj Sanjay Shah, Flor Miriam Plaza-del-Arco, Dirk Hovy, Maria Liakata, Iryna Gurevych

TL;DR

This paper argues that evaluating AI tools for mental health requires a clinically grounded, interdisciplinary framework that accounts for clinical validity, social context, equity, and user experience. It introduces a taxonomy aligning assessment, intervention, and information synthesis with validity, reliability, implementation, and maintenance, illustrated through five case studies spanning LLM-based assessment, intervention, and information synthesis tools. The authors demonstrate how lifecycle evaluation and cross-disciplinary collaboration can improve safety, effectiveness, and real-world utility, and they outline a maturity model and infrastructure needs to support future work. The work aims to shift AI mental health evaluation from predominantly technical benchmarks to holistic, clinically anchored, and ethically responsible practices with potential for safer deployment in diverse settings.

Abstract

Although artificial intelligence (AI) shows growing promise for mental health care, current approaches to evaluating AI tools in this domain remain fragmented and poorly aligned with clinical practice, social context, and first-hand user experience. This paper argues for a rethinking of responsible evaluation -- what is measured, by whom, and for what purpose -- by introducing an interdisciplinary framework that integrates clinical soundness, social context, and equity, providing a structured basis for evaluation. Through an analysis of 135 recent *CL publications, we identify recurring limitations, including over-reliance on generic metrics that do not capture clinical validity, therapeutic appropriateness, or user experience, limited participation from mental health professionals, and insufficient attention to safety and equity. To address these gaps, we propose a taxonomy of AI mental health support types -- assessment-, intervention-, and information synthesis-oriented -- each with distinct risks and evaluative requirements, and illustrate its use through case studies.

Responsible Evaluation of AI for Mental Health

TL;DR

This paper argues that evaluating AI tools for mental health requires a clinically grounded, interdisciplinary framework that accounts for clinical validity, social context, equity, and user experience. It introduces a taxonomy aligning assessment, intervention, and information synthesis with validity, reliability, implementation, and maintenance, illustrated through five case studies spanning LLM-based assessment, intervention, and information synthesis tools. The authors demonstrate how lifecycle evaluation and cross-disciplinary collaboration can improve safety, effectiveness, and real-world utility, and they outline a maturity model and infrastructure needs to support future work. The work aims to shift AI mental health evaluation from predominantly technical benchmarks to holistic, clinically anchored, and ethically responsible practices with potential for safer deployment in diverse settings.

Abstract

Although artificial intelligence (AI) shows growing promise for mental health care, current approaches to evaluating AI tools in this domain remain fragmented and poorly aligned with clinical practice, social context, and first-hand user experience. This paper argues for a rethinking of responsible evaluation -- what is measured, by whom, and for what purpose -- by introducing an interdisciplinary framework that integrates clinical soundness, social context, and equity, providing a structured basis for evaluation. Through an analysis of 135 recent *CL publications, we identify recurring limitations, including over-reliance on generic metrics that do not capture clinical validity, therapeutic appropriateness, or user experience, limited participation from mental health professionals, and insufficient attention to safety and equity. To address these gaps, we propose a taxonomy of AI mental health support types -- assessment-, intervention-, and information synthesis-oriented -- each with distinct risks and evaluative requirements, and illustrate its use through case studies.
Paper Structure (12 sections, 16 tables)

This paper contains 12 sections, 16 tables.

Table of Contents

  1. Introduction
  2. Observed Practices
  3. Proposed Taxonomy
  4. Case Studies
  5. Study I (Assessment): LLM rating scales for psychometric assessment of patient engagement
  6. Study II (Assessment): Natural language response formats for assessing depression and worry
  7. Study III (Intervention): Evaluating the capabilities of LLMs vs. human therapists to generate personalized interventions
  8. Study IV (Intervention): A clinically-grounded framework for evaluating LM-assisted cognitive restructuring
  9. Study V (Information synthesis): Hierarchical LLM-VAE tool for clinically meaningful timeline summarization
  10. Moving Forward
  11. Conclusion
  12. Details about the surveyed papers