![[2602.16053] Multi-Objective Alignment of Language Models for Personalized Psychotherapy](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16053] Multi-Objective Alignment of Language Models for Personalized Psychotherapy
Summary
This article discusses a multi-objective alignment framework for language models aimed at enhancing personalized psychotherapy, balancing patient preferences with clinical safety.
Why It Matters
Mental health disorders affect over a billion people globally, yet access to effective care is limited. This research addresses the need for AI-driven therapeutic solutions that prioritize both empathy and safety, potentially transforming mental health treatment accessibility and effectiveness.
Key Takeaways
- AI can enhance psychotherapy by aligning language models with patient preferences.
- The study introduces a multi-objective alignment framework that balances empathy and safety.
- Direct preference optimization outperforms traditional single-objective methods in therapeutic contexts.
- Clinician evaluations indicate a strong preference for the new multi-objective approach.
- This research highlights the potential of AI in addressing mental health care gaps.
Computer Science > Machine Learning arXiv:2602.16053 (cs) [Submitted on 17 Feb 2026] Title:Multi-Objective Alignment of Language Models for Personalized Psychotherapy Authors:Mehrab Beikzadeh, Yasaman Asadollah Salmanpour, Ashima Suvarna, Sriram Sankararaman, Matteo Malgaroli, Majid Sarrafzadeh, Saadia Gabriel View a PDF of the paper titled Multi-Objective Alignment of Language Models for Personalized Psychotherapy, by Mehrab Beikzadeh and 5 other authors View PDF HTML (experimental) Abstract:Mental health disorders affect over 1 billion people worldwide, yet access to care remains limited by workforce shortages and cost constraints. While AI systems show therapeutic promise, current alignment approaches optimize objectives independently, failing to balance patient preferences with clinical safety. We survey 335 individuals with lived mental health experience to collect preference rankings across therapeutic dimensions, then develop a multi-objective alignment framework using direct preference optimization. We train reward models for six criteria -- empathy, safety, active listening, self-motivated change, trust/rapport, and patient autonomy -- and systematically compare multi-objective approaches against single-objective optimization, supervised fine-tuning, and parameter merging. Multi-objective DPO (MODPO) achieves superior balance (77.6% empathy, 62.6% safety) compared to single-objective optimization (93.6% empathy, 47.8% safety), and therapeutic criteria outperform g...
