EUVIMEDEUVIMED entdecken
Open Access

Mentalizing Without a Mind: Psychotherapeutic Potential of Generative AI

Karen Yirmiya iD, Peter Fonagy iD

Journal of Medical Internet Research · 10.10.2025 · 27 · e79156

DOI: 10.2196/79156 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Medical Internet Research, online erschienen am 10.10.2025. Verfasst von Karen Yirmiya und 1 weiteren Autor:innen. Beteiligte Einrichtung: Clinical, Educational and Health Psychology, University College London, London, United Kingdom. Hauptthemen laut MEDLINE: Artificial Intelligence, Mentalization, Psychotherapy, Theory of Mind. Die Arbeit wurde bisher 40-mal zitiert (Crossref) und nennt 85 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper explores the integration of generative artificial intelligence (AI) into psychotherapeutic practice through the lens of mentalization theory, with a particular focus on epistemic trust—a critical relational mechanism that facilitates psychological change. We critically examine AI’s capability to replicate core therapeutic components, such as empathy, embodied mentalizing, biobehavioral synchrony, and reciprocal mentalizing. Although current AI systems, especially large language models, demonstrate significant potential in simulating emotional responsiveness, cognitive empathy, and therapeutic dialogue, fundamental limitations persist. AI’s inherent lack of genuine emotional presence, reciprocal intentionality, and affective commitment constrains its ability to foster authentic epistemic trust and meaningful therapeutic relationships. Additionally, we outline significant risks, notably for individuals with complex trauma or relational vulnerabilities, highlighting concerns regarding pseudo-empathy, mistaking phenomenal experience for objective reality (psychic equivalence), fruitless ungrounded pursuit of social understanding (hypermentalization), and epistemic exploitation of individuals in whom artificial understanding by AI triggers excessive credulity. Nonetheless, we propose ethically informed pathways for integrating AI to enhance clinical practice, therapist training, and client care, particularly in augmenting human capacities within group and adjunctive therapy contexts. Paradoxically, AI could support psychotherapists in improving their capacity to mentalize, improve their understanding of their clients, and provide such understanding within the moral constraints that normally govern their work. This paper calls for careful ethical regulation similar to that limiting genetic manipulation, interdisciplinary research, and clinician involvement in shaping future AI-based psychotherapeutic models, emphasizing that AI’s role should complement rather than replace the irreplaceable relational core of psychotherapy.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Artificial IntelligenceMentalizationPsychotherapyTheory of MindHumansEmpathyTrust

Stichwörter der Autor:innen

epistemic trustgenerative artificial intelligencementalizationpsychotherapyreflective functioning

Autor:innen und Institutionen

  • Clinical, Educational and Health Psychology, University College London, London, United Kingdom.
  • Department of Psychology, Ben-Gurion University of the Negev, Be'er Sheva, Israel.
  • Anna Freud Centre, London, United Kingdom.

ORCID-Profile: Karen Yirmiya, Peter Fonagy

Zitieren

APA

Yirmiya, K., & Fonagy, P. (2025). Mentalizing Without a Mind: Psychotherapeutic Potential of Generative AI. Journal of Medical Internet Research, 27, e79156. https://doi.org/10.2196/79156

Vancouver

Yirmiya K, Fonagy P. Mentalizing Without a Mind: Psychotherapeutic Potential of Generative AI. Journal of Medical Internet Research. 2025;27:e79156. doi:10.2196/79156

BibTeX
@article{Yirmiya2025,
  author = {Yirmiya, Karen and Fonagy, Peter},
  title = {Mentalizing Without a Mind: Psychotherapeutic Potential of Generative AI},
  journal = {Journal of Medical Internet Research},
  year = {2025},
  volume = {27},
  pages = {e79156},
  doi = {10.2196/79156}
}
Kostenlos starten →