Challenges and opportunities in using interpretable AI to develop relationship interventions
Year: 2025
Type: article
Abstract: Objective Although still in its infancy, research shows promise that artificial intelligence (AI) models can be integrated into relationship interventions, and the potential benefits are substantial. This article articulates the challenges and opportunities for developing relationship interventions that integrate AI. Background After defining AI and differentiating machine learning from deep learning, we review the key concepts and strategies related to AI, specifically natural language processing, interpretability, and human‐in‐the‐loop strategies, as key approaches needed to develop AI models that can be integrated into interventions. Method We explore how AI is currently integrated into family life and explore the literature that has served as the foundation for further integrating AI into interventions. The use of AI in therapy contexts is examined, and we identify key ethical challenges that need to be addressed as this technology develops. Results We examine the key challenges and opportunities for using AI, specifically focusing on four key areas: diagnosis of relationship problems, providing autonomous treatment, predicting successful treatment outcomes (prognosis), and using biomarkers to monitor client reactions. Opportunities explored include the development of data‐efficient AI training methods, creating interpretable AI models focused on relationships, the integration of clinical expertise during model development, and combining biomarker data with other modalities. Conclusion Despite the obstacles, integrating AI into interventions has the potential to provide families with personalized support to strengthen bonds and overcome relational challenges. Implications This emerging intersection of AI and family science can pioneer innovative solutions for diverse relationship needs.
Source: Family Relations
DOI: 10.1111/fare.13172
Authors: Daniel J. Puhlman, Chaofan Chen
Institutions: University of Maine, University of Maine System
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4409181520