Toward emotional mediation: generative AI in art therapy for psychosocial health support

Abstract

The American Art Therapy Association defines art therapy as "a regulated mental health profession that uses active art-making, the creative process, applied psychological theory, and human experience within a psychotherapeutic relationship" (1). Similarly, the British Association of Art Therapists defines it as "a form of psychotherapy that uses art media as its primary mode of communication" (2). Since the 1960s-70s, art therapy has been recognized as promoting health and well-being (3,4). As a psychological intervention, it emphasizes emotional externalization via visual expression and fosters patient-therapist emotional bonds, providing a key nonverbal communication method (5,6). Today, art therapy encompasses various media such as painting, sculpture, photography and etc.The development of artificial intelligence (AI) for image generation, language interaction, and pattern recognition has introduced new possibilities for art therapy. Some scholars argue that AI systems are creative, with the capacity to generate novelty (7). Moreover, AI can serve as a source of inspiration and a practical tool for those with little artistic experience (8). Interactive AI systems offer real-time feedback, stimulating users' imagination and modes of expression (9). For example, collaborative tools like Coco Sketch, GauGAN, and SmartPaint, along with AI-enabled apps like Mind Palette, combine generative AI with therapeutic goals to support mental health (10)(11)(12)(13).Current perspectives in the literature suggest that AI use in art therapy is moving from media-based support toward more structured forms of intervention. An integrative review of the role of AI in art therapy conceptualizes AI as an expressive tool focused on creative collaboration, emotional externalization, and symbolic generation (16). For example, a small-scale study employed autonomous painting robots to co-create with users under a Diagnostic Drawing Series (DDS) framework, proposing the potential for an affective human-machine interactions (17). There are also co-creation platforms aimed at reducing technical barriers, wherein therapists have contributed to collaborative strategy design (10). Additionally, recent studies have begun to explore how AI might function in therapeutic contexts, with roles ranging from supportive tools to potential co-creators or semi-autonomous agents (19,16). In one pilot study, AI-generated imagery was situated within the expressive therapy continuum (also known as ETC), demonstrating the possible use of AI to activate creativity across cognitive, emotional, and sociocultural dimensions (19). Following this trend, a further study showed that AI-enhanced drawing games improved engagement and cognitive stimulation in older adults with mild cognitive impairment. However, these activities were not conducted within a formal art therapy setting and did not appear to incorporate a therapeutic relationship facilitated by a trained art therapist, which is often considered a key element in art therapy (20).Second, several emerging studies have explored the potential of AI as an assistive system in tasks such as assessment, emotional recognition, and therapeutic feedback. For instance, AI-assisted systems can support large-scale drawing evaluation and improve interpretability and diagnostic accuracy (21,22). Meanwhile, tools such as TherAIssist allow therapists to use AI to collaborate on procedures related to feedback, tracking, and homework (23). In education, AI has helped with the analysis of emotional expressions in drawings by students with autism (24). Another study shows the capacity of AI to identify emotional cues and guide therapeutic dialogues, aiding patients with posttraumatic stress disorder in emotional release (25). Deep neural networks were used to integrate statistical and qualitative drawing evaluations (22), and reviews of AI interventions (i.e., from drawing agents to chatbots) have highlighted both their innovation and emergin