A new study at Hebrew University explored the balance between AI and human therapists in mental health therapy, focusing on the role of empathy. The researchers propose a hybrid model where AI supports therapeutic processes without replacing the crucial human elements of empathy and emotional engagement. The study calls for further investigation into how AI can enhance therapy while ensuring genuine human connections are maintained.
A new study by researchers from the Psychology Department at the Hebrew University published in JMIR Ment Health has made significant strides in understanding the role of artificial intelligence (AI) in mental health therapy. Their research focuses on the delicate balance between AI-driven interactions and the irreplaceable human touch in therapeutic settings, addressing critical questions about when AI might effectively replace human therapists and when the human connection remains indispensable.
Led by Prof. Anat Perry, the team has carefully defined various aspects of empathy, comparing the empathic capabilities of humans and AI. In the current JMIR paper, the authors delve into how AI versus human capabilities align with the therapeutic needs, considering both the methodologies employed in therapeutic settings and the individual goals of patients. The study emphasizes the nuanced role of empathy in therapy, underscoring that while AI can simulate empathic interactions and sometimes even create the impression of understanding beyond human capabilities, it lacks the ability to genuinely connect on an emotional level, and crucially to genuinely care.
Prof. Perry highlights the core of their findings, stating, “While AI can provide responses that seem empathically correct, true empathy involves an emotional engagement, and signalling of genuine care, that AI simply does not have. Our study seeks to explore this boundary to better understand when AI can be beneficial in therapy and when it cannot.”
The research proposes a novel hybrid therapeutic model where AI supports but does not replace human therapists. This model suggests that AI could effectively handle tasks such as initial patient intake and routine evaluations, and even assist in certain treatment modalities. However, it crucially maintains that human therapists should be involved in situations where deep empathy and compassion are required, ensuring that the therapy remains grounded in genuine human interaction.
This study aligns with emerging trends in the field of mental health therapy, where technology is increasingly integrated into traditional therapeutic practices. Existing models, such as those combining cognitive-behavioral therapy (CBT) with AI-driven tools, have shown promise in enhancing accessibility and efficiency of therapy. For instance, AI applications can offer real-time feedback and personalized recommendations, complementing the therapist’s role and enabling more effective treatment plans.
Though much of the research remains theoretical, it raises empirical questions that are vital for the future of mental health therapy. The team calls on both industry professionals developing AI applications for mental health and academic researchers to consider these insights and the importance of maintaining human elements in therapy.
These theoretical opinion papers serve as a crucial reminder of the need to carefully evaluate the use of AI in mental health therapies, balancing technological innovations with the essential human connections that form the backbone of effective therapeutic relationships.
This is Perry’s third paper on the topic, following an influential Correspondence piece in Nature Human Behaviour last year (https://www.nature.com/articles/s41562-023-01675-w), and a Correspondence on AI, empathy and ethics published with a team of interdisciplinary scholars last month in Nature Machine Intelligence (https://www.nature.com/articles/s42256-024-00841-7).
The research paper titled “Considering the Role of Human Empathy in AI-Driven Therapy” is now available in JMIR Ment Health and can be accessed at https://doi.org/10.2196/56529.