Considering Therapy (and Mentoring): When Does Human Empathy Matter?
Rubin, M., Arnon, H., Huppert, J. D., & Perry, A. (2024). Considering AI-driven therapy: When does human empathy matter? JMIR Preprints. https://doi.org/10.2196/preprints.56529
As AI-driven mental health tools gain traction, a pressing clinical question has emerged: under what conditions can artificial intelligence reasonably substitute for a human therapist, and when does the irreplaceable quality of human empathy become the decisive factor in treatment? Rubin and colleagues (2024) take up this question directly, arguing that empathy, specifically its emotional and motivational dimensions, sits at the heart of the answer. The authors propose that AI may competently handle cognitively oriented therapeutic tasks but falls fundamentally short when genuine affective connection is what the patient needs most.
The authors conducted a narrative review and analytical synthesis of existing literature on empathy, psychotherapy outcomes, internet-based interventions (IBI), and patient perceptions of AI chatbots. They organized their argument across two primary frameworks: the perspective of therapeutic approach (ranging from skill-based cognitive-behavioral methods to relationship-centered psychodynamic approaches), and the perspective of patient needs (conceptualized along two orthogonal axes — desire for practical tools versus desire for human connection).
They found that AI can plausibly replicate cognitive empathy but cannot authentically produce emotional empathy (affective sharing) or motivational empathy (genuine concern for another’s welfare). Because human empathic expression encompasses time, emotional labor, and selective attention, it signals care in a way that AI responses structurally cannot. Research cited in this piece found that therapist empathy correlates substantially with positive treatment outcomes, and that while patients can form a working relationship with digital programs, that relationship differs meaningfully from the human therapeutic alliance, particularly in predicting treatment adherence and dropout.
The authors argue that AI is best positioned for psychoeducation, skill delivery, and symptom monitoring, functions where cognitive understanding suffices. When patients primarily seek human connection, validation, or compassionate engagement, AI’s limitations become clinically consequential. The paper also flags the risk of overdependency on AI tools and notes that awareness of AI involvement tends to reduce perceived authenticity and trust.
Implications for Mentoring Programs
Mentoring, like therapy, operates on a dual axis of practical guidance and relational support. AI tools may efficiently deliver information, structure, and psychoeducation resources, but the sense that a mentor genuinely invests in a mentee’s success cannot be replicated by a system for which every interaction is cost-free and unlimited. Programs that integrate AI tools should do so in ways that preserve, rather than displace, the human relational core that research consistently links to mentee persistence and well-being.
Read the full article here
