a recent study by brown university compared the responses of mainstream ai chatbots (gpt, claude, llama) with those of licensed clinical psychologists in psychological support scenarios. the study invited seven practicing counselors qualified in cognitive-behavioral therapy to engage in simulated conversations with each ai system. subsequently, three senior clinical psychologists conducted a double-blind evaluation of all conversation transcripts, focusing on whether the responses adhered to ethical guidelines and professional standards.
the analysis identified 15 potential risks, with key issues including template-based replies that fail to account for individual contexts, judgment errors in crisis intervention, and so‑called “pseudo‑empathy”—where ai mechanically repeats emotionally charged phrases like “i understand you” without being able to recognize, respond to, or regulate genuine emotional states.
the research team emphasized that the real concern lies not in the imperfections of the technology itself, but in the lack of accountability: human therapists must be licensed, undergo ongoing supervision, and operate under industry regulation, whereas current ai developers face neither licensing requirements nor clinical liability for their outputs. consequently, the study strongly calls for the establishment of a dedicated regulatory framework for ai applications in mental health, covering admission criteria, usage boundaries, risk‑alert systems, and mechanisms for assigning responsibility.
