Review: The opportunities and risks of AI in mental health research and care

In a Review, Nils Opel and Michael Breakspear discuss how artificial intelligence (AI) can be responsibly and effectively integrated into mental health care, given the unique clinical, ethical, and societal challenges of the field. “It is tempting to be blinded or bewildered by the technological appeal of AI and its superhuman accomplishments,” write the authors. “We suggest that the opportunities and contradictions of AI can be reconciled by avoiding this technology-centric allure and instead adopting a human-centered approach…” AI is poised to reshape mental health care. Recent advances in machine learning, language analysis, digital sensors, and large language models have raised hopes that AI could improve diagnoses, monitoring, and treatment of mental health disorders, as well as expand access to care, particularly for underserved populations. However, according to Opel and Breakspear, the use of AI in mental health care presents unique challenges. In this Review, Opel and Breakspear discuss these challenges and the ways in which AI systems and tools could be successfully and responsibly be deployed to improve and perhaps personalize mental health care across the patient experience.

The widespread adoption of AI in mental health care has been slow because many mental health diagnoses are based largely on subjective symptoms and observed behavior, rather than clear biological tests, and often do not reliably predict outcomes. What’s more, there are concerns related to biased training data, privacy, and the ability of AI systems to deliver safe, empathetic care across diverse populations. Such fears are reinforced by several high-profile incidents of conversational AI goading sensitive users to engage in self-harm or reckless behavior. Yet on the other hand, AI could help address challenges in mental health through new approaches to the analysis of large, complex data, such as speech patterns, facial expressions, wearable-device signals, and brain or molecular measurements. This could open the door to more personalized care and potentially new ways of defining or identifying mental illnesses. Moreover, Opel and Breakspear note that the growing role of AI raises important questions about how it should be used in clinicians’ daily work, especially in sensitive areas such as patient privacy, risk assessment, and treatment decisions. Although AI holds transformative promise, the authors argue that given the deeply personal nature of mental health and the stigma that often surrounds it, patient-centered AI systems must be designed to protect privacy and reduce inequalities, with coordinated oversight across science, medicine, ethics, and patent empowerment.