University of Utah Researchers Define Four Tiers of AI Automation to Support Psychotherapists

University of Utah researchers have established a four-tier framework for artificial intelligence automation in psychotherapy, ranging from scripted chatbots to fully autonomous systems. This framework aims to clarify AI's evolving role, primarily as a support tool for human mental health experts rather than a replacement.

The integration of conversational artificial intelligence into mental health care is rapidly changing traditional psychotherapy. Historically, new technology supports human experts, enhancing their capabilities rather than replacing them, states Zac Imel, a professor of educational psychology and lead author of a new study on this topic. This perspective frames the ongoing exploration of how much and what types of automation are feasible and beneficial in therapy.

A team at the University of Utah, including Imel, Vivek Srikumar, and Brent Kious, defined four distinct levels of AI automation within psychotherapy. Their framework, outlined in a new study published in Current Directions in Psychological Science, provides clarity on AI's diverse applications and associated risks.

The first level, Category A, includes scripted systems where chatbots deliver prewritten content following specific decision trees. Category B involves AI evaluating therapists, reviewing sessions to provide feedback or ratings on performance. Category C focuses on AI assisting therapists by suggesting interventions, prompts, or phrasing, with human clinicians delivering the actual care. The most autonomous level, Category D, describes AI providing therapy directly, where an AI agent generates responses and interacts with patients, potentially with human supervision. Each category carries different implications for utility and risk, resembling the varying levels of automation seen in self-driving vehicles.

This structured approach helps delineate the varying impacts of AI integration on patient consent, potential mistakes, and professional responsibility. The researchers emphasize that understanding these tiers is crucial for both users and health systems navigating AI tools. A significant practical application involves improving clinician training and evaluation, areas where AI can efficiently process information that is often labor-intensive for humans.

The research team is actively collaborating with SafeUT, Utah’s statewide text-based crisis line. This partnership focuses on developing AI tools that can evaluate crisis counselors’ sessions, providing targeted feedback for skill improvement and development. This work demonstrates AI's immediate potential to enhance human expertise in mental health care. Future developments will focus on how AI can streamline therapist support systems and continue to refine mental health service delivery while maintaining high ethical standards.