The Therabot Is In: Categorizing Automation of Therapeutic Interactions
Aimed at integrating cutting-edge psychological science into the classroom, columns about teaching Current Directions in Psychological Science offer advice and how-to guidance about teaching a particular area of research or topic in psychological science that has been the focus of an article in the APS journal Current Directions in Psychological Science.
Generative AI has invaded nearly every aspect of our lives. People use AI to write emails, generate recipes, summarize articles, and create presentations (Mollick, 2024). Large language models (LLMs) and generative AI are also modifying the teaching and learning landscape (Bowen & Watson, 2026; Vee et al., 2026). One area that receives a lot of media attention is the use of AI in therapy, with researchers scrambling to test the efficacy of AI chatbots. There is growing evidence that AI therapy can be helpful (Bodner et al., 2026), but questions remain about whether chatbots can empathize (Gabriels & Goffin, 2026).
LLMs, by virtue of being trained on human writing and language, are very capable of simulating human discourse. You can type in a question, and you will get a response that is relevant. Most LLMs can be convincingly human-sounding, but there are currently minimal checks and balances on the content of the exchanges. For use in therapy, special LLMs are trained with millions of therapy sessions to generate relevant responses and provide counselors with more effective statements (Imel et al., 2026). As the sophistication of LLMs increases and more companies strive to automate therapy, it is important to understand how automation can vary so as to guide one’s own use of the technology if the need arises.
Enter a new framework to guide both the creation and use of AI in psychotherapy: Zac Imel and colleagues (2026) organize AI into four major categories, each with different sublevels. Categories vary in the type of human involvement in the automation and the intellectual sophistication of the system.
In Category A, all material used by the AI is created by a human directly for the purposes of therapy. Patient responses may be limited to forced-choice responses or may be open ended, and an algorithm selects the best pre-scripted human response to share with the patient.
In Category B, human therapists’ communications are evaluated by AI. The system evaluates the treatment, summarizes the session, and provides real-time feedback, allowing the therapist to adjust during the session. A human can then use the summaries for quality control or training.
In Category C, human therapists are assisted by AI but are still the primary care providers. Although similar to Category B in that there is a human clinician in the process, here, AI provides the patients with real-time suggestions about clinical interventions. For example, the AI may suggest the provider be more empathetic or prompt the therapist to ask about something the client just said.
Finally, in Category D, AI provides care directly, with varying levels of human supervision. Clinical materials in this category are not written by humans, but there is still human oversight.
The categorization of automation in the context of therapy has many significant ethical and public policy implications. Factors that can shape decision making around the use of AI across the categories include the level of clinical risk, evidence for specific implications, the complexity of the judgment required by the provider, and the availability of human backup (Imel et al., 2026). Many of the considerations regarding automation also map onto higher education—one could imagine a similar framework being used in the creation of something like Teacherbots, for instance.
To better prepare our students for advancements in AI—and to guide our own usage and interactions with LLMs—reflecting on the different categories of automation is time well spent. The classroom activities below provide ways to catalyze and inform conversations on this topic.
Student Activities
Activity #1
Have students explore an AI companion app (e.g., Character.ai, Replika, Kindroid, Nomi, Candy.ai, EVA). Instruct students to “try on” different roles with the chatbot, reflect on the experience in a short paper, and then discuss in a later class. (Users sign up and are given a range of options to customize their companions. All the apps have a free version that will suffice for this activity.) Students can create any number of characters, setting the physical characteristics and the personalities. Once they have laid out the basic parameters, instruct them to interact with the AI persona for no more than one hour (or 30 minutes if you want a shorter assignment). Some examples of the roles students might create include the following:
- A college professor who you can ask for academic advice. Play the role of a first-year student new to college and interact with the “professor,” asking for best ways to study or cope with the stressors of school.
- An AI “friend” to whom you can be a companion. Play the role of a friend, share the events of the day or week, simulate dealing with a difficult situation, and ask for advice.
Activity #2
Have students complete one or more different measures of AI–human interaction and discuss the findings. For example, the AI-Technologies (Computer) Attitude Scale (Grassini, 2023) is a 4-item scale evaluating public perceptions of AI technology. Students use a Likert scale to rate their agreement with the following statements: “I believe that AI will improve my life,” “I believe AI will improve my work,” “I think I will use AI technology in the future,” and “I think AI technology is positive for humanity.”
An important skill for students to develop is to think critically about when and how to use AI. To build AI literacy and direct and guide thinking about AI, have students take the Collaborative AI Literacy Scaleand the Collaborative AI Metacognition Scale(Sidra & Mason, 2026), and reflect on how literacy and metacognition can moderate human interactions with automated chatbots.
Feedback on this article? Email [email protected] or login to comment.
Additional References
Bodner, R., Lim, K., Schneider, R., & Torous, J. (2026). Efficacy and risks of artificial intelligence chatbots for anxiety and depression: A narrative review of recent clinical studies. Current Opinion in Psychiatry, 39(1), 19–25.
Bowen, J. A., & Watson, C. E. (2026). Teaching with AI: A practical guide to a new era of human learning (2nd ed.). Johns Hopkins University Press.
Gabriels, K., & Goffin, K. (2026). Therapy chatbots and emotional complexity: Do therapy chatbots really empathize? Current Opinion in Psychology, 68, Article 102268.
Grassini, S. (2023). Development and validation of the AI attitude scale (AIAS-4): A brief measure of general attitude towards artificial intelligence. Frontiers in Psychology, 14,Article 1191628.
Mollick, E. (2024). Co-intelligence: Living and working with AI. Penguin.
Montenegro-Rueda, M., Fernández Cerero, J., Fernández Batanero, J., & Meneses, E. (2023). Impact of the implementation of ChatGPT in education: A systematic review. Computers, 12(8), Article 153.
Sidra, S., & Mason, C. (2026). Generative AI in human-AI collaboration: Validation of the Collaborative AI Literacy and Collaborative AI Metacognition scales for effective use. International Journal of Human–Computer Interaction, 42(7), 5084–5108.
Vee, A., Watkins, M., & Bruff, D. (2026). The Norton guide to AI-Aware teaching. Norton.
APS regularly opens certain online articles for discussion on our website. Effective February 2021, you must be a logged-in APS member to post comments. By posting a comment, you agree to our Community Guidelines and the display of your profile information, including your name and affiliation. Any opinions, findings, conclusions, or recommendations present in article comments are those of the writers and do not necessarily reflect the views of APS or the article’s author. For more information, please see our Community Guidelines.
Please login with your APS account to comment.