Could AI help move physical therapy education toward a competency-based model that provides students with high-quality, real-time feedback as they learn? Assistant Clinical Professor Mariano Wechsler, PT, DPT, thinks so.
Dr. Wechsler has long been curious about the potential of artificial intelligence in education. Influenced by his extensive reading on the subject and UCSF symposiums focused on AI in research and teaching, he began to see opportunities to integrate the technology into physical therapy training.
"I read a lot about business case scenarios where people were using chatbots to train sales representatives on how to improve their communication with customers. That sounded familiar. If they’re doing it, why aren’t we doing it?” he recalled.
Through trial and error, Dr. Wechsler narrowed his focus to one of the most persistent challenges learners face during their early clinical rotations: developing strong history-taking and communication skills.
"Little by little, I started defining a problem I thought I could solve,” Dr. Wechsler said. "I saw history-taking as a gap." He then began building his own AI tools.
The first chatbot generated patient cases quickly for students to practice. The second acted as a simulated patient that conversed with students via voice features in scenarios such as, “My knee hurts, I was playing soccer, I twisted it and it’s been hurting ever since.”
The third chatbot was a Clinical Reasoning Tutor, designed to provide feedback based on an evidence-based rubric. It evaluated skills such as building rapport, identifying psychosocial factors and creating differential diagnoses, then highlighted strengths and gaps. For example: “You did very well assessing the pain presentation, but you forgot to ask about red flag questions.” Previously, students relied on peer-to-peer role play, which was limited in frequency, realism and feedback. The literature shows simulated patients yield stronger results, and with chatbots, students can practice as often as they want, receiving immediate, individualized feedback.
Testing, Feedback, and Iteration
The project has already generated preliminary data through a student questionnaire and has been accepted for presentation at the 2026 APTA Combined Sections Meeting. Dr. Wechsler is collaborating with Vincent Ann, PT, DPT, to further develop the simulated AI patient through an application to the UCSF Innovation Funding in Education Grant. Multiple DPT faculty colleagues are helping him integrate this technology into the physical therapy curriculum, including Ivan Arriaga, PT, DPT, Tamar Brand-Perez, PT, DPT, and Sarah Pawlowsky, PT, DPT.
“At the beginning, the results I got with the chatbot were not great. It was clunky and hard to work with. But I just kept at it by narrowing the problem and trying different models. Students helped me build and test the model and I worked closely with the UCSF AI Tiger Team. Together, we eventually built something students could use,” he said.
Next steps include testing the quality of the assessments and technical improvements, such as adding multi-session memory to track student progress across trials. Dr. Wechsler is working with developers to build a platform and fine-tune the models.
Toward Precision and Competency-Based Education
This project represents an important step toward precision education, where learners receive individualized feedback tailored to their specific needs. It also serves as a bridge to competency-based education, which emphasizes clinical performance and mastery over time.
By providing frequent, high-quality assessment and feedback, the department is creating customized learning experiences that mirror real-world clinical practice and support each learner’s growth, Dr. Wechsler said.
A Philosophy of Openness and Experimentation
For Dr. Wechsler, the process of experimenting with AI has reinforced the importance of curiosity, persistence and collaboration. His advice to others: “Narrow down your problem and iterate. Read what others are doing, even outside the medical field. And don’t be afraid to reach out, especially at UCSF, everyone has been so open and willing to collaborate.”
While acknowledging concerns about AI in education, for example, that it could hinder students’ clinical reasoning if they become too dependent on it, Dr. Wechsler believes that AI models can be used to create learning activities that were never possible before.
“By listening to our students, collaborating and thinking creatively, we can drastically improve physical therapy education.”