At an accident scene, first responders need to do more than provide treatment. They also need to speak with the injured person: ask where it hurts, listen to the answer, and decide what to do next. How can they practise that conversation in a realistic setting?

In our research, we created a mixed reality traffic accident simulation with a virtual patient powered by ChatGPT. The patient could answer spoken questions, giving first responders a chance to practise both assessment and communication. We tested the simulation with 24 medical first responders.

Participants responded positively to the idea of talking with a virtual patient, and many found the interaction useful for training. The study also showed where the experience needs work: the patient took about three seconds to answer, usually waited to be asked a question, and could not yet carry a conversation like a person in distress.

The project shows how AI could make emergency training more flexible by letting trainees practise conversations with a simulated patient. Our next challenge is to make those conversations faster, more natural, and better guided by the training scenario.

Read the full study in JMIR Formative Research.