Doctors Can Now Practice Tough Talks With AI Patients
Hard medical conversations, rehearsed as often as you like, without needing a human actor.
When doctors deliver bad news badly, care gets delayed, mistakes happen, and patients suffer. You'd think practice would fix that, but rehearsing hard conversations is rare. It usually takes a trained actor playing the patient plus an expert watching and grading you — expensive, slow, and hard to schedule. The original version of this system, SOPHIE, proved in a randomized trial that it helped doctors talk about serious illness, but it only covered one situation and still needed people to run it.
SOPHIE 2.0 fixes that. It's a website anyone can use on their own, with no human in the loop. You talk to an AI patient that responds in the moment, then get personalized feedback. A large language model — AI that reads and writes like a person — acts as the grader, scoring three communication skills: making the patient feel in charge (Empower), being clear rather than vague (Be Explicit), and showing you understand their feelings (Empathize). Its scores lined up closely with what individual human raters gave, a strong sign it's grading fairly.
The team tested it with 59 clinicians and students. Each person did two AI-patient encounters with feedback in between. The response was warm: 92% called it engaging, 86% found it easy to use, and 83% said it felt clinically relevant. Scores went up in the second encounter, which is encouraging.
Here's the honest catch. Because everyone got the training, there was no comparison group, so researchers can't prove the improvement came from practice rather than just familiarity with the format. The AI grader also matches one human rater, not a panel — good, not perfect — and the study didn't measure whether patients actually fare better. Still, if the tool holds up, it means doctors everywhere could rehearse the hardest moments of their job anytime, for very little money.
- SOPHIE 2.0 is a free browser tool where you practice difficult medical conversations with an AI patient and get instant feedback, no human actor or coach needed.
- It covers 24 real-world clinical scenarios and scores three skills: empowering the patient, being explicit, and showing empathy — matching human raters closely.
- Of 59 clinicians and students tested, 92% found it engaging and 83% clinically relevant, though researchers can't yet prove the practice caused the score improvements.
Why It Matters
Better-trained doctors mean clearer, kinder conversations at the scariest moments of your life — and fewer communication-driven medical errors.