AI Standardized Patient
An AI standardized patient is software that plays a patient role in a training encounter, responding to what a learner says in real time, usually by voice, and then scoring the conversation against a named standard, with the learner's own words quoted under each score. The category name borrows from standardized patient work because the goal is the same, a consistent portrayal a learner can be assessed against, with the portrayal generated by a model rather than performed by a trained person. AI standardized patients extend an SP program rather than replace it.
The definitional line between an AI standardized patient and older virtual patient simulation is how open the interaction is. A screen-based virtual patient usually asks the learner to pick from written options, so what the learner practices is selecting a question. An AI standardized patient takes whatever the learner actually says, in their own words and at their own pace, which is closer to the skill being trained when the objective is a high-stakes conversation rather than a diagnostic pathway.
The honest case for the category is capacity, not superiority. Trained SPs, faculty observers, and rooms are all finite, so most programs can schedule only a small number of SP encounters per learner per year, and that is rarely enough repetition for something like disclosing a serious diagnosis. Learners can rehearse against an AI patient late at night and as often as they want, each attempt scored against the standard the program already holds, so they arrive at the SP encounter knowing what worked and what to practice next.
What an AI patient does not do is the part human SPs are uniquely good at. A person who has portrayed a grieving parent many times can tell a learner how one particular sentence landed, and can improvise in a way that reveals something no rubric anticipated. The sensible design is layered, with AI patients carrying volume and repetition while SP encounters carry the graded moments and the feedback only a person in the room can give, and programs that treat the two as substitutes tend to lose what made their SP program worth defending.
What this looks like in a program
- Programs typically use AI patients for the repetitions that come before an assessment, then keep the SP encounter as the graded checkpoint. Early attempts show each participant what worked and what to practice before anyone books an SP.
- Encounters that are transcribed give a competency committee the learner's actual words rather than a recalled impression of them.
- The case still has to be written, so the case library and rubric discipline that ASPE describes for SP work carries straight over.
Last updated September 2026