How to Choose the Right AI Tool (Before It Chooses You)
September 30, 2025
•
5
min read
Every healthcare provider has met a patient who had Googled their symptoms and sworn they were dying. A patient with a headache thought they had a brain tumor. Another saw a friend’s diagnosis of MS and demanded a spinal tap for herself. And of course, none of this is new. After a spate of anthrax cases in 2001 (immediately after 9/11, if you recall), doctors’ offices were flooded with thousands of patients imagining they had been exposed (there were only 21 confirmed cases when all was said and done).
The lesson: powerful tools in the wrong hands cause problems. If you’re anxiously Googling symptoms, WebMD will combine with the anxiety to create a misdiagnosis.
If you’re considering which AI-enabled tools to use with your students, you face a similar dilemma: how do you pick the right tools for the job, consider the negative externalities and manage the implementation, even as the underlying technology changes quickly?
We’ve got a rubric to help.
The Six-Question Rubric
Run every AI tool you’re considering through these six questions before adopting.
You have to understand the technology BEFORE you use it. One way to do that - as we’ve written about before - is creating a model card, where you jot down key information about the model to understand the risks and possible rewards of its intended use. But whether you write it out or merely think it through, this rubric is simple but flexible enough to be used by everyone (faculty, students, patients, etc.)
The rubric doesn’t fix the problems with AI - hallucination, bias, etc. - but it DOES help HUMANS clarify the downsides and decide what to use and when to use it.
The FDA approved an AI tool that provides a differential diagnosis using photographs of skin conditions and medical history.
It was developed using 16,000 cases and a convolutional neural network to output prediction scores across 400 skin diseases.
Should I use this tool?
In a retrospective study, the AI tool was superior to primary care clinicians; use was associated with improved diagnoses
for 1 in every 10 cases. A prospective study in a clinical setting has not been done yet.
When should I use this tool?
You decide to use this AI tool to augment your diagnostic ability for skin conditions where the diagnosis is unclear. You use
it to inform, not override, your decisions regarding treatment, biopsies, and referrals in a way that boosts accuracy,
quality of care, and resource stewardship.
How do I use this tool?
You learn to take clinical photographs of skin conditions as required by the AI tool and generate a differential diagnosis using it.
You do this seamlessly and efficiently during physical exams.
How should I communicate with patients regarding the use of the tool?
You discuss with the patient why and how the tool is being used and answer questions regarding privacy, ultimately building trust and confidence.
What are the “side effects” of this tool?
Foundational knowledge: a convolutional neural network is a “black box.” Guard against cognitive biases from seeing only the final suggestion.
Critical appraisal: Fitzpatrick skin types I and V are under-represented; type VI is absent in the dataset.
Medical decision making: anticipate lower accuracy for under-represented skin types; adjust utilization accordingly.
Technical use: take the appropriate steps when the tool delivers an error message.
Patient communication: explain why your diagnosis may differ from the tool’s suggestion and use shared decision making.
Applying the Rubric to Competencies
If you have the time, I recommend you read the entire original article. For one, the writing’s excellent (even judging by non-research-paper standards). But also, the article maps the AI rubric above onto the competencies required of different learner levels. So, the first question maps like so:
Domain (competency)
Medical Students
Residents
Faculty
Foundational Knowledge
Explain the fundamentals of AI and how AI-based tools are created and evaluated
All prior competencies +
Explain the critical regulatory and socio-legal issues surrounding AI-based tools as they relate to practice
All prior competencies +
Explain, teach, and shape the current and emerging roles of AI in health care
In other words, we expect faculty to know more about AI than their students so they can lead by example. That’s a tall order for an emerging technology, but it’s exactly why you subscribe to this newsletter, right?
Kidding aside, it’s also why rubrics like this can help simplify complex, evolving information. Use the rubric, apply your best judgment, and you’ll be on a solid path.
How to Decide if InSitu is Right for the Classroom
Let’s use this rubric on a possible tool that faculty at health professions programs across the US are starting to use: ReelDx’s InSitu.
If you’re considering using the tool, then start with the rubric. We’ve filled out the table with some imagined responses for US healthcare faculty, but you can tweak them to fit your own.
Question
InSitu Explanation
What is this tool?
InSitu is a ReelDx simulation platform that integrates:
(1) Real patient video.
(2) A large language model (LLM) fine-tuned on transcripts and medical corpora.
(3) A rules engine for evidence-based answers and patient interviewing practice.
Should I use this tool?
Programs use InSitu to practice clinical reasoning, forcing students to interact with lifelike patient bots and engage in clinical reasoning with a preceptor.
When should I use this tool?
Use in class to reinforce diagnosis of the content being consumed.
Ideal when live standardized patients are scarce or costly.
Can also be used for remediation.
How do I use this tool?
Faculty select cases to supplement course material. Students log in to interview AI-powered patient and debate differential diagnosis.
Automatic, immediate feedback on interview provided by AI; feedback or grading can be done by faculty after-the-fact.
How should I talk to students about it?
Instructors are transparent about AI limits — e.g., hallucinations, and taking the feedback with a grain of salt. They emphasize that it’s practice with an interactive partner to help facilitate the transition to the clinic.
What are the side effects?
Privacy: student voices recorded in order to provide feedback (though learner information is de-identified).
Technical use: as with any emerging technologies, there are always unexpected hitches, which is especially true with generative AI
Not a replacement for traditional tools: as with any classroom supplement, this tool does not replace (yet) the other tools students can use to build their clinical reasoning.
In Conclusion
The article introducing the original rubric makes a good point: that AI - if integrated well, can fulfill the “Quintuple Aim” of patient care: better clinical outcomes, patient experiences, provider satisfaction, financial sustainability and health equity.
We could imagine a Quintuple Aim for the classroom as well. AI can improve all of the following:
The emphasis should be on those question marks. Do the tools we introduce meaningfully improve the experience for learners and faculty alike? Does it help or hurt accessibility? Does it meaningfully move the needle on learner outcomes?
We can fulfill these aims only if we choose our tools carefully.
