The “SI” in SI MedEd
A note on how this page is written: The field moves fast and the evidence is uneven. Below, established developments are described as established; emerging or speculative claims are labeled as such. No product endorsements appear on this page — the point is the pattern, not any particular tool.
“Super Intelligence” is the shorthand this family uses for a straightforward idea: modern AI methods can ingest the sprawling public documents of medical education — exam content outlines, admissions statistics, accreditation standards, match data — and return them as something a busy person can read, search, and question. This page describes how that same capability is reshaping medical education itself, in four areas: how students study, how they're assessed, how curricula are managed, and how institutions do the data work accreditation requires.
The most visible change is in how students learn. Large language models can now generate explanations tailored to a learner's level, produce practice questions on a weak topic, turn a dense physiology chapter into a quiz, or role-play a patient for history-taking practice. Medical students have adopted these tools rapidly — this is observable from published surveys of trainee AI use, which consistently show large and growing adoption.
The caveats matter as much as the capability. Language models hallucinate — they can produce confident, fluent explanations of incorrect facts, and medical detail is exactly where an error does the most harm. Peer-reviewed evaluations have shown that while leading models can score at or above passing thresholds on multiple-choice medical benchmarks, that performance does not transfer cleanly to real clinical reasoning, and a model's answer should never be the final word on a factual claim. The responsible pattern emerging in medical schools is explicit: use AI to explain and generate, verify against primary sources — the textbook, the content outline, the primary paper — before trusting the output. Several schools now include AI literacy and verification habits directly in the curriculum, which is a development worth watching.
Assessment is changing more slowly than study, and for good reason: licensing exams and course grades have to be defensible. Still, three shifts are real. First, automated scoring of written and spoken responses — long researched in the NBME's ecosystem — is becoming more capable, letting programs score clinical reasoning narratives and patient notes at scale. Second, AI-generated items are entering practice question pools: models can draft multiple-choice questions aligned to a content outline, with human psychometricians reviewing and calibrating them — the human review remains essential, because generated items carry the same hallucination risk as any model output. Third, adaptive and formative assessment is getting cheaper to build: short quizzes that adjust difficulty, or dashboards that flag a struggling student earlier than a midterm would.
What hasn't changed: the psychometric bar. An AI-assisted assessment still has to demonstrate validity, reliability, fairness across groups, and security of content. The organizations that run high-stakes exams publish their own positions on AI-assisted item writing and scoring; those positions are authoritative, and schools should follow them rather than this page's summary. The honest state of the field is that AI is a promising assistant in assessment work, not yet a replacement for the measurement science that high-stakes decisions require.
Every medical school must map its curriculum to program objectives and accreditation elements — a massive documentation task that has traditionally meant spreadsheets and committee hours. This is where AI methods are arguably most useful right now: natural-language processing can align session objectives to program-level objectives, flag gaps in coverage of a standard's elements, and draft the narrative summaries that go into accreditation self-study documents.
But the boundary is sharp and worth respecting. An AI system can draft a mapping; it cannot verify that a session actually taught what its slide deck claims. Survey teams and accreditation bodies expect evidence — and evidence means the underlying documents and data, not a well-written summary of them. The emerging best practice is that AI drafts, humans attest: every AI-assisted line in an accreditation document should be traceable to a real source and owned by a named person. Nothing about a self-study should be generated and left unverified.
Accreditation self-study has historically been a periodic scramble: assemble the DCI, run the independent student analysis, bind the evidence, survive the visit, and shelve it all for eight years. The data-rich version of this cycle — continuous quality improvement dashboards, automated survey analysis, longitudinal tracking of graduate outcomes — is increasingly feasible with AI-assisted pipelines. Schools are building dashboards that pull from course evaluations, exam performance, and student surveys to give leadership a standing picture of program health, rather than a snapshot assembled under deadline.
The limitation is data quality, not model capability. An AI pipeline over bad data produces confident garbage; the schools getting value from continuous monitoring are the ones that invested in clean data collection first — consistent evaluation instruments, real response rates, and disciplined record-keeping. The technology is the easy part. The governance is the hard part.
Because this page is part of an AI-named project, it owes readers candor about where the hype outruns the evidence:
SI MedEd is itself an example of the thesis: public documents of medical education, digested with modern AI methods and returned as plain-language guides — with the sources linked, the speculation labeled, and the limits stated. If that approach is useful, the guides on this portal are the proof of work. Start with the About page for the editorial approach, and read the disclaimer for what these guides are not.
Reminder: This is an unofficial resource — not affiliated with, sponsored, or endorsed by the NBME, USMLE, LCME, NRMP, AAMC, or any medical school. See the full disclaimer.