Medicine has been "about to be transformed by AI" since expert systems such as MYCIN in the 1970s. The record since then is mixed: narrow, well-validated tools (diabetic-retinopathy screening, stroke triage, ECG interpretation) have earned regulatory clearance and some payment, while the most publicised bet, IBM Watson Health, was sold in parts in 2022. Large language models changed the picture after late 2022. The first category to spread quickly was not diagnosis but paperwork: ambient AI scribes that draft clinical notes from a recorded visit. This section covers the main product categories, how FDA regulates software as a medical device, how AI gets paid for, the evidence so far, the bias and safety record, the payer side (prior authorization and claims algorithms), and the company landscape. The supporting files work through ambient-scribe ROI and imaging-AI base rates in Python and give a diligence checklist.
What the study covers
- 1Overview
- 2A short history
- 3Product categories in more detail
- 4Mechanics
- 5Large language models in clinical care
- 6Evidence so far
- 7Bias, safety and governance
- 8The payer side: prior authorization and claims algorithms
- 9Key numbers
- 10Company landscape
- 11Landmark cases
- 12Pitfalls
- 13How an expert thinks about it
- 14Glossary
Technical exhibits
Worked examples, checklists and case studies included with this study.
Ambient Scribe ROI and Imaging AI Base Rate Worked Example
Stylized ROI of an ambient-scribe rollout for 100 clinicians under four scenarios, and positive predictive value of an imaging algorithm at different disease prevalences (computed in Python)
Healthcare AI Diligence Checklist
Questions for health-system buyers, investors and product teams evaluating a clinical or administrative AI product