An automated doctor model is different in kind to a requirements model; the requirements model has to immediately read a 50 or 200 page specification and get it all right on the day, to the point of near perfection. A doctor model can be helped to read 5000 pages, say, and then acquire a few attributes about the patient - age, gender, a few symptoms, and come up with a diagnosis (that doesn't have to be perfect, or requires another test). The semantic structure will be very large (we have already blown up the dictionary to 50,000 words, which should do for triage or individual specialties), but most of it will be quiescent most of the time.
The automated doctor model and the specialist models will be relying on updates from text - they are going to be too big to do anything much by hand.
We will also be looking at waystations to the goal -a Medical Note.
Birth Defect - timelines
Causation - handling multiple possible causes
Context Sensitive States - a presentation
Diseases - a variety and how to model them
Orientation - a blind system finds its way around
Sample Document (from Merck)
Simple - high can be good, indifferent, or bad
There Are No Signs of Disease - handling No
ToBeSimilar - using similarity to reduce the load
With Cardiopulmonary Bypass – the perils of induction
Medical Knowledge Gallery - pictures of the network