Beat 1 · Concrete
A consult, rule by rule
Symptoms enter on the left; rules fire down a chain — IF this AND that, THEN — until a diagnosis emerges.
symptoms & rule mechanism
a rule firing
diagnosis reached
motion reduced — chain frozen at the diagnosis
Beat 2 · Abstract
The rule base, as a graph
Strip the consult away: it is just conditions lighting a path to one conclusion. Same firing, no patient.
the path that fires
the conclusion it reaches
rules left unfired
motion reduced — frozen with the meningitis path lit
Beat 3 · Interactive
Run the consult — then break it
Toggle symptoms; watch rules fire to a verdict. Then feed an unseen sign — no rule matches, and it stalls.
Toggle symptoms to run the consult
symptom active
rule fired / confident diagnosis
no rule matches — stalled
The boom, in three names
MYCIN · DENDRAL
The first experts
Stanford's DENDRAL (chemistry) and MYCIN (~600 rules for blood infections) proved rules could rival specialists — yet were never deployed clinically.
XCON / R1 · DEC
The money case
Digital's XCON configured VAX orders from thousands of rules, reportedly saving ~$25M a year — the proof that expert systems paid.
The bottleneck
Knowledge acquisition
Every rule had to be hand-extracted from an expert. Rule bases grew brittle and unmaintainable — the wall the Second Winter ran into.