[McC83], discusses some elaboration tolerance problems with the early expert system MYCIN, although the phrase elaboration tolerance isn't used.
MYCIN had a database of bacterial diseases, symptoms, results of tests and treatments and relations among these entities. It asked questions about the symptoms and the results of tests and recommended treatments or more tests. Provided what it was asked to do fitted its limitations, it did as well as medical school faculty in spite of the fact that it knew nothing of patients, doctors, sickness, health, death or events occurring in time. If told, ``I tried the treatment you now recommend yesterday, but the patient died'', MYCIN could only reply ``unrecognized response''.
If a certain antibiotic was contraindicated for patients running a fever, it could not be altered to think of giving the patient aspirin till the fever went away and then giving the antibiotic, because it knew nothing of events occurring in time.
MYCIN was never challenged in these directions by the reviewers of the papers about it.
We will suggest some elaboration tolerance challenges to knowledge bases being developed under the HPKB initiative. The challenges can either be suggestions to the designers of the knowledge bases or to the organization(s) preparing the formal challenges.