EPIC, describing AI Failures are carefully worded and selected and meant to consider the level of failure when understanding MYCIN, a significant and historical rules-based LISP expert system. MYCIN began in 1974 with a team of Stanford MD's and PhDs led by Bruce G. Buchanan and Edward H. Shortliffe, spanning 10 years of MYCIN experiments and culminating with a publication of the classic MYCIN AI study, "Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project" by Buchanan and Shortliffe.
For a Masters project in the mid-80's, Joe Jesson was given the assignment to read (748 pages!), write an analysis of this work, and to write in LISP a backward-chain inference engine.
The book focused on the need for MD's to rapidly identify both the specific bacteria infection agent and corresponding antibiotics.
Twenty two years after MYCIN was released and no longer used by MD’s, Joe's sister, father, and mother were all victims of the lengthy time (~ 3 days) hospitals and labs required to incubate (in a petri dish) and identify both the infectious bacteria and the effective antibiotic. Three members of his family passed after 2 days and just before the answers came in from the lab. Hence his frustration.
We are in the year 2024, and the same test was applied in the 80's but the number of effective antibiotics is reduced! In some cases, referenced to the number zero. IBM's WATSON was introduced in 2011 as the AI question-answering computer that beat Ken Jennings in "Jeopardy". IBM focused WATSON on answering significant medical problems and we will discuss why this system failed even after IBM spent over 60 million dollars and created Medical partnerships. To end on a positive note, we will mention embedded ML and prove smart sensors have been measurably successful. You will find this well-referenced speech fascinating!
Негізгі бет EPIC AI Failures; Two AI Medical Case studies, MYCIN and WATSON - TCF2024, track7
Пікірлер