Open the Pod Bay Doors · Episode 5 of 10 · 30 min
MYCIN Beat Stanford's Doctors in 1979 and Was Never Used
After HAL, real AI spent decades barely able to do anything, and funding collapsed on two continents. The episode follows ELIZA, the ALPAC and Lighthill reports, and MYCIN, which matched or beat Stanford faculty on meningitis antibiotic choices in a blinded 1979 test yet was never deployed. No electronic record to read and no answer on liability stopped it.
Who's who
- Joseph WeizenbaumMIT creator of ELIZA who later argued against its uses
- Edward ShortliffeAuthor of MYCIN as Stanford doctoral work
- Bruce G. BuchananShortliffe's advisor; brought rule-based expert knowledge from DENDRAL
- James LighthillCambridge professor whose 1972 report on AI led to funding cuts
- Donald MichieEdinburgh AI researcher who argued against Lighthill
- John McCarthy+1 more episodeCoined 'artificial intelligence'; debated Lighthill
- John R. Pierce+1 more episodeBell Labs engineer who chaired the ALPAC committee
- Joshua LederbergSecured NIH funding for the SUMEX-AIM computing resource
Where it happens
- Stanford University, California+8 more episodesHome of MYCIN and the SUMEX-AIM computer
- Royal Institution, LondonSite of the televised Lighthill debate on May 9, 1973
- Pittsburgh, Pennsylvania+4 more episodesWhere INTERNIST-1 was developed
- Dartmouth College, New Hampshire+1 more episodeBirthplace of the time-sharing system and BASIC
Timeline
- 1966ALPAC report finds machine translation slower and costlier than human translation
- 1969The Mansfield Amendment pushes defense research toward mission-oriented work
- July 1972Lighthill delivers his survey of artificial intelligence
- May 9, 1973Televised Royal Institution debate on the Lighthill report
- 1974DARPA cancels its speech understanding program
- 1979JAMA evaluation rates MYCIN acceptable on 65 percent of cases versus a faculty mean of 55.5
Still debated
Whether MYCIN's non-deployment was a failure of technology or of institutions is argued, as is whether the AI winters were a funding correction or an intellectual one. Whether today's clinical AI has learned the lesson, or merely gained better plumbing, remains open.
← PreviousHow Kubrick Built HAL 9000 — and Why HAL Actually RefusesNext →Therac-25, MCAS and the Machines Already Overruling Humans
