In 1964, I was fortunate to find an enthusiastic collaborator, Joshua Lederberg, Professor of Genetics and
Nobel Prize winner at Stanford. He too was interested
in the question “Can AI model scientific thinking?” So
our work together began in 1965, after I joined Stanford. As an aside, Lederberg’s mind was one of great
vision and insight, one of the top minds of the 20th
century, in my view. But Lederberg was the gift that
kept giving. In 1966, Lederberg recruited for us Professor Carl Djerassi, one of the most influential
chemists of all time, the father of the Pill [birth control pill] and the head of Stanford’s mass spectrometry
As I said, I’m an empirical scientist, not at theoretical
one. I needed a test bed in which to do these AI exper-
iments. Lederberg suggested the problem that he was
working on, inferring hypotheses about organic
molecular structures from the data taken by an instru-
ment called a mass spectrometer. Lederberg was doing
research for NASA on the design of a Mars probe,
designing a mass spectrometer system for detecting
life-precursor molecules such as amino acids.
In this experimental setting, the test bed, we could
measure, month by month, how well our program —
which was called Heuristic DENDRAL, or later just
DENDRAL for short — was performing compared with
the performance of Djerassi’s PhD students and postdocs on the same problem.
Throughout the 1960s, Feigenbaum — in collaboration with Bruce Buchanan and others — continued
to evolve the model of the organic chemists in Carl
Djerassi’s laboratory, and in particular their capability to interpret the data about particular sorts of
organic compounds from their mass spectrometry
instrumentation. This modeling had two basic features. For one, the artificial intelligence researchers
developed the model of inductive reasoning processes in DENDRAL. In addition, they modeled the
organic chemists’ knowledge as a store of rules,
roughly in the form of “If, Then” statements.
AAAI Archive File Photo.
Allen Newell and Herbert Simon.