The Community for Technology Leaders

In Memoriam

Pages: pp. 10-11

Joshua Lederberg (1925–2008)

Bruce G. BuchananUniversity of PittsburghGraphic:

Figure    Joshua Lederberg at a Laboratory Instrument Computer teletype, 1974. Lederberg inaugurated the Stanford University Medical Experimental Computer (SUMEX), to encourage AI applications in medicine. (photo courtesy of the US National Library of Medicine)

Joshua Lederberg died on 2 February 2008 at the age of 82. The human race lost one of its foremost champions, science lost a giant, and we in computer science lost a pioneer, visionary, advocate, and friend.

Lederberg's many and varied contributions, particularly to the diverse sciences he mastered and bridged, have been well documented, so the focus here is on his contributions to computing. His interest in computing, he told me, started with his experience programming a card reader to do computations in high school. Although his lifelong fascination with programming machines to help us think was sparked then, in his early career he was best known as a molecular biologist.

Lederberg won the Nobel Prize at age 33 for work on the exchange of genetic material between bacteria, a discovery that fueled the revolution in biology. He went to Stanford in 1958 as first chair of the new Department of Genetics.

In 1963, his interest in computing was reignited, because of its intrinsic fascination for him and because of the potential applications of computers to science. Three years later, he "entrepreneured" the ACME (Advanced Computer System for Medical Research) time-sharing system for the Stanford Medical School. It was innovative in many respects, supporting medical research, teaching, and clinical data management. Such an advanced facility was unique among medical schools of that time.

His introduction to modern computing was via BALGOL (Burroughs Algol). He chose a difficult "class project": to enumerate graphs, a task that he had addressed in high school (!) to solve chemical isomerism problems. The computing problem required new graph theory. George Pólya had determined how to count the number of unique graphs with specified numbers of nodes of different orders. But there was no algorithm for naming the graphs uniquely—or for enumerating them with a guarantee that the graphs included all isomeric forms without any duplicates. Lederberg's "dendritic algorithm" (from which came the name Dendral) does just that, and by itself was an advance in the fundamentals of chemistry.

Lederberg formed a collaboration with Edward Feigenbaum in 1965 after meeting the previous year at a workshop on computer simulation of cognition. They proposed exploring the computer modeling of theory formation in scientific thinking and invited me to join them at Stanford in what later became a large team effort. The Dendral algorithm became the generating engine in a heuristic generate-and-test program that discovered solutions to new problems in analytic chemistry.

His notational algorithm for chemical graphs was straight-forwardly mapped into a generating algorithm for acyclic graphs, written in Lisp. Under his leadership, a team of talented mathematicians and computational chemists subsequently developed a computationally efficient generator of cyclic graphs. This generator of a complete set of possible graphs, without duplicates, was the core of the Dendral program, and the first real-world application of AI problem-solving methods. In AI terms, the generator defined the solution space for any problem requiring a graph as an answer.

Such a combinatorial generator is useful practically if it can be constrained to generate a sufficiently small number of candidate answers. Thus entered heuristic pruning and evaluation into the AI toolkit. Lederberg saw that a program could use some of what he knew about mass spectrometry as heuristics to constrain the generator. He later recruited his friend Carl Djerassi to provide the necessary deep domain knowledge of mass spectrometry.

Feigenbaum and I had the pleasure (perhaps "thrill" is more descriptive) of working with him to formulate the model of knowledge-based systems. Lederberg wrote of that work:

My interest in AI has little to do with my background as a biologist, a great deal with curiosity about complex systems that follow rules of their own, and which have great potentialities in preserving the fruits of human labor, of sharing hard-won traditions with the entire community. In that sense, the knowledge based system on the computer is above all a remarkable social device, the ultimate form of publication. (

Our routine meetings would occasionally turn into a remarkable Lederberg "fireworks show" of ideas. We used to say that one of these meetings with Lederberg could generate five years of work for us. And what a struggle it was to keep up with a mind that could see solutions before the rest of us could articulate the problems!

Having gained confidence with the solution of individual hypothesis-formation problems in analytic chemistry, we returned to our original motivation: modeling theory formation. Our choice was to infer rules (constituting an empirical theory) for interpreting mass spectra from primary data (that is, from known molecular structures and their associated mass spectra). Lederberg dubbed this learning program "Meta-Dendral." In 1976, it became the first computer program to have its learned results published in the refereed scientific literature as a new scientific contribution.

An overview of the Joshua Lederberg papers archived at the National Library of Medicine summarizes Dendral's significance:

The greatest significance of Dendral, however, lay in its theoretical and scientific contribution to the development of knowledge-based computer systems. It was the ambition of Dendral's creators to transfer the principles of artificial intelligence from the realm of chess and other strictly controlled settings in which they had been formulated during the 1950s, to real-world problems facing biomedical researchers and physicians. They wanted to show that computers could become experts within a concrete knowledge domain, such as mass spectrometry, where they could solve problems, explain their own conclusions, and interact with human users. (

Lederberg's information-technology interests were broad and visionary. He played a key role in conceiving the systems required for NASA's Viking spacecrafts to collect samples from the Mars surface, analyze them, and send data back to Earth. The Lederberg/Levinthal Instrumentation Research Laboratory developed signal-processing algorithms for collecting and assigning meaning to data from a mass spectrometer. He and Feigenbaum persuaded Darpa to let the Stanford group's computer connect to the embryonic Arpanet (later the Internet) to facilitate collaboration with other groups around the country doing AI in medicine. This was the AIM part of the SUMEX-AIM (Stanford University Medical Experimental Computer for Artificial Intelligence in Medicine) computer facility, of which Lederberg was the founding principal investigator.

Lederberg was always interested in communication among scientists and was an early, and lifelong, enthusiast for the Science Citation Index and similar tools. He served on numerous editorial boards. In 1978, he published his vision of how networked computers would transform science by enabling colleagues to collaborate across long distances, with insights that have stood the test of time ("Digital Communications and the Conduct of Science: The New Literacy," Proc. IEEE, vol. 66, no. 11, 1978, pp. 1314–1319). In his many public-service roles, advising US presidents since Eisenhower on matters including disarmament, national defense, biological warfare, bioterrorism, and space exploration, he brought an appreciation of the promise and realities of computing into policy. At the same time he was thinking of these broad implications, however, he was having fun at a practical level writing programs to manipulate his email files!

Of course, Lederberg was involved in molecular genetics all this time. It was paradoxical that our group was working in chemistry, medicine, and other fields, but not his field, molecular genetics. With the discovery of recombinant DNA techniques, this changed. In 1976, he agreed that the time was right and our group started the MolGen project. (This was a pun that tickled Josh. Obviously, "MolGen" stood for Molecular Genetics, but it also reversed the parts of the name he gave Dendral's top-level function, GenMol, for Generate Molecules.) MolGen was the first large-scale effort in computational molecular biology, which is now a huge field.

The Times (London) wrote, "The benefits to mankind of his work have been enormous and they increase each year." For his very active life of public service, as well as his scientific achievements, he was awarded the highest honors. Among them are the Nobel Prize for Physiology or Medicine (1958), the National Medal of Science (1989), the Presidential Medal of Freedom (2006), the American College of Medical Informatics Morris F. Collen Award (1999), and the ACM/AAAI Allen Newell Award (1995). The Newell Award citation reads: "For pathbreaking contributions to the application of computer science research in chemistry and biology; and for leadership in building a computer-networked community of workers in these areas."

Josh inspired many to follow his vision and to share his curiosity about far-ranging questions of science and public policy. He held everyone's respect by mastering all levels of problems, from the data structures of programs to their geopolitical implications. He wanted us to feel a sense of responsibility for demystifying computing for the general public and for explaining clearly to lawmakers the implications of publicly funded research. But instead of lecturing us, he led by example—writing a weekly column for the Washington Post explaining new scientific findings, for example.

He also believed that generous acknowledgment of ideas encouraged collaboration and that collaboration was central to science. There was no one more generous in acknowledging our contributions, yet there is no one to whom we are more indebted for ideas, inspiration, and encouragement in our attempts to add to the enterprise of science that he shaped so significantly and loved so deeply.

About the Authors

Bruce G. Buchanan is University Professor of Computer Science Emeritus at the University of Pittsburgh. Contact him at
67 ms
(Ver 3.x)