In any field at a given time, a relatively small set of generally accepted paradigms is subscribed to by workers in the field. In the case of artificial intelligence, one of these, arguably the most successful, is the technology known as expert systems. Computer science is a young field and AI is even younger (although not much), so it is often possible to trace the exact history that has led to present practice.
For expert systems, the progenitor is pretty clearly the DENDRAL system, although many other projects and ideas have had significant influence. This paper describes the history of DENDRAL, outlining its problem background, development history, present state, and lessons learned. I recommend it highly not only to AI practitioners but to software developers in general. Many of the experiences and observations apply generally to all software. The writing style is clear, and the level of exposition makes it mostly accessible to anyone in computer science. (Section 3 contains perhaps a bit more detail than many readers will be interested in, but the introductory paragraphs of each section provide an adequate summary).
We can hope that other development teams for well-known projects will find the time for their own retrospectives. Santayana’s comment that “Those who cannot remember the past are condemned to repeat it” has already been verified too many times in computer science.