The DENDRAL system is an expert system that helps chemists determine the molecular structure of an unknown substance. It began development at Stanford University in the United States in 1965 and was successfully developed in 1968. It is the result of collaboration between Feigenbaum and chemist J. Lederberg. In the mid-1960s, Lederberg proposed an algorithm that could list all possible molecular structures based on input mass spectrometry data, and over the next three years, he explored with Feigenbaum and others methods for establishing knowledge systems using rule representation, culminating in the creation of the DENDRAL system. The aim was to utilize this system to complete the task of listing all possible molecular structures in a shorter time than manual efforts. DENDRAL is the world’s first successful expert system, and its emergence marks the birth of a new field in artificial intelligence—expert systems. Since then, various different expert systems have been established.
Application Domains
Molecular structure of substances
The DENDRAL system primarily utilizes mass spectrometry data as its original information. The entire system can be divided into three functional parts.
① Planning: Using mass spectrometry data and chemists’ experiential knowledge regarding the relationship between mass spectrometry data and molecular structure to form several constraints on possible molecular structures.
② Structure Generation: Using J. Lederberg’s algorithm to provide some possible molecular structures, controlling the expansion of these possibilities with the constraints generated in the first part, ultimately yielding one or several possible structures.
③ Evaluation: Using chemists’ knowledge of mass spectrometry data to assess and rank the results provided in the second part. Finally, a molecular structure diagram is produced.
The development of expert systems once flourished but has since become obscure. Expert systems generally consist of six components: knowledge base, database, inference engine, user interaction layer, interpreter, and knowledge acquisition module, varying in structure based on different needs. Among these, the knowledge base and inference engine are the core components of the system.
The knowledge base accurately, concisely, and effectively converts expert knowledge into a language understandable by machines, commonly using representation methods such as production rules, frame representation, and semantic networks.
The inference engine is the “thinking” structure of the expert system, solving problems by simulating the expert’s thought processes, primarily through forward reasoning, backward reasoning, and mixed reasoning.
Expert systems have technically gone through four stages: incubation, emergence, maturity, and development. In 1956, the Dartmouth Conference was held in the United States, where the term “artificial intelligence” was first adopted, marking the official birth of the AI discipline. Subsequently, artificial intelligence rapidly developed in three directions:
One is machine thinking, such as machine proof, machine learning heuristic programs, and expert systems for chemical analysis and medical diagnosis;
The second is machine perception, such as machine vision, machine hearing, text and image recognition, natural language understanding, as well as perceptrons and neural networks;
The third is machine behavior, including intelligent control systems with self-learning, self-adaptive, and self-organizing characteristics, cybernetic animals, and intelligent robots. In 1965, Edward Feigenbaum, a computer scientist at Stanford University in the United States, began developing the world’s first expert system for inferring chemical molecular structures, DENDRAL, marking the impending birth of the subdiscipline of “expert systems” within artificial intelligence. In 1968, DENDRAL was successfully launched, opening a new branch of artificial intelligence—”expert systems”.
In the 1970s, expert system technology matured and was widely used in other fields. Edward H. Shortliffe and others at Stanford University began in 1971. In 1976, they completed the first medical expert system, MYCIN, for diagnosing, treating, and consulting on blood infections. Richard O. Duda and others at Stanford Research Institute began their work in 1976.
In 1981, they completed the geological exploration expert system, PROSPECTOR. In 1977, Feigenbaum proposed the concept of “knowledge engineering,” greatly promoting the development of knowledge-based expert systems and their development tools, such as the skeletal expert system development work EMYCN, KAS, knowledge acquisition aids TEIRESIES, SEEK, and general knowledge representation languages LISP, PROLOG.
In the 1980s, expert systems began to leave the laboratory and enter the market. In 1981, Professor William F. Clocksin of the University of Hertfordshire published “PROLOG Language Programming.” In 1982, the first commercial expert system, R1, successfully ran at Digital Equipment Corporation (DEC). In 1983, Professor Barbara Hayes-Roth of Stanford University published “Building Expert Systems.”
In 1985, Professor Paul Harmon of the University of California published “Expert Systems: Artificial Intelligence in Business.” It is reported that nearly one company in this field was founded every week, and as expert systems and their tools became increasingly commercialized, a knowledge industry aimed at producing and processing knowledge emerged, ushering in the “golden age” of expert systems.
However, due to the overly narrow application domains of expert systems, the “bottleneck” in knowledge acquisition, and difficulties in uncertain reasoning, commercial demand sharply declined in the late 1980s, leading to the coldest “winter” in the history of artificial intelligence represented by expert systems. In the 1990s, expert systems began to enter a slow development period, with research shifting towards the integration of knowledge engineering, fuzzy technology, real-time operating technology, neural network technology, and database technology.
Scan to follow for more information