Blast from the past - an article on AI from 1983

tl;dr; - Japanese "expert systems" were going to take over the world.

I saved an article from 1983 that was predicting the future of computers and AI. At the time there was a lot of talk of the “fifth generation” of computers that were being build by the Japanese, and “expert systems” which were destined to revolutionize the world.

Today I used Google AI Studio to extract the text of the article from the above photos.

Here it is, as it appeared in “Daily News” in April of 1983:

Japanese Computer Challenge

By EDWARD EDELSON
Science Editor of The News

Second of two articles

IN 1981, THE JAPANESE invited some American observers to attend a meeting on what they call their Fifth Generation Computer Systems program. Visitors such as Richard Dolen of the Office of Naval Research carried home reports about a research effort with a breathtaking goal: to make supercomputers that can see, listen, talk and think - caused something of a debate among computer authorities in this country. Some, including Dolen and Edward Feigenbaum of Stanford University, believe that the Japanese effort is the clearest and most urgent challenge yet to American dominance of the computer field.

Others, including Jacob Schwartz of New York University’s Courant Institute of Mathematical Sciences, say that the Japanese Fifth Generation challenge is not all that serious. That estimation is based on a judgment that the Japanese are setting unrealistic goals.

AT THE HEART of the debate are different evaluations of the field called artificial intelligence, which has been a part of computer research since the 1950s. Artificial intelligence has several subdivisions, one of which is using robots for manual labor. Another effort that is drawing attention and controversy is the drive to produce computers capable of doing the intellectual tasks that now require the human mind.

Feigenbaum, a leader of the nation’s artificial intelligence community, says that such programs usually called "expert systems,” represent the future of the computer, and that artificial intelligence will come to dominate the field.

Schwartz, on the other hand, says that “artificial intelligence is possible--over the next couple of centuries. It will take that long to produce computers that can dwarf human intelligence,” Schwartz says.

What can be said thus far is that several expert systems have been developed and that two or three of them are in actual use. A program called DENDRAL works out the structure of organic molecules. Another, called R-1, designs components for customers of Digital Equipment Corp. A third, PUFF, diagnoses respiratory disease at a California hospital.

ENTHUSIASTS SAY THAT these expert systems will be followed by many more. The work of developing such expert systems, however, is not as easy as it seems.

In essence, an expert system is simply a codification of the knowledge and reasoning used by human experts. PUFF, for example, is built on the diagnosis methods used by a single physician, who was unusually good at explaining how he reached decisions. R-1 was developed by interviewing engineers who design computer systems.

Such interviews require what are called “knowledge engineers,” specialists who can coax the information out of people and put it into a form suitable for the computer. John McDermott, the computer expert at Carnegie-Mellon University who developed R-1, estimates that there are only 50 to 100 knowledge engineers in the United States.

Expert system development is also time-consuming. The working estimate is that it takes one hour for one knowledge engineer to extract one computer-real rule from an interview. An expert system contains hundreds or even thousands of rules, and the system must be refined constantly to fill in gaps that arise because most human experts haven’t thought about all the assumptions they use to make decisions.

FEIGENBAUM BELIEVES THAT the future of expert systems lies in taking the job of what is called “knowledge acquisition” from humans and giving it to computers. He envisions the time when a human expert will be "interviewed” by answering questions that are flashed on a computer screen. The programs to make that dream possible will take many years to develop, Feigenbaum acknowledges.

Even the Japanese program does not have knowledge acquisition as a goal. Its goals are many accomplishments far beyond current computer capabilities. If the Japanese succeed, computers of the 1990s will be able to translate from one language to another, respond to spoken commands, pick one speaker out of several analyzing speech samples, solve complex problems, and do much more.

Schwartz and Feigenbaum agree on one point. With Japanese programs, they say, American computer research is fragmented and uncoordinated. Unless Americans get organized, they say, the dominant supercomputers – and the thinking programs--of the future probably will be Japanese.