AI Emerges From the Shadows

By Stephen J . Shaw

from the November 1983 issue of Mini-Micro Systems magazine

Slowly but earnestly, companies in the artificial intelligence (AI) field are bringing out the tools to build the heralded Fifth Generation computer systems. It’s likely to take several years for any AI-based products to have a major impact in the commercial market. But 1983 could be the year AI emerges from its R&D cocoon.

Many significant product developments are coming from recently formed companies devoted exclusively to AI. More than 20 new enterprises have appeared during the past two years, notes Louis Robinson, editor of AI Reports and former executive director of the American Association for Artificial Intelligence (AAAI). Those 20 enterprises form the vanguard of the nascent AI industry. That industry is devoted to revamping the relationship between humans and computers by developing systems that can perform logical inferences, draw on the knowledge of experts in specific applications and be accessed in “natural” languages- that is, in terms familiar to human operators.

Such AI systems were displayed and described at a recent five-day AAAI conference. Testifying to AI’s broadening appeal, the conference attracted more than 2,000 participants, including 75 representatives of the general and trade press. More than 30 companies sponsored technical exhibits, and many more were represented on panels along with university and government research organizations.

There was a minor flurry of product releases at the conferencefrom both computer industry bulwarks such as Digital Equipment Corp. and Xerox Corp. and recent AI start-ups. But offerings aimed at further AI development applications outnumbered systems and programs for end-user applications. A broad range of strictly commercial AI products appears to be several years away.

“It’s hard to build commercial products in any relatively new field of science,” commented Larry Harris, president of AI Corp. “It’s especially hard in Al because of the long development cycle, two to five years from prototype to product.”

Harris told several hundred conference participants that more industry-specific products have just begun to be developed. He demonstrated AI Corp. ’s Intellect software package, a general-purpose natural-language program that allows database access by entering conversational English onto a keyboard. IBM Corp. announced in June that it will include Intellect in its Information Center product line.

Other AI hardware and software products featured at the conference include:

• The Lambda system from Lisp Machine Inc. (LMI) offering both UNIX and LISP in 32- or 40-bit architecture. Instruction execution on the 68000 processor, which contains the Berkeley UNIX native software environment, can reach 1 million instructions per second (MIPS) with cache memory and a 10-MHZ processing speed. LMI also announced that Texas Instruments Inc. had acquired a 25 percent interest in LMI. The Lambda uses Tl’s NuMachine.

• General E lectric Co. displayed a computer-aided troubleshooting system, the CATS-1 for locomotive-engine-repair applications. GE has placed the system in beta test at the Conrail and Burlington Northern railroads. The $40,000 system incorporates a PDP-11/23 processor.

• DEC showed off the XSEL expert system that is not yet available commercially, says a DEC spokes-woman. It has been used successfully in-house to assist sales personnel in defining system configurations for customers. The XSEL system reportedly has helped DEC reduce the number of systems that is shipped in improper configurations, requiring follow-up sales calls and reorders.

• Symbolics Inc. began shipping its 3600 symbolic-processing system in June to customers including Bell Laboratories, GE, Mitre Corp. and Fairchild Industries Inc. Symbolics president Russell Noftsker estimates that the company has reached sales of $3 million monthly, largely due to the 3600 system used as an R&D tool to develop AI expert systems.

• Another recent start-up, Inference Corp., introduced an inference engine, called ART, that can be combined with knowledge bases to construct a variety of expert systems. A license agreement has been reached with Control Data Corp. to use ART in CDC product development, states CDC company president Dr. Alexander Jacobson. Inference is attempting to reach a similar deal with DEC and is searching for a joint venture partner to develop an expert financial-system package.