Firms ‘Teach’ Computers to Get Smarter : Artificial Intelligence Reaches Point Where It Goes Commercial
About 20 touring Japanese businessmen crowded into a small meeting room in an office building on Wilshire Boulevard last week to hear how the intimidating logic of a Sherlock Holmes could be harnessed and put to work for the likes of Kawasaki Steel or Matsushita Electric.
Computer scientists from Silogic, a young Los Angeles firm, were using one of the fictional British detective’s more spectacular deductions (there was chalk on Watson’s left hand; therefore, he had decided not to invest in South African securities) to illustrate what their software can do. And never mind that their product speaks only English.
“We’ll have available a Japanese language interface early next year,” sales director John Weil assured the group.
Serious Business
Clearly, these people aren’t hawking vegetable slicers. This is the selling of artificial intelligence, and it’s getting to be serious business. The technology that endows computers with certain human-like mental powers has emerged from the laboratory and, for better or worse, entered the real world.
“There is a buck to be made in artificial intelligence now,” said Woody Bledsoe, a pioneer in the field and president of the American Assn. for Artificial Intelligence, an academic organization that hosted an uncharacteristically noisy and crowded biennial conference at UCLA last week.
The scientific meetings were overshadowed by a trade show where 60 exhibitors drew about 5,500 attendees--more than triple the number at the previous conference. The action spilled over into hospitality suites in the Beverly Hilton, Westwood Marquis and other hotels, where products could be privately shown, drinks served and deals cut.
“This conference used to be a sort of quiet, pastoral setting,” said Stephen Crocker, head of the computer science laboratory at the nonprofit Aerospace Corp. of El Segundo, a federal contract research center. He betrays some ambivalence about the commercialization.
“It’s the payoff after all those years of research,” Crocker said. “But you name it, and it’s been oversold. The most tantalizing claims are the least likely to be true. . . . Caveat emptor is good advice.”
Indeed, “artificial intelligence” is something of a misnomer when applied to today’s commercially available products. Critics say most are merely “expert systems” that are fed highly specialized knowledge and in turn can interpret an electrocardiogram or forecast an avalanche. Diagnosing, monitoring or interpreting from a limited amount of information hand-fed to a computer is a far cry from the level of reasoning powers being sought in the next generation of computers.
Personal-computer pioneer Alan Kay, who is now a “fellow” at Apple Computer, says today’s systems have about the same intelligence as a termite. And he bemoans the flight of academicians to fill a mushrooming demand for AI researchers, at the expense of pure scientific study in the field.
Doing Practical Things
“It’s getting hard to find any graduate students who are in graduate school anymore,” Kay said. “They’re all out doing practical things. Some of those things are so practical that they’re on a completely different pathway than what needs to be done.”
Moreover, AI will remain a tiny part of the computer industry for years to come. Even at an expected 45% annual growth rate from last year’s $450 million in revenues, AI would be a $4-billion market in 1990--less than 5% of what is expected to be a $150-billion computer industry.
But for now, AI is a bright spot in an otherwise depressed industry. Not only is there rapid growth in sales, but also a shrinkage in price: Computer systems designed for AI that used to cost $100,000 are dropping to the $30,000 range. Some software programs are now being written for personal computers, a boon to those who might not have access to more costly systems.
And, as Kay acknowledged at a panel discussion last week, “Termites can do marvelous things.”
Once considered a slightly arcane sub-category of computer science--”We march to a different drummer,” Bledsoe says--artificial intelligence has lately gained visibility because of its importance in the well-publicized international race to develop the fifth-generation computer. At the University of Texas, according to Bledsoe, fully half the students entering the computer-science graduate program are specializing in artificial intelligence.
Corporate Funding
While AI funding historically has come almost entirely from the Pentagon, the field gained new commercial credibility when NASA earlier this year found privately available AI technology feasible for use in the space shuttle. Venture capitalists are rushing in, and such industry heavyweights as IBM, Hewlett-Packard and Digital Equipment in the past few weeks announced their first AI products.
They join about 50 little companies that have appeared in the past few years to design artificial-intelligence software and to make computers specially designed for the technology. Many of the entrepreneurs came from the AI labs at Massachusetts Institute of Technology, Stanford, Carnegie-Mellon and other schools.
With such names as Artelligence, Expertelligence and Syntelligence, these start-up companies are rushing into a marketplace that, according to Hambrecht & Quist analyst Jeffry Canin, “is poised for explosive commercial growth.” It is a chaotic, acronym-crazed market with little standardization: The products “speak” such often-incompatible computer dialects as INTERLISP, LISP MACHINES LISP, FRANZLISP, MACLISP, NIL, IQLISP, VAX/LISP, T-LISP, LISP/VM and SYMBOLICS-LISP. (LISP is derived from List Processing.)
“It reminds me of the early Wild West days of personal computers,” said Sung Park, who spent last week demonstrating a sophisticated financial software program just introduced by Palladian Software of Cambridge, Mass., one of about two dozen AI software developers.
Response to Knowledge
Artificial intelligence is no easier to define than natural intelligence, but it is typically described as a science that empowers a computer to perceive, reason, plan, act and use language. Ultimately, scientists hope to create computer memories that can “learn,” but most of the current technology is based on the computer’s response to a body of knowledge fed into the system by human experts.
AI technology draws partly from the study of logic, as illustrated by the Sherlock Holmes example. Holmes’ creator, Arthur Conan Doyle, takes readers through a mental exercise. Holmes has inferred that Watson will not invest in some property in South Africa. Here’s how:
He notices chalk on Watson’s left hand and deduces that Watson has being playing billiards. Watson plays billiards only with an associate who, Holmes knew, had tried four weeks before to persuade Watson to invest with him in some property on which he had a one-month option. Since Watson’s checkbook remained locked in Holmes’ drawer, the detective reasoned that Watson had decided not to participate in the deal.
“How absurdly simple!” Watson exclaims.
Early AI research has already joined the mainstream of computer technology, embodied in such areas as computer graphics and networking. Thousands of physicians have office machines that read electrocardiograms, another modest example of artificial intelligence. The space shuttle will use an “expert system” developed by Los Angeles-based Inference Corp. to monitor its navigation console and make continual selections of data, a task normally performed by three humans watching more than 100 radar indicators on a display screen.
Broad Price Range
The commercial examples of current AI software defy generalization. Target Technologies, a Menlo Park, Calif., newsletter, identifies one AI software program from McGraw-Hill that costs $49.95. A more serious example would be Palladian’s $95,000 “financial adviser.”
Palladian, founded last year by entrepreneur Philip Cooper, former General Electric executive John Karcanes and several MIT professors in finance, management and artificial intelligence, calls its program a “thinking partner” for senior finance executives of big companies. It claims a backlog of $1 million in orders from 13 companies, including Coopers & Lybrand, McKinsey & Co., Cigna Corp., Ryder System, Champion Paper and the finance arm of Digital Equipment.
Drawing on knowledge fed into the program from about a dozen finance and computer experts, the system performs rapid financial analysis and flags financial projections that it considers questionable. In a financial plan for a new product, for instance, the program itself raises the question of how margins might be affected by unexpected off-shore competition.
The Palladian program runs on specially designed artificial-intelligence computers made by Symbolics, another MIT offshoot based in Cambridge, Mass., with a plant in Chatsworth, and by Texas Instruments, one of the established computer firms that have made a presence in the AI field. Such so-called dedicated AI computers--the other makers are Xerox and Los Angeles-based LISP Machine Inc.--represent the other major segment of the fledgling industry.
Expects Rapid Growth
So far, most of the machines are bought by the industry itself, government and academia. According to Hambrecht & Quist, Symbolics’ three biggest orders were from MIT, the Microelectronic & Computer Technology Research consortium in Austin, Tex., and Carnegie Group, another new AI start-up. But the Fortune 500 share of this business, now 30%, is expected to rise steadily. From revenues of $200 million this year, market researcher Gartner Group of Stamford, Conn., says the market for AI computers will reach $1 billion by 1990.
The machines are used in computer-aided design and manufacturing, diagnostics, robotics, vision inspection systems and other industrial functions.
F. Stephen Wyle, founder and vice chairman of 5-year-old LISP Machines, says that market will shift quickly to lower-price, desk-top computers that will soon account for half the units--but only 20% of the revenues. Wyle said LISP will stay above the fray with its costlier, more sophisticated systems, which it now sells “for the most part” to Fortune 500 firms such as Rockwell, Grumman, Gould, Westinghouse, Texaco, Exxon and General Dynamics.
Hambrecht & Quist expects that the number of AI computers, now about 1,500, will reach 75,000 in five years.
