Some managers still think that artificial intelligence--the decades-long effort to create computer systems with human-like smarts--has been a big flop. But executives at most companies on the BW50 list know better. Artificial intelligence (AI) is often a crucial ingredient in their stellar performance. In fact, AI is now a part of a swath of industries as broad as the BW50 itself. AI software helps engineers create better jet engines. In factories, it boosts productivity by monitoring equipment and signaling when preventive maintenance is needed. The Pentagon uses AI to coordinate its immense logistics operations. And in the pharmaceutical sector, it is used to gain new insights into the tremendous amount of data on the human genome.
Last year, for example, Abbott Laboratories Inc. (ABT
) opened a high-throughput lab for concocting novel drug candidates. James B. Summers, Abbott's advanced-technology honcho, boasts that the new lab's six researchers can do the work of nearly 200 scientists in Abbott's older chemical-synthesis facility. He credits this enormous speedup to various robotic and data-processing systems, most of which incorporate AI.
The finance industry is a real veteran in such technology. Banks, brokerages, and insurance companies have been relying on various AI tools for two decades. One variety, called a neural network, has become the standard for detecting credit-card fraud. Since 1992, neural nets have slashed such incidents by 70% or more for the likes of U.S. Bancorp (USB
) and Wachovia Bank (WB
). Now, even small credit unions are required to use the software in order to qualify for debit-card insurance from Credit Union National Assn.
Like banks, retailers collect huge amounts of data. Wal-Mart Stores Inc. (WMT
), for instance, harnesses AI to transform that raw data into useful information. Wal-Mart consolidates point-of-sale details from its 3,000 stores. Data-mining systems sift instantly through the deluge to uncover patterns and relationships that would elude an army of human searchers. Data-mining software typically includes neural nets, statistical analysis, and so-called expert systems with if-then rules that mimic the logic of human experts. The results enable Wal-Mart to predict sales of every product at each store with uncanny accuracy, translating into huge savings in inventories and maximum payoff from promotional spending.
Similar data-mining methods, augmented with AI software that can parse the text of electronic messages and translate foreign-language phone conversations, are being used by U.S. and allied intelligence agencies in the war on terror. Although the National Security Agency doesn't talk about its Echelon system for eavesdropping on all forms of telecom traffic, it snags far more data than Wal-Mart's intranet. Without AI, even the entire adult population of the U.S. couldn't begin to filter all the material NSA collects, which is why the Defense Advanced Research Projects Agency (DARPA) has long been a primary sponsor of research on AI. Echelon almost certainly helped nab al Qaeda's Khalid Sheik Mohammed in Pakistan, and NSA's data-sleuthing system also flagged advance warnings of the September 11th atrocity. Unfortunately, these alerts weren't screened by human analysts until after the fact.
Data-mining software that "understands" human language is now finding some unexpected applications in medicine. For example, a multidisciplinary collaboration led by researchers at Carnegie Mellon University is using natural-language processing methods to chart the "grammar" of various sequences of amino acids in proteins, then correlating the sequences with protein shapes and functions. Similarly, IBM (IBM
) is working with the Mayo Clinic to unearth now-hidden patterns in medical records. This approach has already had a significant success: Switzerland's Novartis (NVS
) used it to discover that infant leukemia has three distinct clusters, each of which would probably benefit from tailored treatments. Caroline A. Kovac, general manager of IBM Life Sciences, expects that mining the records of cancer patients for clustering patterns will turn up clues pointing the way to "tremendous strides in curing cancer."
Another data-mining thrust extracts information from images, including TV footage. Virage Inc. (VRGE
) offers a "motion-mining" system that integrates language-processing software from vendors such as BBN Technologies Inc., which has developed speech-recognition and translation tools for Arabic languages. At CNN, motion-mining helps human editors monitor raw news feeds from affiliates 24 hours a day. Virage's software also transforms video clips into digitized files that can be stored in searchable libraries.
DARPA's AI research typically sparks more civilian than military applications. That's certainly the case for DAML, or DARPA Agent Markup Language. This is an expanded version of the HTML (hypertext markup language) now used for writing Web pages. DAML generates Web data that can be processed by any software running on any brand of computer -- a sort of Rosetta stone for the Internet. Early DAML adopters have created Web sites where companies complete myriad transactions -- even automatically handling all supply-chain buying and selling -- assisted by AI software programs called intelligent agents.
Agents are also a chief focus of a wide-ranging project at IBM's Research Div., called the Agent Building & Learning Environment (ABLE). Along with Unisys (UIS
), Sun Microsystems (SUNW
), and Hewlett-Packard (HPQ
), IBM sees AI as the route to computers that automatically adapt to changing workloads and heal themselves after crashes. Today, complexity undercuts reliability, says Nagui Halim, head of distributed computing research at IBM. "Customers keep adding lots of different software pieces, not all of which are designed to work together." Smart, adaptive software would be able to oversee giant, complex systems that otherwise have no ability to adjust to spikes in demand.
Under ABLE, IBM's strategy is to build stronger, engineering-caliber versions of AI software "to make AI more ruggedized, more successful," says Halim. These will be used to create an integrated toolbox of AI methods, plus an expert-system manager for picking the right tool or combination of tools for each specific job. "We believe ABLE could have significant commercial impact in the near term as the nervous system helping to control the bones and muscles of conventional systems," Halim adds. That and all the other progress being made should give an overdue boost to AI's sometimes dismal reputation among managers.
MARCH 24, 2003
By Otis Port in New York, with Michael Arndt in Chicago and John Carey in Washington
Get BusinessWeek directly on your desktop with our RSS feeds.