The New Expert Systems

AI - left for dead years ago - lives on in the guise of packaged analytic applications

Arthur C. Clarke's Heuristic Algorithm Logician (HAL), of 2001: A Space Odyssey fame, is the fictionalized example of the ultimate decision-support application gone haywire. As part of their goal to concede mission-critical decision-making to software, HAL's human architects designed that software to create its own business rules as necessary. Unfortunately, HAL created a new rule during the mission to the effect that the humans tagging along were counterproductive.

Although we never did construct a "real" HAL, artificial intelligence (AI) luminary Marvin Minsky has said that many of the necessary technological ingredients are available. One such ingredient, the so-called expert system architecture, emerged in 1965 when Edward Feigenbaum created "knowledge engineering." His team's development of Dendral, a computer program that mimicked human decision-making by applying rules to a knowledge base painstakingly "extracted" from human domain experts, presaged a boom in commercial expert systems in the late 1980s. (Mrs. Fields' Cookies is the classic example of a company that used knowledge engineering to considerable business advantage during that time.)

The expert systems market died out in short order for various business reasons, but it's well appreciated that AI technology lives on in other guises, particularly in the sophisticated data analytics that drive high-end business intelligence solutions.

However, few people specifically appreciate the high-level similarities between classic expert system architecture and that of emerging packaged analytic applications. In fact, these apps are evolving toward an expert system-like model in which they will crystallize human domain knowledge for the purposes of automated business optimization. And when voice-recognition technology advances sufficiently (Microsoft is making huge investments here), who knows? HAL might begin to look pretty feasible.

What's Bred in the Bone

Both expert systems and analytic apps separate knowledge from process (apply sets of rules to data that resides elsewhere), support time-based analysis to make business forecasts (how many chocolate-chip cookies to bake or infant car seats to stockpile on a given day based on historical patterns, for example), and help business users focus on problem-solving rather than software implementation. (Thanks to IDC's Henry Morris for some of the definitions here.)

What's most important, however, is the origin of the business rules involved: In expert systems, they are derived from marathon interviews with human domain experts. In packaged analytic apps, they are increasingly based on "best practices" in the form of metrics. Thus, the vendors involved now look to differentiate themselves partly on the quality of their data models.

For example, Sagent Inc. likes to remark of its Insight for Retail Banking app that the product is "built by banking professionals," and Business Objects considered its 13,000-large customer base a great asset when developing its new Business Objects Analytics suite. And considering that Hyperion Solutions Corp.'s Balanced Scorecard implementation already supports custom benchmarking methodologies, it's not hard to imagine the eventual packaging of customer domain knowledge into verticalized modules.

No One is Replaceable

While writing this column, I couldn't help but imagine this interaction between David Bowman, a futuristic line-of-business manager, and his trusty Data Analysis Logician (DAL):

Rate This Article Rating: 1 (best) 2 3 4 5 (worst) Comments: Optional e-mail address:

DAL: Good morning, Dave. You're a bit late today. Was traffic bad?

Dave: Good morning, DAL (gulping a lukewarm vanilla latte after settling into his seat). The freeway was a mess.

DAL: I can imagine. Dave, there are new sales and inventory reports available for the Midwest region. Would you like to review them now?

Dave: Maybe later, DAL.

DAL: I think you should know that SKU XY3456E is in very low supply across outlets in central California. Would you like to issue a purchase order to a local supplier?

Dave: That sounds like a good idea, DAL, thanks.

DAL: Of course, Dave. And by the way, you're fired.