
                         Roanoke Times
                 Copyright (c) 1995, Landmark Communications, Inc.

DATE: SUNDAY, January 22, 1995                   TAG: 9501240027
SECTION: EDITORIAL                    PAGE: D-3   EDITION: METRO 
SOURCE: EDWARD TENNER
DATELINE:                                 LENGTH: Medium

TO ERR IS HUMAN (AND COMPUTER)

ONCE THERE seemed to be strength in inexactitude. Proctor & Gamble made hundreds of millions of dollars by claiming Ivory soap to be not 100 percent but 99 and 44/100 percent pure. Pure ``what?'' kids always asked. ``Pure soap,'' their mothers explained. Even Mussolini, Hitler and Stalin knew better than to claim 100 percent of the vote. It was always 98.7 or 95.3 or so.

Today nothing less than perfection will satisfy the information culture. We are not so tolerant of tolerances. Thousands of kids get pure 1600s on their college boards. And millions of consumers are getting replacement computer chips from Intel Corp. because a bug in the ``floating point unit'' of its flagship Pentium microprocessor is sometimes off by a wee fraction of a percent. The question appeared to be whether the error frequency matched the phases of the moon (rival IBM's position) or direct asteroid hits (Intel's).

After Intel announced details of its replacement offer, it became time for another look at this parable of precision. No doubt, the problem was real and potentially disastrous for some users. A fraction of a percent now and then won't unbalance a checkbook, but could blow up an options strategy. And it's not the sort of bug that locks up the computer or looks weird. It's potentially disastrous because it's infrequent and subtle. If Intel had disclosed the problem as soon as it discovered it, instead of letting mathematicians find out for themselves, it probably could have soothed most users, especially the text-and-graphics crowd.

Instead, the Pentium panic has turned up still other bugs. The industry magazine PC Week reports that the Windows 3.1 Calculator program has always given the answer 0.00 when users substracted 2.10 from 2.11, but nobody had noticed - until they started using the calculator to test for the Pentium bug. Intel has company.

The computer industry has only itself to blame. The legalese on software license agreements says that the product isn't necessarily good for anything; the sales pitch on the box says it can do everything. The whole point of computer marketing and computer interfaces is to induce a fantasy of perfect rationality. Computers turn out displays and text that look deceptively 100-percent right. That's why tens of millions are sold every year. Slide rules and even manual typewriters reflected human frailty. A hairline might be a trifle on this side or that; a letter could show a fatigued touch or an erased strikeover. But computer output just looks right and convinces people that they are right. A student who prints out a term paper in 500 dot-per-inch 12-point Postscript Palatino is unhappy settling for a ``B.'' Psychologists call this the illusion of control.

Business-school researchers have been studying how people actually use spreadsheets and discovering that users are perfectly capable of making big mistakes without help from the Pentium floating point unit. Experiments by Raymond R. Panko and Richard P. Halverson Jr. at the University of Hawaii suggest that while people are not bad at entering data in the cells of spreadsheets, a small error rate in entry can cascade into a large rate on the bottom line - up to 80 percent.

In real life, most spreadsheets probably have errors. Even if the commercial software has no bugs, the formulas that users construct often do. As software and home-brewed applications become more complex, there will be more rather than fewer mistakes.

If the process is shaky, data are even worse. They rarely are as accurate as a Pentium chip on one of its off days. The economist Oskar Morgenstern showed years ago that governments could not even keep very close accounts of closely monitored events like gold bullion transfers (``Economics is a two-digit science,'' Morgenstern once declared.) Do you really think your company's sales data - or the latest census results -are better?

So while you're waiting for that replacement chip (the Repentium, as wags are calling it), cool down and lighten up. Pentiums didn't bankrupt Orange County. They didn't pound the peso. And perfect or not, they won't psych out the Fed. The episode is a lesson not just for Intel but for all of us - 99 and 44/100 percent looks pretty good after all.

Edward Tenner, a visiting fellow at Princeton's Department of Geological and Geophysical Sciences, is finishing a book on unintended consequences of technology. He wrote this for Newsday.

- Los Angeles Times-Washington Post News Service

by CNB