Ghost in The Machine

The first sign that something had gone haywire in AT&T's long-distance telephone network came at 2:25 p.m. last Monday, when the giant map of the U.S. in the company's operations center in New Jersey began to light up like a football scoreboard. For reasons still being investigated, a computer in New York City had come to believe it was overloaded with calls, and it started to reject them. Alerted to New York's troubles, dozens of backup computers across the U.S. automatically switched in to take up the slack -- only to exhibit the same bizarre symptoms. People trying to place long-distance calls all over the world suddenly began to hear busy signals and recorded messages blandly informing them that "all circuits" were busy.

Thus began the worst computer breakdown in the history of the U.S. telephone system. The incident was also a vivid reminder of how susceptible America, and the world, has become to computer failures -- natural and man-made. In 20 years of intensive automation, everything from supermarkets to stock exchanges has been computerized. Last week businesses and consumers were forced to face up to a downside of technology that becomes apparent only when the new systems fail. Said Steven Idelman, chairman of Omaha-based Idelman Telemarketing: "When things go wrong in a computer environment, they go wrong in a big way."

Things stayed wrong at AT&T for nine hours last week. Of the 148 million long-distance and 800-number calls placed with the company that day, only 50% got through. Hotels lost bookings. Cars went unrented. The number of calls to the American Airlines reservation system fell two-thirds. Idelman had to send 800 phone workers home for the day; he estimates he lost about $75,000 in sales. All told, the breakdown cost AT&T some $60 million to $75 million in lost revenues. Said AT&T Chairman Robert Allen: "It was the worst nightmare I've had in 32 years in the business."

Phone-company technicians traced the problem to a single "failure of logic" in the computer programs that route calls through the AT&T network. Like many programming bugs, it stemmed from an improvement on the original system. By carrying information about who is calling whom on a separate channel, or band, from the call signal itself, AT&T has been able to reduce the time between dialing and ringing from as much as 20 seconds to as little as four seconds. But the refinement inadvertently made the system more prone to breakdowns. Last week's glitch spread rapidly among the 114 computers in AT&T's network in part because they all contained the same programming error.

The collapse of its network came at a time of increased vulnerability for , AT&T. Although Ma Bell still carries 70% of the U.S.'s long-distance traffic (down from 90% five years ago), it has been fighting a rearguard action to keep its customers from defecting to its feisty competitors, MCI and US Sprint. The glitch simultaneously deflated AT&T's multimillion-dollar "reliability" advertising campaign and handed its competitors a once-in- a-career sales pitch. "An important message to everyone whose telephone is the lifeline of their business," began a print ad rushed out by US Sprint after the breakdown. "Always have two lifelines."