Science: A Star Is Born

In Arthur Clarke’s space odyssey 2001, the most intriguing character is not flesh and blood but an extraordinary computer called HAL (for Heuristically programmed ALgorithmic computer). Amazingly lifelike, HAL speaks with flawless diction, calmly disposes of almost any problem and—when it turns “psychotic”—does the same to most of the human crew. Could Clarke’s fantasy also be prophecy? Perhaps. In a chilling augury of the cybernetic future, scientists at Caltech’s Jet Propulsion Laboratory have now created a computer that in some ways approaches HAL’s versatility. In at least one vital respect it actually rivals its fictional counterpart: it can diagnose its own flaws and figure out ways to overcome them without any human help.

Billions of Miles. Called STAR (for Self-Testing And Repairing), the experimental electronic whiz is the prototype of a computer that is being designed for a most demanding assignment: guiding the first unmanned expeditions to Jupiter and the other outer planets later in the decade. These far-ranging planetary “grand tours” will carry spacecraft billions of miles from earth to the very edge of the solar system and take up to eleven years to complete. Unfortunately, even the most reliable of contemporary computers would almost surely suffer a failure of some of its components long before the end of such missions.

Nor could designers anticipate all the other hazards that computers might encounter in distant space—a burst of gamma radiation, the impact of a tiny meteor, an unexpected power loss. Even mission control would be powerless in such a crisis. If trouble occurred near Neptune, for example, radioed data describing the problem would take four hours to reach the earth. By that time, the moment for a crucial maneuver might well have passed, and the entire mission could be jeopardized.

With STAR, controllers should rest easy. Built by a Lithuanian-born J.P.L. engineer named Algirdas Avižienis, 38, the computer consists of ten separate units, each designed to perform a specific function (computation, logic, communications, memory and monitoring). More important, some of the units always stand by as spares. Thus, if any of STAR’S working parts should falter, it can quickly mobilize a replacement.

The really remarkable feature of Avižienis’ brainchild is a specially gifted unit dubbed TARP (for Test And Repair Processor). Like a zealous office manager always peering over the shoulders of his clerks, TARP can almost instantly spot errors, determine who has made them, and take steps to discipline or replace the wrongdoer. It constantly monitors the specially coded messages —or interoffice memos, as Avižienis calls them—that pass between the units, and immediately reacts to deviations from normal in the computer chatter. “It’s as if a person were to start mispronouncing or slurring words,” explains Avižienis. “Illness or intoxication would be immediately suspected.”

Majority Rule. Still, even if TARP suspects a malfunction, it does not react impetuously. Before calling in a standby, it will first give the balky unit a chance to redeem itself by letting it repeat the task. But what if TARP itself is having an off day? Adapting an idea first proposed by the late mathematician John von Neumann, Avižienis divided TARP’s brain into three independent lobes. If one lobe detects an error not subsequently confirmed by its two partners, the outvoted lobe will also be dismissed. Later, it may be given another chance. But if it continues to disagree with its partners, they will perform an electronic lobotomy on it and instruct a replacement to do the job. Explains Avižienis: “The majority kicks out the minority, brainwashes a spare with what it knows, and then continues.”

Like HAL, the self-repairing computer is not entirely foolproof. It could not, for example, cope with a flurry of failures that knocked out parts faster than they could be replaced. But in most test situations, STAR has proved a stellar performer. On one occasion, J.P.L. scientists taught it to play blackjack, then momentarily shut down its electrical power, hoping to unsettle the computer brain. No luck. STAR immediately revived, tersely acknowledged the disruption, resumed the game exactly where it had left off—and won.

In its present form, STAR occupies more than 100 cubic feet of space. For the “grand tours,” Avižienis hopes to compress it into two cubic feet and reduce its power needs to 50 watts—less than most ordinary light bulbs. Avižienis thinks that such a tiny, trusty brain also might be useful closer to earth: monitoring the guidance systems of supersonic aircraft, controlling high-speed trains, and even standing watch over the vital functions of seriously ill patients in hospital wards.