By Wilson da Silva
SYDNEY – Computers are too unreliable to monitor nuclear power stations or manage air traffic and should not be used in areas where lives could be at stake, says a new study.
Mishaps are so common that “computers should not be entrusted with life-critical applications”, said the study which examined computer and software failures over the past 20 years.
Australian scientists Tom Forester, of Griffith University in Queensland, and Perry Morrison, of the University of New England, cite 36 examples of computer failures worldwide that have killed people or caused huge damage.
Details of their findings were released ahead of scheduled publication in the June issue of the British academic journal Futures.
The scientists say computers are increasingly used for complex tasks on which many human lives depend – but there is no guarantee that computers will work or that their software is error-free.
Computer experts know there is no “adequate explanation why computers are so unreliable and thus why a computerised society is so vulnerable to computer failure”, the study says.
“Computers are inherently unreliable...prone to catastrophic failure and...their very complexity ensures that they cannot be thoroughly tested before use.”
Among the mishaps quoted:
– Failure of a computer-controlled radiation treatment machine in Texas which delivered fatal doses to two cancer patients.
– A single hyphen left out of a program for an Atlas-Aegena rocket carrying a probe to Venus caused it to tumble wildly and forced U.S. launch authorities to explode it in the air.
– 22 fatal crashes of the fly-by-computer UH-60 Blackhawk helicopter used by the U.S. Air Force.
– In July 1989, 104 failures in a single day of the Los Angeles air traffic control computer.
– A full-scale nuclear alert in the United States triggered by a faulty chip in a communications computer.
The study says programmers who write software for air traffic control systems and other “life-critical” functions should be licensed. Software manufacturers should guarantee their products, but few do.
The researchers say recent studies found that the best programmers can be more than 25 times as competent as the worst, and that many software supervisors do not understand their programmers’ work.