108 
As an example, consider the fire at Browns Ferry Nuclear Plant on 
March 22, 1975. The design of the plant included redundant safety systems 
so that it appeared that the failure of all of them would be very unlikely. 
Nevertheless, the safety systems were all disrupted simultaneously by a fire 
that destroyed their electrical cables. A costly shutdown--hundreds of 
millions of dollars--of the plant has resulted. Disaster of catastrophic 
proportions was avoided only because equipment was available to cool the 
reactor that was not even part of the sophisticated safety system. The 
safety system failed totally. 
Another indication of the failure to take component interaction into 
account is visible in the uncritical acceptance of the linear hypothesis with 
respect to survivability of organisms. Here, it is assumed if X organisms 
out of N survive in a certain medium, then the probability of a single 
organism surviving is X-over-N, and usually N is very large, and X-over-N is 
very small. This is a dangerous assumption. Enfeebled organisms will be 
competing for scarce nutrients that may be available only in trace amounts, 
and may kill each other off. A few such organisms may have a much better 
chance of surviving at least long enough to exchange some of their DNA. 
Similarly, a large number of organisms might destroy some other orga- 
nisms that could act as their host, while a small number might survive 
as parasites. 
The weaknesses of oversimplified reliability theory is not a matter 
merely of academic speculation. Fault tree analysis, which is a simplistic 
form of reliability theory that is being used here, was used to predict the 
reliability of the main engine in the Apollo service module. It predicted 
a reliability of .9999, one failure in 10,000. The highest reliability 
actually achieved in tests was .96. The theory underestimated failures by a 
factor of one in 400. "The space program example showed approximately 20 
percent of the ground test failures, and over 35 percent of the in-flight 
malfunctions and failures were system malfunctions that were not identified 
as credible prior to their occurrence, despite a very thorough failure mode 
analysis made prior to the tests or in-flight trials." The citation is in 
my paper. 
The fact that fault tree analysis has not worked well for complex 
physical systems does not augur well for its application to biological 
systems. The application of a very unsophisticated version of reliability 
theory suggests a premature acceptance of mechanistic reductionism by those 
who use it. Probabilities are presented as if those parameters are definite, 
objective determined numbers. In fact, these probabilities are themselves 
random variables at best. It is important, therefore, to know what the 
confidence intervals of the assessed probabilities are. It is also important 
to appreciate the subjectivity of the assessments. To depend upon a failure 
estimate of 10 -n , we must determine a confidence interval on the order of 1 
minus 10 -n . What good does it do to say that the probability of disaster is 
only 10~ 25 if the chance that we are wrong is even five percent. 
[ 312 ] 
