A recent article in the September issue of IEEE Spectrum is titled Why We Fall Apart. It analyses human aging using reliability theory, simplifying the same authors' 2001 paper in J. theor. Biol. The Reliability Theory of Aging and Longevity.
Reliability theory is pretty mundane stuff1, although for complex systems the calculations are complicated (hence the software tools). It's used by engineers to make predictive models of failure. Among other uses, this is how designers calculate the MTBF figures you see quoted in disk drive specifications.
The essential point of the article is this: you can model the human body as being comprised of a combination of irreplaceable, redundant, non-aging parts, some of which are defective to begin with. Doing so predicts the salient characteristics of human mortality:
- The infant mortality period, which, not ironically, is a term from both reliability theory and population dynamics.
- The normal working period.
- The aging period with exponentially increasing failure probability described by Gompertz 200 years ago. That is, after age 25 or-so, the probability that you'll survive another year declines very rapidly.2
- The post-aging or late-life mortality period with linear failure probability. That is, after age 95 or-so, your probability of surviving for another year is bad, but pretty much the same as it was the year before.
The post-aging characteristic of mortality statistics is related to convergence, the fact that an 80 year-old Indian has a similar life-expectancy to an 80 year-old Dane, although the life expectancies of Indians and Danes are quite different. This reliability theory model accounts for convergence and late-life mortality nicely.
A consequence of this explanation of aging is that one way to control it might be to avoid developmental damage in vitro. Don't skimp on those anti-oxidants, mothers-to-be!
 See an overly simple tutorial here.
 Actually, mortality increase during aging is not the same in living things as in redundant machines. Failure rates in such machines follow the Weibull distribution. But that's because classical reliability theory assumes that when machines are created, all their parts are working. If you assume that there's a high probability that components are faulty, you pretty much get a Gompertz distribution.