A reliability engineer might say that as a first approximation, the time-till-failure of the human animal (really just an electro-mechanical-chemical machine) has a governing equation which is a linear combination of many Weibull equations [0] and Arrhenius equations [1]. Mechanical failure is known to be governed by Weibull, and chemically-driven devices fail according to Arrhenius.

From this claim it isn't hard to cook up superexponential-ish results over certain timescales by tuning the fitting parameters and combination coefficients. But that doesn't mean the underlying failure physics here are truly superexponential.

But as a better approximation, human death may be governed by a system of differential equations with primarily stochastic coefficients and plenty of strongly nonlinear operators. So after our biomedical engineers remove the first few bottlenecks at ~105 years old, these curve shapes might change dramatically to reflect the true complexity of the underlying physics.

[0] http://reliabilityanalyticstoolkit.appspot.com/mechanical_re... [1] http://reliawiki.com/index.php/Arrhenius_Relationship