Basic HTML Version

There is another way in which function relates to robustness. Organisms are designed for
reliability, so that even with the genetic variations they undergo, or the environmental variations
they must endure and exploit, they still work. In general, the more important the function, the more
securely it is backed by redundant processes or alternative ways of achieving it. This is perhaps
the most obvious connection between robustness and organic design; organisms are generally
much better at this than human engineers.
Good and bad explanations
A good explanation ought to be something that is not too sensitive to detail and is potentially
generalizable. You do not need to get a general law out of it, but it should apply to at least some
range of other situations that are sufficiently like it. There are regularities in nature that are what I
call sloppy, gappy generalizations. These are endemic in the compositional sciences, where we are
trying to understand upper-level behavior in terms of the parts of which the system is composed.
It is generally the case that you do not get an exact match between the regularities at a lower level
and those at the upper level when you try to bring them into register. The match generally only
works for a limited range of conditions as specified at the lower level. For instance, the Brownian
motion affecting small but visible particles is produced by local imbalances in the lower-level
collisions, where they do not quite average out, as they would for larger particles. In addition,
Brownian motion shows scale dependence. An insectivorous songbird will see a more jagged
Brownian motion than we would because it is seeing things at a flicker fusion of 70 frames a
second, rather than at the human rate of about 24. That bird is able to track shorter zigzags in the
motion resulting from the local imbalance of bacteria-sized particles with molecular collisions. A
butterfly takes an apparently random flight path to make it harder for a bird to catch; and the bird
evolves the ability to sample the butterfly’s location more frequently so as to increase its chance
of catching it.
There are many kinds of bad explanations. If you have to meet a number of constraints to
make a good explanation – and here I have ignored many, many dimensions in that regard – the
corollary is that there are going to be many ways to fail. So, for example, if you insist on giving
an explanation in terms of the second derivative of a function and you are talking to a high school
algebra class, that is a bad explanation for that context, because you are presupposing calculus,
which they cannot yet understand. You are violating an entrenchment constraint by not providing
information that is a precondition to understanding your explanation, so it is the wrong explanation
for that audience. Here is another problem: if they disagree with some of your background
presumptions, you may not be able to get started. This is treating explanation in a different way
than I was before; I am exploring the issue of what is explanatory to a given audience, what is
likely to convince them. Thus, as Tversky and Kahneman (1974) showed, salience is important to
an explanation. For instance, people are far more worried about an airplane trip than a car trip,
even though the casualty rate for car mileage is more than 100 times as high as for air mileage. On