[Hans Blom, 970717b]
(Bill Powers (970715.1010 MDT))
This post is mostly about "causes" and "explanations", but first
Such definitions are not hierarchical; they are essentially
circular. They are also not based on observations but on a theory.
Only if given the complete "theory" F = m * a, for instance, one
can define any of the three concepts in terms of the other two:
F = m * a
m = F / a
a = F / m
Algebraically this is true, but with respect to observations it is
not. It is possible to measure F and a independently, but not m.
I'm not quite sure what you mean here. How is something measured
"independently"? That can have two meanings, I guess. The first is
that we have bodily sensors that "measure" the variable in question
directly. Our skin, for instance, might have force sensors. The
second -- a more precise method -- would be by direct comparison,
using any of our senses, of a quantity with some known quantity of
that variable. Lengths, for instance, used to be visually compared to
the length of a certain rod in Paris.
Otherwise, all measurements are indirect. If you analyse what actual
measuring devices actually do to convert some physical quantity into
something that we humans can experience "directly", you'll find a lot
of conversions based on physical laws such as the above. The
principle is always the same, however: find an applicable law, fix
some of its terms to known constants, and if you can already measure
the remaining variables, you can "measure" (compute) the new one.
Take, for instance F = m * a. If we fix m, which is easy to do, and
if we can already measure a, we have a "measurement device" for F.
Note that in this device F is hierarchically higher than m and a, but
this might be different in a device that computes a based on
measurements of F and m.
But this is taking us to far into measurement theory for a discussion
here, I guess.
I think there are different kinds of statements that people use as
explanations. Those that refer to observables are disprovable; those
that invoke hypothetical entities are not. If I say your car stopped
because it's out of gas, you can look in the gas tank to see if my
explanation is believable.
This is the usual approach, yes. But it has its limitations, as we
have found out in several "intelligent alarms" applications, where
the goal is to find out, as precisely as possible, which "cause"
(e.g. a disconnect of the hose that ought to connect a patient to a
respirator) explains an abormal (multi-dimensional) measurement (e.g.
an abnormally low respiratory pressure and an abnormally high oxygen
concentration in the expired gas; the measurements are performed at
the ventilator side rather than the patient side of the circuit). To
do so, we essentially discover/learn a "perceptual input function"
which combines all primary measurements into one conclusion "signal"
for each possible problem, that indicates whether the problem
currently occurs or not.
This approach usually works fine, but it is based on the assumption
that _only one fault occurs at the same time_. That is often a
reasonable assumption, but it is not full-proof. In your car example,
for instance, although unlikely, there might be another coexisting
problem as well. I will find your explanation reasonable because
there is no gas in the tank indeed, yet when I provide the car with
gas it doesn't solve the problem. Is your "explanation" still
believable now? Yes and no. Yes, because you have given a partial
explanation. No, because your explanation is incomplete.
What we have found is that the more problems can coexist, the more
difficult it is to identify the actually occurring ones. We have also
found it a useful heuristic to identify only _one_ problem at a time.
Solve that problem, and if there is another one, it will then present
itself so that it can be solved in turn. Since solving a problem
takes valuable time and since a patient can survive a major problem
for only three minutes, it is essential to order the problems that
must be investigated according to their importance for survival. This
approach correlates with the human tendency to concentrate on one --
the most important -- goal at a time, which sometimes causes a lot of
irritation in others who believe that a different problem is far more
important and ought to be investigated first.
But if you say you robbed a bank because of a trauma that occurred
when you were young, there is no way to check to see if the trauma
occurred (that is, an effect inside of you, not the circumstances
that supposedly created it).
Why do we humans want to find/invent/construct/attribute a "cause" so
often? Because a cause identifies an intermediate variable that
results in -- or at least contributes to -- an outcome, a perceptual
result. If we can identify a cause _and if we can manipulate it_, we
will have discovered a way in which we can control better. Thus,
humans have a great desire to identify causes, and for a good reason:
doing so will most likely result in improved control.
Identifying a cause is not simple, however, because there usually are
a great many variables that contribute to a certain outcome. Anyone
of those may be identified as "the" cause. This shortsightedness is,
again, useful, because it leads to immediate action, which -- if this
variable is indeed an important one -- decreases the control error.
It is like hill-climbing, where one dimension is optimized in turn.
We know that this is a suboptimal method, but we also know that in
high-dimensional spaces it may be a lot faster (and easier to
construct an algorithm for!) than methods which optimize in all
dimensions simultaneously. It is also like constructing an input
function weight after weight, stabilizing one weight factor at a
It makes little sense, however, to identify "causes" that cannot be
manipulated, as in your second example. If someone tells you that
"the cause that I'm alive is the fact that the atmosphere contains a
sufficient amount of oxygen", you might want to replace "the cause"
by "a cause", but otherwise your spokesman is correct. But mentioning
such a cause is hardly meaningful, because it cannot be manipulated.
Similarly, once a youth trauma has occurred it cannot be undone
anymore; it's in the past. Yet, treating the patient may still lead
to him being able to cope better in the future and thus to prevent
reoccurrence of the problem.
Note that even in this case -- I do somewhat trust psychologists,
strange as it may seem -- one of the most important "causes" has
probably been found: an alternative, far more likely one (e.g.
youthful bravado and an unlucky "yes, I dare!" bet) would surely have
surfaced. In a situation where there are many simultaneous "causes",
discovering the most important one is quite a subjective and fallible
Anyway, discovering "causes" serves our being in control. Usually...