actually know empirically
have the form of the traditional causal laws, except that they are not
to be regarded as universal or necessary. "Taking arsenic is followed by
death" is a good empirical generalization; it may have exceptions, but
they will be rare. As against the professedly exact laws of physics,
such empirical generalizations have the advantage that they deal with
observable phenomena. We cannot observe infinitesimals, whether in time
or space; we do not even know whether time and space are infinitely
divisible. Therefore rough empirical generalizations have a definite
place in science, in spite of not being exact of universal. They are the
data for more exact laws, and the grounds for believing that they are
USUALLY true are stronger than the grounds for believing that the more
exact laws are ALWAYS true.

Science starts, therefore, from generalizations of the form, "A is
usually followed by B." This is the nearest approach that can be made
to a causal law of the traditional sort. It may happen in any particular
instance that A is ALWAYS followed by B, but we cannot know this, since
we cannot foresee all the perfectly possible circumstances that might
make the sequence fail, or know that none of them will actually occur.
If, however, we know of a very large number of cases in which A is
followed by B, and few or none in which the sequence fails, we shall in
PRACTICE be justified in saying "A causes B," provided we do not attach
to the notion of cause any of the metaphysical superstitions that have
gathered about the word.

There is another point, besides lack of universality and necessity,
which it is important to realize as regards causes in the above sense,
and that is the lack of uniqueness. It is generally assumed that, given
any event, there is some one phenomenon which is THE cause of the event
in question. This seems to be a mere mistake. Cause, in the only
sense in which it can be practically applied, means "nearly invariable
antecedent." We cannot in practice obtain an antecedent which is QUITE
invariable, for this would require us to take account of the whole
universe, since something not taken account of may prevent the expected
effect. We cannot distinguish, among nearly invariable antecedents, one
as THE cause, and the others as merely its concomitants: the attempt to
do this depends upon a notion of cause which is derived from will, and
will (as we shall see later) is not at all the sort of