Realism and Artifact Kinds 199
the artifact precisely forms the specifi c connection between a certain input-output function
and a certain structure. For practical reasons related to generic use, we may be satisfi ed
with a general characterization of function, but for more specifi c uses—for example, when
we need to substitute a component—such a general functional characterization is no longer
adequate. We need to appeal to a fi ner-grain structural description of the artifact, and that
is what the author’s design provides.
The same criticism might be leveled in a different way. Someone could insist that we
are not entitled to say that a characterization including a specifi cation of a structure is
functional and thus that in the end I fail to provide an authentic functional characterization
of artifact kinds. These criticisms might be correct.
We can be driven by considerations concerning some expressions of use to regard physi-
cal structure as something extraneous to function, something belonging to the device that
performs the function and not to the function itself. We say that an object with structure
S performs function F, hence it seems that the function can be identifi ed independently of
S. But if for similar reasons we can say that a pen and a pencil have the same function
even if they perform it in a different way, it also seems perfectly correct to say that a pen
and a pencil have different functions of the same type; the same can be said of a bicycle
and a motorbike or a Ferrari and a Fiat 500. The fact that we can describe function at a
very high level of abstraction does not imply that descriptions mentioning both the input-
output information and a physical characterization of the structure cannot be regarded as
a more detailed characterization of the function itself. Thus criticisms of this type seem
more allied to terminological rather than substantial points. The idea is not new in the lit-
erature. Think, for example, of the famous approach proposed by David Marr in Vision
(1982). According to Marr, to understand a certain function or process, it is necessary to
analyze it at three different levels: the level of what and why, the level of representation
and algorithm, and fi nally the level of physical realization. According to him, the levels
of the what description—the input-output level—and the physical structure are of necessity
involved in the comprehension of the same phenomenon, that of the performing of a certain
function.
An item of a historical-structural, that is to say, nonfunctional kind, would be described
in words like “something that has been built with such and such a structure under certain
conditions.” I claim that the condition of the origin and the structure are not suffi cient for
the categorization of artifacts, which is why mere historical-structural kinds would not be
appropriate for the classifi cation of artifacts. As has just been observed, talking of function
does not involve the exclusion of considerations concerning structure. Instead it allows
one to choose the appropriate level of structural description; one can describe a function
at a very general level or at a less general level by introducing more details concerning
the structure. For this reason, the mere introduction of a reference to the structure in the
criteria for the classifi cation of artifacts is not suffi cient to conclude that these criteria are
not functional criteria.