THE PRODUCTIVITY OF THOUGHT

CHAPTER 2. SOME HISTORY AND

PHILOSOPHY OF SCIENCE

One way in which scientific theories are confirmed by data is by finding

instances in which observations conform to a theory's law-like generalizations.
So, when Newton's theory along with initial and boundary conditions enables an
astronomer to accurately predict the successive positions of the planets, one has
some measure of confirmation for Newtonian gravitation theory. Such
articulation of theory in order to fit the available data is a paradigmatic scientific
activity. Yet, it appears that not all scientific reasoning works this way.
Sometimes scientific reasoning attempts to show that, even though two theories
are able to accommodate a given body of data, one theory can accommodate it
or explain it better than can its rival. Copernicans thought they had such
reasoning to provide against Ptolemaic astronomy. Charles Darwin thought he
had such reasoning against the creationism of his day. In these cases, the
Copernicans and the Darwinians thought they had at least a prima facie reason
to believe that their theory better explains a given body of data. While one
paradigmatic scientific activity is fitting theory to data, another paradigmatic
scientific activity is showing how one theory better fits a given body of facts
than does its rival.
A central contention of this book is that the systematicity arguments
involve this latter kind of scientific reasoning. While fitting theory to data is a
part of what the systematicity arguments are about, they are also essentially
committed to more than this. They involve some notion of a preferred kind of
account of the data. The initial motivation for this contention is the need to
make sense of various Classicist comments such as the following:

A Connectionist can certainly model a mental life in which, if you can

reason from P&Q&R to P, then you can also reason from P&Q to P .... But
notice that a Connectionist can equally model a mental life in which you get
one of these inferences and not the other (Fodor & Pylyshyn, 1988,
pp.47-48, italics in the original),

The point of the problem that systematicity poses for a Connectionist

account is not to show that systematic cognitive processes are possible
given the assumptions of a Connectionist architecture, but to explain how
 THE PRODUCTIVITY OF THOUGHT

systematicity could be necessary ... given those assumptions (Fodor &

McLaughlin, 1990, p. 202, italics in the original).

More recently, Fodor has added these very suggestive comments,

Human cognition exhibits a complex of closely related properties-including

systematicity, productivity, and compositionality-which a theory of
cognitive architecture ignores at its peril. If you are stuck with a theory
that denies that cognition has these properties, you are dead and gone. If
you are stuck with a theory that is compatible with cognition having these
properties but is unable to explain why it has them, you are, though
arguably still breathing, clearly in deep trouble (Fodor, 1996, p. 109).

These passages clearly indicate that the Classicists, at least at times, think that
something more is needed in the systematicity arguments than merely
accommodating the data, hence provide some initial reason to explore the
possibility that the systematicity arguments have a structure in common with
other explanatory arguments in the history of science. A second, and ultimately
more significant, reason to explore the foregoing possibility is that finding this
more demanding explanatory standard ultimately indicates a strength that
Classicism has against its rivals. This more significant reason will unfold over the
course of the next several chapters.
This chapter is the first step in what is meant to be a methodical
exploration of the kind of reasoning that underlies the systematicity arguments.
It will attempt to clarify the nature and significance of explanatory arguments
where scientists have maintained that one theory can explain a given body of
data better than can a rival. A reasonable way of doing this is through some
examples taken from the history of science. The first pair of examples are taken
from the history of Copemican and Ptolemaic positional astronomy. They involve
what are now textbook examples of the methodological advantages of
explanatory elegance over coverage of data. The second pair of examples is from
Charles Darwin's Origin of Species. In all these examples we see that there exists
an empirical regularity for which two theories provide accounts. One theory,
however, provides a better explanation for this regularity, hence giving us some
defeasible reason to prefer one theory over another. The examples will be
described in a preliminary way in sections 2.1 and 2.2 before a more extensive
analysis is advanced in section 2.3. In this more extensive analysis, we see ways
to refine what Fodor and Pylyshyn say these explanatory arguments demand. In
 THE PRODUCTIVITY OF THOUGHT

addition to examining these historical explanatory arguments, we shall make

some initial steps to relate them to the systematicity arguments.

COPERNICAN AND PTOLEMAIC ASTRONOMY

Quite early in the history of astronomy, it was observed that, in addition
to the multitude of stars that remained fixed relative to each other, there exists a
small number of bodies that wander through the stars.1* While it was not
possible to describe precisely the motions of these wanderers, the planets, some
gross regularities were quite evident. Among the simplest was the fact that, if
one traced the motions of Mercury, Venus, and the Sun through the fixed stars
over the course of several years, one would find that Mercury and Venus were
never found in opposition to the Sun. That is, one could never find the Sun at
one point in the sky and Mercury or Venus at another point separated by 180° of
arc. More specifically, Mercury could always be found within about 28° of arc
from the Sun and Venus always within about 45°. Put in other words, Mercury
and Venus displayed a limited elongation from the Sun. In this respect, Mercury
and Venus differed from the "superior" planets Mars, Jupiter, and Saturn, which
could be found in opposition to the Sun.

1. Cf., e.g., Matthews, (1994), Niklasson and van Gelder, (1994), van Gelder and Niklasson, (1994),
Cummins, (1996b).

A second significant qualitative regularity in the motions of the planets was the
fact that the superior planets always go through retrograde motion at
opposition. Retrograde motion refers to the change in direction of the motion of
a planet through the fixed stars. Although usually moving through the sky in a
west to east direction relative to the fixed stars, the superior planets periodically
slow in their eastward movement, begin to move westward for some time, then
again slow and resume their eastward movement. (See Figure 2.1.) Here, then, we
have two important astronomical regularities that must be explained: 1) Mercury
and Venus are never found in opposition to the Sun and 2) the superior planets
always go through retrograde motions while in opposition to the Sun.
 THE PRODUCTIVITY OF THOUGHT

Figure 2.1. Retrograde motion.

Both Copernican and Ptolemaic theories were able to provide accounts of
the motions of the planets that were in rough accord with the phenomena, albeit
in quite different ways. Consider, first, the accounts available in what is to us the
more familiar Copernican heliocentric system. On the heliocentric hypothesis,
the ordering of the planets in terms of their distances from the Sun can be fixed
as Mercury, Venus, Earth, Mars, Jupiter, and Saturn. Given this order, the two
phenomena we have alluded to fall out automatically, of necessity one might say.
By visual inspection of Figure 2.2, it is clear why Mercury can never appear in
opposition to the Sun. Given that they orbit the Sun with a radius smaller than
that of the Earth, the appearances must be as they are. The account of Venus is
essentially the same. Further, by visual inspection of Figure 2.2, one can see that
retrograde motions must be due to the faster moving Earth overtaking a slower
moving superior planet. The superior planets do not genuinely reverse their
direction of motion; they only appear to do so against the backdrop of the
supposedly fixed stars. Further, it is clear from this why retrograde motions of
the superior planets must always occur when the superior planets are in
opposition.
Like the Copernican system, the Ptolemaic geocentric system could
provide a qualitatively accurate account of the foregoing phenomena using the
basic construction of an epicycle attached to a deferent.2* (See Figure 2.3. The
line from the earth to D is the deferent. The line from D to the planet is the
epicycle.

2. The Copernican system also postulated epicycles and deferents, but these were not needed in
the explanations of the motions of Mercury and Venus relative to the Sun or the occurrence of
retrograde motion of superior planets during opposition.
 THE PRODUCTIVITY OF THOUGHT

The epicycles, deferents, and the sizes of the planets and the sun are not
drawn to scale.) In order to account for the fact that Mercury is never far from
the Sun,

Figure 2.2. The Copernican system.

Figure 2.3. A Ptolemaic model.

one supposes that the Sun lacks an epicycle and that the deferents of Mercury
and the Sun are collinear. The collinearity of the deferents accounts for the
proximity of Mercury and the Sun, while the motion of the epicycle accounts for
the deviation of Mercury from the Sun. An analogous account can be given for
Venus.
​ The Ptolemaic account of retrograde motions and oppositions is
somewhat more complicated. According to the Ptolemaic account, certain
combinations of the relative lengths and rates of rotation of the epicycle and
deferent produce retrograde motion. One can get a qualitative sense of what this
is like through examination of Figure 2.4. Given epicycles and deferents of
roughly the proportions shown in the figure, if one has an epicycle that rotates
very slowly relative to the rotation of the deferent, observers on the central
Earth will not observe retrograde motion. If, however, one has an epicycle that
rotates very rapidly relative to the rotation of the deferent, Earth-bound
observers will see numerous periods of retrograde motion. The very
phenomenon of retrograde motion, thus, involves some measure of parameter
fixing. One has to add yet other adjustments in order to have the superior
 THE PRODUCTIVITY OF THOUGHT

planets go through retrograde motions at opposition, namely, keeping the solar

deferent parallel to the epicycles of the superior planets and keeping the period
of the superior epicycles equal to one solar year. One can get a rough qualitative
sense of the way this works through Figure 2.4. For ease of illustration, the
period of the epicycle is one solar year and the period of the deferent is two
solar years.3*

3. Cf, Fodor & Pylyshyn, (1988), pp.17-28.

In 2.4a, the planet (gray) is in retrograde motion while in opposition to the Sun
(white). In 2.4e, the planet is in normal eastward motion in conjunction with the
Sun.

Figure 2.4. The Ptolemaic model of retrograde motion.

Finally, in 2.4i, the planet is again in retrograde motion in opposition to the Sun.
​ Here we can make some preliminary observations. While both Copernican
and Ptolemaic astronomy are able to offer accounts of the phenomena, the
Copernican account is superior. Think of the motions of Mercury, Venus, and the
Sun. The Copernican heliocentric system explains the fact that Mercury and
Venus never stray far from the Sun based on the way the two inferior planets
must appear from a planet having a larger orbit. By contrast, in the Ptolemaic
 THE PRODUCTIVITY OF THOUGHT

geocentric system, one must conjoin to the geocentric hypothesis the

hypothesis that the deferents of Mercury, Venus, and the Sun are collinear. The
idea, very roughly, is that the heliocentric hypothesis leads naturally, or of
necessity, to the phenomena with Mercury, Venus, and the Sun, where the
geocentric hypothesis requires the addition of a further arbitrary or ad hoc
hypothesis about the collinearity of deferents. Still speaking at an intuitive level,
the arbitrariness appears to lie in the fact that the Ptolemaic account can as
easily allow deferents to be collinear--as in the case of Mercury, Venus, or the
Sun--as allow them to move independently--as with Mars, Jupiter, and Saturn.
​ Think now about the motions of the superior planets. In the heliocentric
Copernican system, retrograde motions must occur at opposition, given the
proposed nature of retrograde motions. At an intuitive level, one might say that
it is a law of nature that the superior planets go through retrograde motions at
opposition and the Copernican heliocentric hypothesis necessitates this
phenomenon. By contrast, the problem with the Ptolemaic account of these
phenomena is that one must invoke geocentrism, along with the hypothesis that
the epicycles of the superior planets are parallel to the Sun-earth deferent
(along with still other hypotheses relating the relative lengths and speeds of the
superior epicycles and deferents). The hypothesis relating the epicycles of the
superior planets to the Sun-earth deferent, however, is in some sense an
arbitrary hypothesis of Ptolemaic astronomy. While the Ptolemaic machinery
can be arranged to allow this parallelism, it can also be arranged not to have this
parallelism.

DARWINIAN EVOLUTION AND CREATIONISM

Jumping to another historical period and switching scientific disciplines,
we can find many additional illustrations of what is apparently the same form of
argumentation in the Origin of Species. Until 1859, the orthodox scientific theory
about the nature of species was that each species had been specially created by
a "miraculous" act of God. In the Origin, however, Darwin championed the view
that new species were transmuted forms of pre-existing species. All living
species were the common descendants of one or a few original species. While
Darwin was able to convince the majority of his fellow biologists that species do,
in fact, evolve, he did not rely primarily on the fossil record. Indeed, the Origin
treated the fossil record as something of an embarrassment. Rather than relying
on the fossil record, Darwin repeatedly applied the strategy we have been
discussing to laws or regularities in biogeography, embryology, morphology, and
taxonomy. Here we will review two of Darwin's biogeographical arguments.
 THE PRODUCTIVITY OF THOUGHT

In Chapter V of the Origin, Darwin notes that if one examines the blind
animal forms in the deep limestone caves in the United States and in Europe,
one finds that the American blind cave forms resemble American sighted surface
forms, where the European blind cave forms resemble European sighted surface
forms. Here is a biogeographical empirical regularity: blind cave forms resemble
the sighted forms from the spatially adjacent surface. Why is there this
biogeographical regularity? What explains it? Darwin envisioned two accounts,
an evolutionary account and a creationist account. According to the evolutionist,
the patterns of resemblance between cave and surface forms is due to the fact
that the blind forms in American caves are transmuted descendants of the
sighted forms on the American surface, where the blind forms in the European
caves are transmuted descendants of the sighted forms on the European
surface, and descendants tend to resemble their ancestors. According to the
creationist, God created all life forms and chose to distribute animals on the
surface according to some plan whose pattern is only vaguely manifest in the
biogeographical data. The distribution of surface and cave forms is merely one
component of this overall plan. Which account is superior? Evidently, the
evolutionary account. The deep limestone caves in Europe and the United States
are about as similar environments as one could imagine, says Darwin. So, there is
reason to think that God could as easily have placed the same blind forms in
both the American and European caves, as he could have placed different blind
forms in the caves. Indeed, it would have beenjust as much within God's infinite
power to place the blind forms that resembled European sighted surface forms
in American caves and the blind forms that resembled the American sighted
surface forms in European caves. The problem for the Creationist account
appears to be that, even supposing God to have created the world, He could have
done what the creationist purports He did as easily as He could have done
something else. The creationist account, while it is able to save the phenomena,
after a fashion, does not save the phenomena in the right way. The creationist
theory has what appears to be an arbitrary assumption about it, namely, that
God chose to distribute life forms in one way, when He could as easily have
chosen to distribute them in another way. The evolutionary account does not
have this sort of arbitrariness about it. This gives us a defeasible reason to prefer
the evolutionary account to the creationist account.
As a second example, consider Darwin's account of the biogeography of
Batrachians, given in Chapter XII of the Origin:

Bory St. Vincent long ago remarked that Batrachians (frogs, toads, newts)
have never been found on any of the many islands with which the great
 THE PRODUCTIVITY OF THOUGHT

oceans are studded. I have taken pains to verify this assertion, and I have
found it strictly true. I have, however, been assured that a frog exists on the
mountains of the great island of New Zealand; but I suspect that this
exception (if the information be correct) may be explained through glacial
agency. This general absence of frogs, toads, and newts on so many oceanic
islands cannot be accounted for by their physical conditions; indeed it
seems that islands are peculiarly well fitted for these animals; for frogs have
been introduced into Madeira, the Azores, and Mauritius, and have
multiplied so as to become a nuisance. But as these animals and their
spawn are known to be immediately killed by sea-water, on my view we can
see that there would be great difficulty in their transportal across the sea,
and therefore why, on the theory of creation, they should not have been
created there, it would be very difficult to explain (Darwin, 1859, p. 393).

Here we have to do some reconstruction to see why Darwin thinks that

evolution provides a better explanation of batrachian biogeography than does
creation. According to the theory of evolution, batrachian forms first appeared
on the relatively old mainland, rather than the relatively young oceanic islands.
But, because sea-water kills batrachians and their spawn, thereby hindering
their migration across oceans, one finds that (almost without exception) there
are no batrachians on oceanic islands. Evolutionary theory has a story to tell
about batrachian biogeography, but so does creationism. Creationism can simply
maintain that God brought life to the planet and distributed it according to a
plan. The problem with the creationist account, even allowing for the existence
of God, is that it appears that God's plan could as easily have placed batrachians
on oceanic islands as not. The evidence for this latter claim is that naturalists
had already observed that it is possible for humans to transport batrachians to
Madeira, the Azores, and Mauritius and have them survive quite well. The
creationist can invoke some plan to try to explain the distribution of batrachians
across the surface of the earth, but the hypothesis of a plan is a kind of arbitrary
addition to the creationist hypothesis. Again, speaking intuitively and very
roughly, the idea in these explanatory arguments is that some regularity must
occur as a kind of natural consequence of a central hypothesis, not as the
product of a central hypothesis conjoined with an additional arbitrary
hypothesis.

WHAT THESE EXPLANATORY ARGUMENTS HAVE IN COMMON

In the foregoing two sections, the morals of the story were described in
rough and intuitive terms not chosen entirely at random. As we shall see, the
 THE PRODUCTIVITY OF THOUGHT

intuitive terms were of the sort that Fodor & Pylyshyn, (1988), at times, used to
describe what is involved in the systematicity arguments. What we would like to
do now is try to move beyond these rough and ready analyses to a more refined
analysis. At the outset, it is worth noting that drawing attention to interesting
and suggestive explanatory arguments from the history of science is one thing,
providing a philosophically adequate analysis of them is quite another. While the
examples might be on to something methodologically important, exactly what
constitutes this importance may be quite elusive. Thus, while it is reasonable to
predict that there will be philosophical problems with the analysis, it would be
rash to infer from that alone that there is nothing to the explanatory intuition
underlying these historical cases and the systematicity arguments. In other
words, a maximally compelling rejection of the significance of our arguments
taken from the history of science would not only explain what is wrong with the
analysis that is about to be offered, it would also indicate how the intuitions that
they involve are ill-founded. So, on to work.
As there are a number of potentially problematic dimensions to these
arguments, it is perhaps best to begin with some simple, less controversial
observations, observations that are not the less important for their simplicity
and obviousness. Clearly one of the most obvious features of these arguments is
that in each case we have two theories, both of which have some story or
account to give regarding the things to be explained (the explananda). The
theories in each debate-both heliocentrism and geocentrism, both evolution and
creationism-have stories to tell about why the putative astronomical and
biogeographic facts are the way they are. The arguments are not about fitting
the data or which theory saves the phenomena; they are not about which theory
is able to exhibit the desired regularities. Rather, they are about the ability of
one theory to provide a superior kind of account. In the astronomical cases, the
point is not that there are various facts about the motions of the planets that
Ptolemaic astronomy cannot accommodate. Some Copernican arguments--such
as the changing apparent brightness of Mars at opposition--are about a theory
being able to provide some account of the data, but the explanatory arguments
involving the limited elongation of Mercury and Venus and the retrograde
motions of the superior planets are not like that. In the biogeographical cases,
Darwin certainly knew that the creationist could call upon God's infinite power
to distribute life across the surface of the planet in any number of ways, so he
knew that the challenge he was issuing was not to accommodate the facts. The
challenge was to explain them in a desired way.
Consider, now, something about the elements of the foregoing arguments.
In each of our historical cases we have two theories which offer accounts of
 THE PRODUCTIVITY OF THOUGHT

putative regUlarities. In the astronomical cases, we have the regularity that

Venus and Mercury are never found in opposition to the Sun and the regularity
that the superior planets go through retrograde motion while in opposition to
the Sun. In the biogeographical cases, we have it that blind cave forms tend to
resemble the forms of the surrounding environment and that Batrachians are
not found on remote islands. What is the status of these generalizations? Are
they mere empirical generalizations or perhaps laws of nature? In various
developments of the systematicity arguments, it is held that one must explain
why systematicity is lawful or nomologically necessary.4*

4. The phrase "Kepler's laws" merely reflects common usage and should not be taken to indicate
that Kepler's laws are genuine nomological laws, rather than mere empirical generalizations.

No reason is given for this requirement nor is there any obvious reason for it.
Nor does anything in the foregoing examples motivate this requirement. In fact,
the foregoing examples suggest that it may be enough that the regularities be
mere empirical generalizations. Recall that the textbook examples of empirical
generalizations are Kepler's laws.5* The first Keplerian law states that planets
move in ellipses with the sun at one focus, the second states that the radius of a
planet sweeps out equal areas in equal times, while the third states that the
square of the period of a planet is proportional to the cube of its distance from
the sun. Each of these generalizations describes the motions of the planets or
some feature of the motions of the planets without any implications regarding
the causes of these motions. In the goodness of time, Newton's laws provided
the causal underpinning to these generalizations. Like Kepler's laws, the
generalizations concerning the limited elongation of Mercury and Venus, the
retrograde motion of superior planets at opposition, and the biogeography
ofBatrachians and cave forms appear to be mere empirical generalizations
awaiting some mechanism that would account for them. Geocentrism and
evolution provide these mechanisms. Thus, Classicist contentions
notwithstanding, there is reason to think that the explananda in these
arguments need not be lawful regularities, but may be mere empirical
generalizations.

5. Fodor and McLaughlin threaten to conflate these two issues in the following passage: The
point of the problem that systematicity poses for a Connectionist account is not to show that
systematic cognitive processes are possible given the assumptions of a Connectionist
architecture, but to explain how systematicity could be necessary--how it could be a law that
cognitive capacities are systematic--given those assumptions (Fodor & McLaughlin, 1990, p. 202).
 THE PRODUCTIVITY OF THOUGHT

Here the proposed lawlike status of the explanandum is not clearly distinguished from the
relationship between the explanandum and the explanans.

Having said something about the nature of the generalizations (the

explananda) in these sorts of arguments, we ask about the relationship between
the thing to be explained (the explananda) and the thing doing the explaining
(the explanans).6* We may begin with Fodor and McLaughlin's suggestion that
the explanans must necessitate the explanandum in these sorts of arguments.
The very first thing to observe is that the kind of necessity involved here is not
clear. The so-called "received view" of explanation in the philosophy of science,
the tradition of Hempel and Oppenheim's (1948) Deductive-Nomological (DN)
model of explanation, had it that explanations are a kind of deductive argument,
hence that the explanans logically entails the explanandum, hence that the
relation between explanans and explanandum is one of logical necessity. Yet,
one of the persistent problems of the Deductive-Nomological model was the
attempt to have logical relations, which the Logical Positivists took to be
relatively unproblematic, do duty for nomological relations that seem to underlie
genuine explanations. That is, the DN model was constantly threatened by
counterexamples in which a set of premises allowed one to deduce some
explanandum, but not thereby explain it.7* Given these problems, it appears that
we should examine the conjecture that, in these explanatory arguments, the
explanans nomologically necessitates, rather than logically necessitates, the
explanandum.

6. A nice account of this may be found in Ruben, (1990).

7. Recall Fodor & Pylyshyn, (1988), pp. 47-48.

If We look to our examples from positional astronomy, we may observe

that, given only the Copernican configuration of the sun, Mercury, Venus, and
earth, it does not nomologically follow that Mercury and Venus must never
appear very far from the sun. The Copernican system of hypotheses generates
the relevant appearances only given certain other auxiliary hypotheses, such as
the rectilinear propagation of light and the absence of obstructions that might
prevent the propagation of light from the sun to the earth. So, as things stand,
the explanans does not nomologically necessitate the explanandum. We may,
however, try to defend the Fodor-McLaughlin view that the explanans should
necessitate the explanandum if we take the foregoing historical examples to
present mere explanation sketches in Hempel's sense (Hempel, 1965, pp.
423-424). That is, we may suppose that the Copernican narrative mentioned only
some of the salient factors that will figure in a true, complete explanation.
 THE PRODUCTIVITY OF THOUGHT

Pursuing this line, we might, therefore, reformulate or perhaps flesh out, the
Fodor-McLaughlin idea of the explanans nomologically necessitating the
explanandum in terms of complete explanations: in a complete explanation (one
in which no relevant hypotheses are omitted from the explanans) the explanans
nomologically entails the explanandum. With this refinement in place, it appears
that, given a complete Copernican account, including the hypothesis that light is
propagated rectilinearly and that there are no occluding objects, it is
nomologically necessary that Mercury and Venus never appear very far from the
Sun. Further, given the rectilinear propagation of light and the absence of
occluding objects, the hypothesis that a period of retrograde motion consists of
the earth overtaking a slower-moving superior planet does nomologically entail
that retrograde motions will appear only at opposition. So, there is reason to
believe that, in a complete explanation of the kind under consideration here, the
explanans nomologically necessitates the explanandum.
But, what about the evolutionary cases? Insofar as biological evolution is
fundamentally a probabilistic process, the evolutionary cases seriously threaten
the present formulation of the Fodor-McLaughlin analysis. Even given the
assumption that Batrachians first appeared on the mainland and that saltwater
constitutes a barrier to their migration to remote islands, it is not nomologically
necessary that there will be no Batrachians on remote islands. A lack of
Batrachians on remote islands is at most rendered unlikely by the background
facts. It is possible, as Darwin himself suggests, that vigorous glacial action could
provide an effective means by which Batrachians might safely traverse the ocean
to remote islands. Similarly, given that sighted organisms live on the surface
near limestone caves, it is not nomologically necessary that they will migrate
into the caves and lose their eyes. The gradual reduction in the eyes is subject to
the probabilistic vicissitudes of the selection process. So, provided that the
evolutionary and astronomical cases involve arguments of the same form, it
appears that we cannot say that a complete explanans nomologically
necessitates its explanandum.
The evolutionary cases show that an explanans nomologically
necessitating or entailing the explanans is not itself a necessary features of these
arguments. The Ptolemaic explanations, however, further show that hypotheses
nomologically necessitating the explanandum is also not sufficient for the
desired form of explanation. If we think of the complete set of Ptolemaic
hypotheses relating to the motions of Mercury, Venus, and the Sun, it turns out
that they nomologically necessitate the explanandum. The same holds for the
set of Ptolemaic hypotheses relating to the retrograde motions of the superior
planets. Further, one might be tempted to argue that the Creationist's set of
 THE PRODUCTIVITY OF THOUGHT

hypotheses relating to God's creation oflife on earth nomologically necessitates

a pattern of distribution of blind cave forms and Batrachians.
If an explanans nomologically necessitating the explanandum is neither a
necessary nor a sufficient condition in these explanations, what else, or what
more, contributes to the presumed superiority of the Copernican and Darwinian
explanations? In our initial commentary on these cases, we suggested that the
Ptolemaic and Creationist accounts relied on arbitrary hypotheses. Perhaps
something can be made of this idea. It is, of course, not a problem that
explanations have recourse to auxiliary hypotheses for explanation and
confirmation. Perfectly good explanations quite freely make use of auxiliary
hypotheses. Perhaps the problem has to do with the auxiliary hypotheses being
in some sense arbitrary. One way to try to unpack this arbitrariness is to say that
arbitrary hypotheses are those that one could as easily take as leave, consistent
with one's central scientific hypothesis.8* In Ptolemaic astronomy, we might
identify geocentrism as the central hypothesis. Given this hypothesis, one can as
easily have the deferents of Mercury, Venus, and the Sun collinear as not.
Further, given the Ptolemaic hypothesis of geocentrism, one can as easily have
the superior planets go through retrograde motions at opposition as not.
Similarly, it appears that the problem with the creationist explanations of
biogeography is that, within the creationist framework, one hypothetical plan of
divine creation is just as good as another. The rough idea, then, is that the
preferred kind of explanation does not rely on arbitrary auxiliary hypotheses,
where by an arbitrary hypothesis we mean something like a hypothesis such that
both it and its negation are consistent with the central hypothesis of the theory.

8.For example, Dennett, (1991), van Gelder and Niklasson, (1994), Cummins, (1996b), and
Hadley & Hayward, (1997).

This initial analysis, inspired by comments from Fodor, et aI., requires

some refinement. The Copernican account of the apparent planetary motions
relies on the background assumption that light is propagated rectilinearly. But, a
heliocentric theory can as easily take as leave the hypothesis that light is
propagated rectilinearly. The Darwinian account of Batrachian biogeography
relies on the background assumption that saltwater constitutes a migration
barrier. But, of course, the theory of evolution is consistent with salt water being
a barrier to Batrachian migration and with it not being such a barrier. So, if we
are to sustain the intuition that Darwin and Copernicus were on to something
with their explanatory arguments, it cannot be that the arbitrariness of an
 THE PRODUCTIVITY OF THOUGHT

auxiliary hypothesis consists of both it and its negation being consistent with a
central theoretical hypothesis. Such a standard of arbitrariness would disqualify
even the Copernican and Darwinian accounts from counting as genuine
explanations.
There are, however, other features ofthe Copernican and Darwinian
auxiliary hypotheses that set them apart from their Ptolemaic and Creationist
counterparts. The Copernican and Darwinian auxiliary hypotheses were
confirmed independently of the explanations in which they were deployed,
where the Ptolemaic and Creationist hypotheses were not. Sixteenth century
scientists had reasons, independent of geocentrism, heliocentrism, and the
motions of the planets, for thinking that light travels in straight lines. Darwin
assumes that it is a (relatively) well-known or well-established fact that seawater
kills Batrachians and their spawn. By contrast, there were no independent
checks on the Ptolemaic hypothesis regarding the collinearity of the deferents of
Mercury, Venus, and the Sun. Nor were there independent checks on the plan of
God in creation. So, one reasonable conjecture is that independently confirmed
auxiliary hypotheses are acceptable for explanatory purposes, where auxiliary
hypotheses that are not independently confirmed are not acceptable.
Unfortunately, this analysis has a subtle weakness. Were this the source of
the superiority of one account over the other, one might suppose that the
Ptolemaic and Creationist accounts were simply not as strong or compelling as
they might be. They would be better if each of the hypotheses invoked were
independently confirmed. In other words, all things being equal, the more
evidence an explanatory argument has for each of its constituent hypotheses the
better the explanation. The superiority of one account over the other would,
thus, be a matter of degree. This, however, seems not to be the nature of the
dissatisfaction found in the Ptolemaic and Copernican case. The common take
on these historical episodes is that one account constitutes an explanation,
where the other does not--period. In fact, examination of these cases often
provokes the much stronger claim that Ptolemaic astronomy cannot explain why
Venus and Mercury are never found in opposition to the Sun and that Ptolemaic
astronomy cannot explain why retrograde motions occur only at opposition.
Consideration of biogeography often provokes the charge that Creationism
cannot explain the facts. This sort of commentary suggests that more is going on
than that the accounts are not as well-grounded as they might be. It suggests a
deeper dissatisfaction with the accounts.
In order to capture this deeper dissatisfaction, we might venture the
following: the problem is that the auxiliary hypotheses are arbitrary in the sense
that one cannot confirm the auxiliary short of establishing the truth of the
 THE PRODUCTIVITY OF THOUGHT

theory. In the Ptolemaic case, there is no way to confirm the hypothesis that the
deferents of Mercury, Venus, and the Sun are collinear, short of confirming the
Ptolemaic geocentric system. Perhaps more obviously, there is no way to
confirm the nature of God's plan in distributing life across the surface of the
globe, short of confirming the Creationist theory that God in fact had a plan in
creation. Put in other words, the various arbitrary auxiliary hypotheses are
arbitrary in the sense that they can state whatever needs to be stated in order to
get the explananda to come out as they should. They are to this extent ad hoc,
"free parameters".
So, in these inference to the best explanation arguments about empirical
generalizations, it appears that the principle that favors one explanation over
another is that an explanation must not rely on hypotheses that can only be
confirmed given the truth of the theory. If this is the correct analysis of these
arguments, then it makes some sense of these explanatory showdowns. If the
entire aim of these sorts of showdowns is to confirm one theory against its
rivals, then the fact that one theory cannot provide this sort of confirmation,
without relying on a hypotheses that presupposes the independent confirmation
has to seem deeply problematic. This seems to capture the sense in which
arbitrary or ad hoc hypotheses are methodologically suspect.
To this point, we have talked of central and auxiliary hypotheses without
having offered any characterization of them. This terminology is, of course,
common in the philosophy of science, capturing to some extent the pragmatics
of science. The central hypotheses are those that are of most pressing interest
to a scientist, while the auxiliaries are oflesser concern. The central hypotheses
are those of such importance that they bear names, such as evolution,
creationism, heliocentrism, geocentrism, uniformitarianism, and corpuscularian
ism, where auxiliary hypotheses are typically specified by descriptions. Here it
should be noted that the distinction between central and auxiliary hypotheses is
required to do no philosophical work, hence that its invocation is innocuous.
What we have found to be problematic in Ptolemaic and Creationist cases is
some feature of the set of hypotheses. The set of hypotheses is problematic
insofar as one (or more?) of them might be varied to fit the explanandum without
independent check. The problem that some theories have in explaining certain
regularities is that they must rely on free parameters.
We have just said that what matters in separating the "good" explanatory
accounts from the "bad" is something about the set of hypotheses invoked in the
explanation. In particular, it is something about one set of hypotheses relying on
a hypothesis that is in some sense arbitrary or ad hoc. It does not matter how we
identify particular hypotheses as either central or auxiliary. In this same vein, we
 THE PRODUCTIVITY OF THOUGHT

should add that it does not matter how we name groups of hypotheses. It does
not matter whether by the name "Ptolemaic astronomy" we mean simply the
hypothesis of geocentrism or some more extensive set including the hypothesis
that there are epicycles and deferents or some still more extensive set including
hypotheses about the particular position, lengths, and rates of rotations of each
epicycle and deferent. The same holds for use of the name "creationism." A given
set of hypotheses has or lacks whatever explanatory power it has, quite apart
from how they are grouped under various names. While obvious at this point, it
may be less than obvious in the later, more contentious, context of systematicity
arguments.
The last few paragraphs have offered a range of reasons for analyzing
particular episodes in the history of science in a particular manner. The proposal
is that there are various empirical regularities that call for accounts in which a
central theoretical hypothesis, in conjunction with confirmed auxiliary
hypotheses, bring about the desired regularity. Furthermore, the account may
not rely on hypotheses that cannot be confirmed short of confirming the
hypothesis under test. The ability of a theory to provide explanations meeting
these conditions seems to be an important indicator ofthe truth. This is not to
say that the ability guarantees the truth of the theory; only that the ability is a
defeasible bit of evidence for the truth of the theory. This feature bears
emphasis. In his discussion of systematicity, Robert Cumm ins, (1996b), describes
the explanatory standard we have in mind in terms of explanations that are
"principled" versus those that are not. He, then, has this to say about principled
explanations:

Being principled is a virtue in explanation. But the virtue is purely

methodological. Principled explanations are easier to test, but they are no
more likely to be true. If there are independent reasons for preferring a
connectionist architecture (as clearly there are), then the methodological
weakness of the ensuing explanations of our sensitivity to systematicity in
language must simply be swallowed. If you are convinced that the mind is a
network, you should not be dismayed by the fact that your explanation of
our sensitivity to linguistic systematicity is not as easily tested as the other
guy's (Cummins, 1996b, p. 608).

If, in fact, the systematicity arguments share a form with the arguments we have
just reviewed, then the examples have something to say about Cummins's
analysis. Cummins claims that "principled" explanations are more easily tested,
but they are no more likely to make their background theories true. The idea
 THE PRODUCTIVITY OF THOUGHT

about ease of testability is an interesting idea, perhaps correct, and surely

worthy of further exploration. Yet, the examples from biogeography and
positional astronomy do suggest that explanations ofthe sort that Darwin gave
provide us with good reasons to think that evolution actually took place.
Explanations of the sort that Copernicus could have given, give us reason to
think that heliocentrism is really true. Cummins is, of course, correct in pointing
out that explanatory arguments are not the only sort of evidence that may be
brought to bear in a scientific dispute and that other evidence may lead to an
alternative assessment of rival theories. He is right insofar as he is reminding us
that theory choice must be made on the basis of all the available evidence.
Perhaps Connectionists, for example, can give us good reason to believe in their
theory of cognition, even if Connectionism does not explain the systematicity of
thought. One might also suggest that theories can have methodological virtues
other than those displayed in the kinds of arguments we are here exploring.
These are points to which we shall return. Here, however, we may note that
these points only show what should have been evident all along, namely, that
these "principled" explanatory arguments are merely defeasible indicators of the
truth. Observations about defeasibility, however, do not show that "principled"
explanations are to be preferred only for methodological reasons. From the
historical cases surveyed so far, it appears that "principled" explanations are
guides to the truth.
​ To emphasize the defeasibility of this form of argument, we might note a
curious example where the argumentative strategy lead to a false theory. An 18th
Century evolutionist, Demaillet, apparently asked why it is that

To every species of parrot, whose plumage is the most diverse, the rarest
and the most singularly marked of all the birds, there corresponds a fish
colored like them, in black, brown, grey, yellow, green, red, violet, gold, and
blue, and these markings appear on the same parts ofthe fish as they do in
such a bizarre manner on the plumage ofthese very birds (cited by Richard
Owen in his review of Darwin's Origin., reprinted in Hull, 1973, p. 199).

Demaillet's implication was that the birds had evolved rather directly from the
fish, which provides a neat diachronic explanation of the regularity. Demaillet's
argument can go wrong, if it turns out that, in fact, there is no bona fide
icthyo-ornithological regularity of the sort that Demaillet offers. As we shall see,
a number of critics of the systematicity arguments argue that there is no
bonafide explanandum involved in these arguments.9* Yet another moral is that
we should allow for the possibility of even true regularities turning out to be
 THE PRODUCTIVITY OF THOUGHT

mere coincidences. Be all of this as it may, the final conclusion to draw seems to
be that there is only some presumption in favor of the need to provide the
preferred style of explanation for a given regularity. An unexplained regularity is
a kind of anomaly for a theory. A theory may survive an anomaly, but it must
recognize them, especially anomalies of the form that have helped topple well
entrenched theories, such as Ptolemaic astronomy and creationism.
​ Having sketched a proposal for a structure to these explanations, it is
important to observe that not all scientific explanations in good standing have
this structure. 10* Four kinds of cases come to mind. A first and obvious
illustration is the explanation of particular events, such as the crash of the
Hindenberg or the extinction of the dinosaurs. In the arguments that have been
at the focus of this chapter, the explanandum has been supposed to be some
empirical generalization and empirical generalizations are not particular events.
So, explanations of particular events may not fit this model. In addition, there
are instances of structural facts whose explanations are properly explained by
recourse to a thing's substructure. Why does water expand when it freezes? This
has to do with the structure of the water molecule and the arrangement of its
molecules into a crystal. Molecules in the liquid water are able to move about
freely and independently which allows them to move closer to each other than is
possible when the molecules must assume the fixed arrangement of the solid
crystalline form. A third class of explanations that appears not to fit this model
are explanations of particular processes, such as how proteins are made.

9. In defense of a model of certain systematic relations in thought, Hadley, (\ 997), sketches

cases of explanations that he supposes do not have this structure.
10. For the details of how these conclusions are reached experimentally through a habituation
paradigm, the reader is referred to Karmiloff-Smith, (1992), pp. 67-72.

The story in this case is that DNA, in conjunction with certain nucleotides and
enzymes, makes RNA which, in conjunction with amino acids and other
enzymes, give rise to the proteins. A fourth class involves explanations of what
things are, for example, what the Krebs cycle is, what sunspots are, and what
structure of DNA is. These explanations, as well, seem not to have the form of
the arguments from biogeography and positional astronomy. There is no reason
to think these four cases constitute a mutually exclusive and jointly exhaustive
taxonomy of possible scientific explanations. They almost certainly do not. The
point here is simply that there are multiple reasons for thinking that not all
scientific explanations must be like those found in the Copernican and
Darwinian examples we have reviewed.
 THE PRODUCTIVITY OF THOUGHT

​ The existence of legitimate scientific explanations that do not fit this

pattern does not, of course, diminish the importance of the kind of explanation
we have seen in astronomy and biogeography. Nor do they show that the
particular examples we have seen have been incorrectly analyzed. They do,
however, raise other sorts of difficult questions. To begin with, it is unclear
exactly how we are to categorize explanations and explanatory demands. This is,
to some extent, an unavoidable consequence of the lack of a sound general
theory of explanation. If we are to have a philosophically satisfactory
classification of explanations, it must divide them into theoretically interesting
classes. Yet, in the absence of a satisfactory general theory of explanations, we
don't know what the theoretically interesting classes are. Another significant
question raised by explanations that do not fit our pattern is when the demand
for our preferred type of explanation must be met. If not all explanations must fit
this pattern, which explanations must?
​ Alas, no answer to these questions will be forthcoming here. Nevertheless,
it would be rash to dismiss the scientific challenge the Copernican and
Darwinian explanatory arguments pose. Good judgement would suggest that the
Copernican and Darwinian arguments ought not be ignored simply for lack of a
philosophical theory to ground them. Science should not have to wait for a
philosophical seal of approval on its reasoning. It appears that whatever is going
on in the Copernican and Darwinian arguments, it is an important indicator of
truth. We should take as relatively fixed points in our debate the idea that the
Copernican and Darwinian theories are on to something that philosophers
would do well to characterize, in spite of local failures to characterize them
properly. So, as far as cognitive science is concerned, we need to attend to what
the Copernican and Darwinian cases have to tell us.

SOME BROADER IMPLICATIONS OF OUR EXPLANATORY

STANDARDS
Our principal concern with the history of science has been to illuminate
the structure of the systematicity arguments. Yet, the explanatory standard
found in these episodes would be worth reflecting upon, even were we not to be
concerned with the systematicity arguments. There are other regularities in
cognitive science that might also benefit from scrutiny in the sorts of terms we
have found operative in the history of science. To close out this chapter in
support of this contention, this section digresses to review two contemporary
and completely unrelated areas in which explanatory standards make a
 THE PRODUCTIVITY OF THOUGHT

difference. For those only interested in the Classicism/Connectionism debate,

this section can be skipped without loss.
Developmental psychologists often recognize domains of cognitive
development, such as language, mathematics, and physics. Annette
Karmiloff-Smith, (1992), however, postulates, in addition, microdomains. As the
name would suggest, a microdomain covers a part of some domain. Thus,
KarmiloffSmith suggests that, within the domain of language acquisition, we
might recognize three microdomains corresponding to noun, verb, and pronoun
acquisition. Karmiloff-Smith also proposes that development might, or might
not, take place at different rates in different microdomains.
By hypothesizing multiple, independently developing microdomains,
Karmiloff-Smith has more parameters to use in fitting available developmental
data. Microdomains can be isolated and timed to fit many, many patterns. Yet,
the potential weakness in this flexibility in data coverage looks to be comparable
to Ptolemaic astronomy's weakness in its flexibility in fitting the data of
planetary positions. Judicious selection of microdomains and the freedom to
vary independently the time course of development in diverse microdomains
leaves one with auxiliary hypotheses with enough flexibility to allow the theory
to fit essentially any pattern of development without thereby providing a
genuine explanation.
Karmiloff-Smith's discussion of the developing child as physicist bears out
this point. If a 4-month old infant visually observes the joint and continuous
movement of parts of an occluded object, this suggests to the infant the
presence of a single object. If a 4-month old cannot see the joint and continuous
movement of an object, but can use two hands to feel two parts of an unseen
thing move together continuously, this also suggests to the infant the presence
of a single object. II Here is a fact to be explained: Why is it that continuous
movement in both the visual and tactile modalities leads a 4-month old child to
postulate a single object? One possible answer is that the ability to use
continuous motion in the visual modality and the ability to use continuous
motion in the tactile modality are two microdomains that arrive at the same
developmental stage at the same time. Another possible answer is that the ability
to use continuous motion is one domain, so it must happen that the ability to
use it in one modality goes with the ability to use it in another. Either way,
however, it seems that Karmiloff-Smith's theory must rely on auxiliary
hypotheses about the individuation of microdomains and (perhaps) the timing of
development of microdomains, auxiliary hypotheses that seem to admit of no
independent confirmation. Here we have a puzzle worthy of further
examination.
 THE PRODUCTIVITY OF THOUGHT

Not to lay the burden for meeting this explanatory standard too heavily
with more experimentally oriented cognitive psychologists, a second example
challenges Classicism with a homely folk psychological hypothesis. Folk
psychology appears to have it that, as a matter of nomological fact, actions
require an agent to have both a belief and a desire. So, suppose that an agent
believes that there is water in the refrigerator. If the agent has no desires, folk
psychology has it that the agent will perform no action. By the same token,
suppose that the agent desires a drink of water. Without any beliefs about how
the agent might go about getting a drink, the agent will again perform no action.
So, here is some regularity to be explained: in normal cognitive agents, actions
require both beliefs and desires. A viable form of this requirement, if there really
is one, might well add something about the relationship between the content of
the belief and the desire somehow being related, so that a belief about water
being in the refrigerator and a desire for a cup of coffee will produce no action.
But, for present purposes we may forbear this refinement. All we care about is
why both beliefs and desires are required for action.
Perhaps such a folk psychological regularity will be denied, but if not, we
may ask how, for example, Classicism might explain such a regularity. To begin
with, we recall that, according to Classicism, believing that p is a matter of
standing in some sort of computational relation to a mental representation r that
means that p and that desiring that q is a matter of standing in some sort of
computational relation to a mental representation that means that q. The natural
thing to do, then, is to build upon the metaphor of a "belief box" and a "desire
box". This metaphor might be cashed out in terms of something like a two-tape
Turing machine. One tape has the mental representations that represent the
objects of belief, while the other tape has representations that represent the
objects of desire. The read-write head in our hypothetical two-tape Turing
machine will simultaneously scan both tapes and will process information only if
the head scans something on both tapes. So, rather than having Turing machine
instructions, such as "S0 1 0 S2" (which are interpreted to mean "if in state So
scanning a 1, write a 0 and go into state S2"), our new two-tape Turing machine
will have instructions such as "S0 0 1 0 0 S2" (which are interpreted to mean "if in
state So scanning a 0 on the first tape and a 1 on the second tape, then write a 0
on the first tape, a 0 on the second tape, and go into state S2"). In a twotape
Turing machine of the sort just described, there will be no processing, no
cognition, no action, if there are not representations on both tapes. If either
beliefs are totally lacking or desires are totally lacking, then there will be no
action. This conforms to the explanandum before us.
 THE PRODUCTIVITY OF THOUGHT

Having an account, of course, is not all there is to the explanatory

standard we have surveyed. We need to consider whether or not we have had
recourse to an arbitrary hypothesis and the worry is that we have. Given the
hypothesis that propositional attitudes are matters of standing in computational
relations to mental representations, we can as easily have it be the case that
processing depends on both beliefs and desires as that it depends on beliefs
alone or desires alone. That is, we have already seen how we might have a
computational theory of mind realized in a system in which both beliefs and
desires are necessary for processing, but, of course, the standard case involves
the equivalent of only one Turing machine tape. That is, Turing machines
compute given only very localized information about what is on a tape. When a
Turing machine computes simple arithmetic functions, there are only markers
on the tape. It need not be the case that there is something corresponding to a
belief and something corresponding to a desire in order for processing to take
place. Now, it is, of course, true that "belief boxes" and "desire boxes" can be
realized on a single Turing machine tape. That is beside the point. The problem
is that there need not be any functional realization of both a belief and a desire
in order to get a Turing machine to compute. Further, what holds for Turing
machines holds for other Turing-equivalent computing devices. So, given
Classicism as a central hypothesis, it appears that only by the addition of a
problematic auxiliary hypothesis--one that can be tested only given the truth of
the theory under question--will we generate the desired empirical
generalization about the need for both beliefs and desires. If this is so, however,
then we don't have the sort of explanation we have seen to be so important in a
range of scientific episodes.

TAKING STOCK
The present chapter has an important moral. It is that there is good
reason to believe that not all ways of saving the phenomena are equally good. At
some level, both Copernicus and Ptolemy were able to save the phenomena
regarding the motions of the planets. At some level, both creationism and
evolution were able to save the phenomena regarding the distribution of life
forms across the surface of the planet, the nature and distribution of vestigial
organs, and the commonalities in morphological structures. Yet, at a deeper
level, the Copernican and Darwinian theories appear to provide superior
explanations of commonly accepted empirical regularities. The intuition is,
roughly, that the accounts provided by some theories do not amount to genuine
explanations because they are ad hoc or rely on arbitrary hypotheses.
 THE PRODUCTIVITY OF THOUGHT

The history of science suggests that there are explanations, then there are
explanations. The history of science tells us that there is something of epistemic
interest going on in these historical episodes. What it doesn't tell us is just how
to characterize what is going on in these episodes. Providing such a
characterization is a job for the philosopher of science. The core of the analysis
proposed here is that, in these cases, there are empirical generalizations for
which we want some explanation. An explanation will involve a single "central"
hypothesis that may, in itself, lack any confirmation other than that which it
might obtain via its role in the explanation at hand. In addition, the single
"central" hypothesis may have recourse to any number of additional,
independently confirmed "auxiliary" hypotheses. If a theory is able to provide an
explanation meeting these criteria, this will provide a rather strong, but still
defeasible, reason to think that the theory providing the explanation is true.
Such arguments, however they are characterized, cannot constitute the
whole of scientific reasoning that might be offered in support of a theory. In fact,
given the apparent structural features of these arguments, where two theories
are both able to provide accounts of a single empirical generalization, such
arguments are bound to be relatively rare in the history of science. Such
arguments must, therefore, constitute only a rarely used means of confirming a
theory with much o the confirmatory work being done elsewhere. Moreover,
such scientific virtue as these explanatory arguments may embody need not be
assumed to constitute the whole of the scientific virtues. No reason at all, for
example, has been given to think that there are no "supra-empirical" virtues,
such as beauty, simplicity, or parsimony, that might support one theory over
another. No reason at all has been given to think that there might not be some
other feature of a theory to recommend it over its rivals. Rather, the present sort
of argument appears to constitute a rarely used, but nonetheless extremely
important, defeasible indicator of the truth of theories. In subsequent chapters,
we will have the opportunity to explore some of the alternative empirical and
"supra-empirical" bases upon which one might try to evaluate competing
theories of cognitive architecture.