C R U C I A L IN S TA N C E S A N D F R A N C I S

B A C O N ’ S Q U ES T F O R CE RTA IN T Y
Daniel Schwartz

Francis Bacon’s method of induction is often understood as a form of eliminative in-

duction. The idea, on this interpretation, is to list the possible formal causes of a phe-
nomenon and, by reference to a copious and reliable natural history, to falsify all of
them but one. Whatever remains must be the formal cause. Bacon’s crucial instances
are often seen as the crowning example of this method. In this article, I argue that this
interpretation of crucial instances is mistaken, and it has caused us to lose sight of why
Bacon assigns crucial instances a special role in his quest for epistemic certainty about
formal causes. If crucial instances are interpreted eliminatively, then they are subject to
the two problems related to underdetermination raised by Duhem: (1) that it is impos-
sible to be certain one has specified all of the possible alternatives and (2) that an ex-
periment falsifies a whole theory, not just a single hypothesis in isolation. I show that
Bacon anticipates and aims to dodge both of these problems by conceiving of crucial
instances as working, in the ideal case, through direct affirmations that are supported
by links to more foundational knowledge.

Introduction
Francis Bacon’s method of induction is often understood as a form of elimina-
tive induction (e.g., Jardine 1974, 126; Cohen 1980, 222; Gaukroger 2001,
139). The idea, on this interpretation, is to list the possible formal causes of
a nature—“nature” here meaning something like a phenomenon—and, by ref-

Contact Daniel Schwartz at Rowan University, Department of Philosophy and Religion Studies,
201 Mullica Hill Rd., Glassboro, NJ 08028 (schwartzd1@rowan.edu).
I wrote much of this article during a postdoctoral fellowship at the University of California, San
Diego. I would like to thank the Anthem Foundation for Objectivist Scholarship for their grant sup-
porting that fellowship. I would also like to thank the many individuals who provided valuable feedback
on earlier versions of this article, including Donald Rutherford, Darryl Wright, and two anonymous
referees.

HOPOS: The Journal of the International Society for the History of Philosophy of Science, vol. 7 (Spring 2017).
2152-5188/2017/0701-0005$10.00. © 2017 by the International Society for the History of Philosophy of Science.
All rights reserved. Electronically published February 21, 2017.

130
 Schwartz | SPRING 2017

erence to a copious and reliable natural history, to falsify all of the possible causes
but one, which (in one of Bacon’s metaphors) is left behind like the residue af-
ter an alchemical experiment (Bacon 2004, 255, Novum organum Bk. II, aph-
orism 16). Whatever remains must be the formal cause.
Bacon’s crucial instances are often seen as the apotheosis of this method. As
they have come to be understood, crucial instances are supposed to help us
choose among the last remaining contenders for the true cause of a nature by
eliminating all possible causes but one. In this article, I argue that this interpre-
tation of crucial instances is mistaken, and it has caused us to lose sight of why
Bacon assigns crucial instances a special role in his quest for epistemic certainty
about formal causes. For when they are conceived of in the traditional elimina-
tive way, crucial instances are subject to concerns about underdetermination.
But when they are conceived of as noneliminative, the same concerns do not
arise.
It is already coming to be recognized in the literature that elimination is, at
most, one part of Bacon’s method. For example, there has been increasing atten-
tion paid to the natural historical stage of the method—to the role played by
literate experience (Pastorino 2011) and to how Bacon’s speculative natural phi-
losophy is intermixed with the experimental work (e.g., Rees 1975; Anstey
2012). Others, including Urbach (1987) and McCaskey (2006), deny that Ba-
con is an eliminativist at all and offer alternative interpretations according to
which he is, respectively, either more Popperian or more Aristotelian. Bacon’s
method has three main (very much intermixed) stages: (1) the gathering of nat-
ural history, (2) an eliminative stage in which the data in the natural history are
used to rule out possible formal causes, and (3) a stage in which additional elimi-
native and noneliminative resources are used (the nine auxilia outlined in Novum
organum II.21). Among those additional resources, Bacon provides the most in-
formation about Instances with Special Powers, among which crucial instances
are just one kind.
But when it comes to these crucial instances, it is difficult to find a satisfac-
tory noneliminativist account. One recent exposition by Claudia Dumitru notes
that one of Bacon’s examples of a crucial instance relies on direct confirmation
rather than elimination, but Dumitru goes on to assert that that instance’s “log-
ical structure is highly compatible with the classic Duhemian view of the crucial
experiment,” which is eliminative, because the value of a crucial instance “al-
ways comes from confirming one hypothesis while refuting others” (2013, 54).
This interpretation leaves Bacon vulnerable to Duhem’s concerns—wrongly so,
in my view.
This article offers a new noneliminativist interpretation of crucial instances.
In contrast to Dumitru and others, I deny that their logical structure is compat-

131
HOPOS | Bacon’s Quest for Certainty

ible with the classic Duhemian account. For Bacon believes that some crucial
instances offer direct and conclusive evidence of the formal cause of a nature
and that in such cases the other alternatives do not need to be falsified as part
of the process—although their falsity follows, as a matter of course, once the
true cause is discovered. As a result, I argue, Duhem’s well-known objections
to crucial instances are inapt, vis-à-vis Bacon’s conception of them.

What Crucial Instances Are

Crucial instances are the fourteenth kind of experiment on Bacon’s list of “In-
stances with Special Powers” ( praerogativis instantiarum), which occupies most
of Book II of the Novum organum. Each kind of instance on this list offers some
special value in the course of doing natural history and thinking inductively on
that basis. Crucial instances, though, are singled out for their role in concluding
an investigation with finality. Although it would be wrong to think that crucial
instances are supposed to help us arrive at certain knowledge of formal causes in
every case (since Bacon allows for their having varying degrees of exactness and
says that some are better than others), Bacon does say that the interpretation of
nature is sometimes completed with them—and they are the only Instance with
Special Powers about which he says this. Indeed, he explicitly contrasts crucial
instances with “probabilistic reasoning” (rationes probabiles) and frequently uses
“linguistic markers of certainty,” as Dumitru notes, throughout his examples
(Dumitru 2013, 54; Bacon 2004, 339, Novum organum II.36). For Bacon, the
champions of mere probabilistic or credible reasoning are the Academic skeptics,
whom he criticizes for seeking to “destroy the certainty of science” (1861–79,
2:687–88, my translation). So probabilistic reasoning is to be contrasted with rea-
soning that allows for certainty.1
The general pattern of crucial instances—which are so-called not because
they are crucially important but because they help point the way like signposts
(cruces) when one is at a crossroads—is described by Bacon as follows:

When in the investigation of any nature the intellect is finely balanced,

and so that it is uncertain as to which of two or sometimes more natures
the cause of the nature under investigation should be attributed or as-
signed (on account of the frequent and normal concurrence of many na-

1. But see Hacking (1975) for some of the complexities involved in interpreting early modern talk
of probability. As Hacking notes, “probability” was associated with what was approved or should have
been approved. This conception is broadly compatible with that of the Academic skeptics, and it is this
conception of “probability” that Bacon appears to contrast with “certainty.”

132
 Schwartz | SPRING 2017

tures), Crucial Instances indicate that the partnership of one of the natures
is (in relation to the nature under investigation) constant and indissolu-
ble, while that of another is variable and separable—whence the question
is settled, and the former nature is accepted as the cause while the other is
set aside and rejected. (Bacon 2004, 319–21, Novum organum II.36)

Each example is introduced with the call to “let the nature under investigation
be [sit Natura Inquisita]” such-and-such, a formulation that makes it clear that
the goal of crucial instances is not to choose among just any hypotheses but spe-
cifically to choose among hypotheses about the formal causes of natures.
The abstract description above does not correspond to every example that
follows it (not all of them “settle the question,” as I will discuss in a moment),
but it is still worth noting that within the abstract description Bacon does not
say that crucial instances work by means of falsification. The falsification of the
wrong hypothesis and the confirmation of the correct hypothesis are treated as if
they were on a par with each other—two conclusions, one affirmative and one
negative, inferred from a single crucial instance. Bacon offers no further insight
here into the logical structure of that inference, leaving it up to us to reconstruct
what we can from his examples.
Note also that Bacon states that the need for crucial instances arises from a
particular problem, “the frequent and normal concurrence of many natures.” I
take this to mean that two natures—such as the nature of earthly rotation and
the nature of cosmic rotation—would often (if instantiated) produce many of
the same effects. Sometimes all of our currently known instances are compatible
with either of two or more natures being the cause. Crucial instances help re-
solve deadlocks of this kind.
But they do not always do so with certainty. On this point, Hacking has it
right: “Later philosophy of science made crucial experiments absolutely deci-
sive. The picture is that two theories are in competition, and then one single
test conclusively favours one theory at the expense of the other. Even if the vic-
torious theory is not proved true, at least the rival is knocked out of action. That
is not what Bacon says about instances of the fingerposts” (1983, 249–50). But
let us not take this line of thought too far, for some crucial instances are sup-
posed to be absolutely decisive. In fact, a survey of Bacon’s 10 examples re-
veals two important distinctions among the crucial instances: first, whether
they purport to establish a conclusion with certainty—for some of them do
and some of them do not—and second, whether they point directly and affir-
matively to one nature as the cause of the nature under investigation or instead
work negatively and by elimination. If there are only two possible causes, as
in some of Bacon’s examples, then these eliminative crucial instances might lead

133
HOPOS | Bacon’s Quest for Certainty

us to affirm the remaining cause, but if there are three or more possible causes, as
in other examples, then the conclusion drawn from such a crucial instance will
not be affirmative at all. They might tell us with certainty only that some nature
is not the cause we are looking for.

Examples of Crucial Instances in Bacon’s Work

Here, then, are brief summaries of examples of each kind, although I will quickly
turn my attention to the most ambitious variety, the kind of crucial instance
that aims to establish with certainty and affirmatively what the cause of a na-
ture is.2 Note that this initial survey aims as far as possible for neutrality with
respect to competing interpretations of the logical structure of the examples,
which are pulled from the list in Novum organum II.36 (Bacon 2004).
Certain and affirmative crucial instances are as follows:
Diurnal Motion: The diurnal motion is caused by either terrestrial or
heavenly rotation. Information about (what amounts to) the linear veloc-
ity of objects at key points on the earth and in the heavens can be used to
falsify the Copernican account and affirm that the whole cosmos partic-
ipates in the diurnal motion.
Annual Motion: Bacon denied that the annual motion of the planets was
real and noted that real spiraling motions that proceed in the same direc-
tion as the diurnal motion could make the same predictions without the
need to invent entities such as epicycles. He proposes the following cru-
cial instance to test his view. If a comet can be found that permanently
and continuously moves opposite the diurnal motion (as opposed to sim-
ply alternating between retrograde and prograde motion), then an annual
motion is possible (“some such motion could exist in nature”; Bacon 2004,
327).
Weight: Weight is caused by either a nonrelational property of the heavy
object or an attractive force exerted by the earth. Information about the
weight of objects at various elevations (gathered through an experiment
involving two different types of clocks) can be used to falsify the account
that states that weight is a nonrelational property and confirm that it is
caused by the attractive force of the earth.

2. Bacon provides no examples of merely negative and probable crucial instances, and it seems pos-
sible that such examples would not be important enough to qualify as Instances with Special Powers at
all. For that reason, I omit this category.

134
 Schwartz | SPRING 2017

Flame in a Flame: The transitory nature of flame is caused by either the

cessation of the cause of the flame (such as the removal of a fuel source) or
the antagonistic force of the air. If a flame that it is shielded from the air
by a surrounding flame does not go out but instead takes on a spherical
shape, then the antagonistic force of the air must be the cause of the tran-
sitory nature of flame.
Here is an example of a certain but merely negative crucial instance:
Tide Species: On the assumption that one has already established that
the motion of the tides is an up-and-down rather than side-to-side one,
the rising and falling motion of the tides is caused either by an attractive
force that pulls up on the sea, the expansion of a fixed mass of water, or by
the inflow of additional water from beneath the ocean. If an attractive
force pulls up on the sea, it must pull up on the middle. Otherwise, by
pulling up on the sides at regular intervals, it would create a side-to-side
motion rather than a rising-and-falling one. Therefore, if one finds that it
is false that water bulges in the middle of the sea at the same time as low
tide on two opposite shores, then this “magnetic” account is falsified with
certainty. However, two other possible causes remain, such that the cru-
cial instance has not established the cause affirmatively.
Here is a probable and affirmative crucial instance:
Ordinary Flame: The transitory nature of flame is caused either by the
removal of the cause or by the antagonistic force of the air. The shape of a
flame (broader at the base and narrower at the tip) indicates that air is an-
tagonistic to flame, for the more time the air has to interact with the flame
as it proceeds upward, the narrower the flame becomes. Bacon says that
his other crucial instance concerning flame is a more exact one (accu-
ratior), implying that this one is not quite certain.
And the following crucial instance may be seen as either certain and affirma-
tive or certain and negative, depending on Bacon’s intentions. Dumitru refers to
it as an instance of refutation but also notes that Bacon might anticipate the re-
sulting confirmation without stating it explicitly:
Tide Genus: Tides are caused by either a side-to-side or an up-and-down
motion. Information about the time of high tide at various key points
around the globe could be used to falsify one account of the cause and
presumably affirm the other. (Bacon does not explicitly affirm the re-
maining account, though—hence the ambiguity.)

135
HOPOS | Bacon’s Quest for Certainty

My primary concern in this article is with the first kind of crucial instance, the
kind that aims to confirm a hypothesis with certainty. Now that I have noted
that less ambitious varieties are possible, I would like to put them aside in order
to discuss the first set of examples in more detail and make some observations
about their logical structure.

Affirmative and Certain Crucial Instances

In Bacon’s presentation, the first crucial instance that clearly fits this category
deals with the diurnal motion. Two possible causes are considered, the rotation
of the earth and the rotation of the heavens. Bacon’s characterization of this sec-
ond option is a departure from the traditional Aristotelian-Ptolemaic model. He
instead has in mind what he calls a cosmical motion, which he defines in the
Thema Coeli as one that “the heavenly bodies receive by consent not of the heav-
enly bodies alone but of the whole universe” (Bacon 1996, 181). The notion of
“consent” plays an interesting role here. The idea is that everything in the uni-
verse, from the stars and planets to the winds and from the oceans to compass
needles, participates in a single daily motion from east to west. Bacon does not
have in mind a motion that penetrates from the outside in. That would be an
example of what he calls mutual motion (ad invicem), which is motion that is
communicated from one particular thing to another. Instead, everything has an
built-in impulse to rotate in accord with the whole cosmos, and it is because
everything in the universe has this common nature that the diurnal motion ex-
ists.
The crucial instance in effect involves plotting the motion of objects at var-
ious key points spread out at various distances from the earth (and depths below
the earth) and at various terrestrial and celestial longitudes and latitudes. If ev-
erything, from the stars on down to the earth’s crust, shares in the same east-west
motion, then the motion is a cosmical one. One might object that a motion
could be communicated from the outside in and give rise to the same motions.
There is evidence in the Historia ventorum that Bacon recognizes this compli-
cation since he seeks additional crucial instances to rule out the possibility that
the sun causes the winds by heating the air, which would make this motion ad
invicem (Bacon 1996, 181).
Observe that there is no step in Bacon’s discussion in which he states that the
instance would falsify the geokinetic theory. Presumably, it would, but there is a
question whether the falsification of the geokinetic theory is logically prior or
logically posterior to the confirmation of the geostatic theory. And in this case,
it is not clear how falsification could come first. The instance has nothing to do

136
 Schwartz | SPRING 2017

with determining whether we indeed observe what we would expect to observe

if the earth were turning. Rather, we observe what we would expect to observe if
the cosmos were turning—we first discover that many parts of the cosmos are
turning together with the whole, and then we infer that all parts of the cosmos
are turning together with the whole. Inasmuch as this theory of cosmic motion
is incompatible with geokinetic theories, we must reject geokinetic theories as a
result.
In the next example, Bacon notes that a comet could prove the possibility of
a real annual motion that proceeds in the opposite direction as the diurnal mo-
tion. It is difficult in this example to determine what the initial set of alternatives
is supposed to be. In Bacon’s other examples, he indicates the set of alternatives
with the formula: necesse est ut. Not so in this example, even though he mentions
a number of alternatives throughout the discussion. Some possibilities that the
text allows include (1) a planetary cause versus a terrestrial cause, (2) a real cause
versus an illusion, (3) a planetary cause versus a terrestrial cause versus an illu-
sion, and (4) a motion in the same direction as the diurnal motion versus a mo-
tion in the opposite direction as the diurnal motion. Note that the conclusion is
affirmative and seemingly certain, but what is affirmed is not the actual exis-
tence of the annual motion of the planets but the possible existence of such a
motion—presumably because Bacon is reluctant to extrapolate from the mo-
tion of one comet to the motion of the planets. No matter how we carve up
the initial set of alternatives, it is difficult to see how the conclusion is derived
from or related to them. There is not a single alternative in any of the four op-
tions above which a comet definitely affirms or falsifies, yet the conclusion is the
definitive affirmation of the possibility of the annual motion.3
The best explanation of the text is that even if one of the above sets of alter-
natives is in the background, the conclusion does not actually emerge from those
alternatives. The alternatives motivate the inference but form no part of its log-
ical structure. A comet of the right sort is crucial because it would show, directly
and affirmatively, the actual existence of motion in the direction opposite the di-
urnal motion. Then there is a further inference: if the motion actually exists in a
comet, then it is possible for it to belong to the planets.
The same analysis holds true for the more certain of the two crucial instances
concerning the transitory nature of flame. What the experiment seems to make
evident is the fact of the antagonistic force of the air. It does this by means of an
apparatus that shields some of the flame from the air without changing any

3. Note, by the way, that the lack of such a comet would prove nothing. There would then be no
instance, let alone a crucial one.

137
HOPOS | Bacon’s Quest for Certainty

other variables that we have reason to believe will affect the flame. If the flame
loses its fleeting nature, Bacon seems to reason, then the air must be the cause of
that nature. Here, there is no step in which Bacon tries to falsify the competing
account according to which a continuous flame, like a continuous sound, is ac-
tually a succession that must be generated anew from moment to moment. In-
deed, the evidence Bacon provides is irrelevant to eliminating that alternative
except insofar as that alternative is mutually exclusive with the alternative that
he does affirm.
These examples support my claim that some instances work by means of di-
rect affirmation, not by falsification. But some crucial instances do appear to
work via falsification. In the example concerning the genus of the tides, the ini-
tial set of alternatives is used to logically ground an eliminative inference. For
the sake of completeness, it is worth briefly reviewing how that example works.
Bacon lists two possible kinds of cause: a side-to-side, progressive motion or
an up-and-down, rising-and-falling motion.4 If the motion of the tides is pro-
gressive, then a rise on one shore must be compensated for by a fall in water level
on some other shore. But if the motion is vertical then a high tide on one shore
should be roughly simultaneous with a high tide on an opposite shore, and a low
tide with a low tide.
This reasoning leads to the crucial instance, which consists in a set of obser-
vations detailing the times of high tide and low tide at various key points around
the world. If one selects key points that have nearly simultaneous high tides and
finds that there is no outside source that could be supplying the water to both
simultaneously, then the motion has to be a rising-and-falling one.
The selection of key points on opposite sides of the Atlantic would not be suf-
ficient. For example, simultaneous high tides in Florida and Spain do not decide
the matter, according to Bacon, because the Atlantic Ocean is narrow enough that
it might behave like a river, with water coming from the Indian Ocean, passing to
the south of Africa, and then flooding into the space between North America and
Europe so that the opposite shores have simultaneous high tides.
But if there were two more simultaneous high tides, one on the Pacific side of
the Isthmus of Panama and one on the eastern side of China, then there is no
body of water remaining to supply all of the additional water, and a progressive
motion would be ruled out. The South Pacific is much wider than the Atlantic,
and the passage south of South America is much narrower than the passage

4. A progressive motion is like the motion of water in a bowl when one tilts it from side to side—the
water falls on one side while it rises on the other. Rising motions involve either expansion of a constant
mass, vertical locomotion, or an increase in mass from another place.

138
 Schwartz | SPRING 2017

south of Africa, such that a sufficiently vast quantity of water could not enter
the Pacific basin quickly enough to make the opposite shores flood simulta-
neously.
So a progressive motion has been eliminated. If we assume that Bacon in-
tends for us to anticipate the affirmative conclusion that tides are caused by a
rising-and-falling motion, then the conclusion is affirmative, and it has been
reached by eliminating the alternative.
So to sum up this survey of Bacon’s examples, there are some relatively clear-
cut cases in which the alternatives motivate the inference but form no part of
its logical structure. The example concerning the genus of the tides is an excep-
tion, if we assume that Bacon intends for us to anticipate the affirmative con-
clusion. In the other examples surveyed, the affirmative step comes first, and
the nonaffirmed alternatives can be rejected afterward, insofar as they are in-
compatible with the affirmed hypothesis.
As for the logical structure of the affirmative inference itself, it is impossible
to provide a single analysis that fits all of Bacon’s examples. The flame-in-a-
flame case seems to rely on something like Mill’s method of difference: the
air is taken out of play, and if the flame loses its fleeting nature then the air
must be the cause of that fleeting nature. But the diurnal motion case is very
different. It seems to involve modeling the motion of a representative sample of
objects in the universe. And the case of the comet is different still. If we have
the right background knowledge, then we can directly observe whether a comet
is moving in the direction opposite to the usual direction of everything else in
the heavens. Perhaps all we can say is that Bacon is a pluralist about how affir-
mative inference works. The important point, for my purposes here, is that these
inferences are affirmative.

The Problem of Underdetermination

Before tackling the problem of underdetermination, let me say a word about the
utility of opposing Bacon and Duhem. One reason to bring Duhem to bear on
Bacon (or Bacon to bear on Duhem) is that, among the detractors of crucial ex-
periments, Duhem is relatively moderate and fair-minded, and he shares with
Bacon a commitment to drawing on the history of science to inform our phi-
losophy of science. Indeed, some post-Duhemian forms of the problem of
underdetermination are so radically weak or so radically strong that they are al-
ready widely accepted or rejected. At the weaker end, nobody could reasonably
deny that hypotheses are sometimes at least temporarily underdetermined by
the evidence currently available. It is important to point out that there is no rea-
son to think that Bacon would take issue with the idea of transient or even re-

139
HOPOS | Bacon’s Quest for Certainty

current transient underdetermination (see Stanford [2006] for a plausible ac-

count of recurrent transient underdetermination).
But the version of underdetermination defended by the early Quine in Two
Dogmas of Empiricism, according to which we may hold onto any theory “come
what may,” is so radically all embracing—it has been called permanent and ubiq-
uitous in contrast to Duhem’s more transient and localized underdetermina-
tion—that most reasonable commentators have already rejected it. (For a use-
ful survey of the underdetermination literature that emphasizes this contrast
between Duhem and Quine, see Pietsch [2012].) Bacon would reject Quine
too, but Duhemian underdetermination, which is confined to physics and
based on episodes in the history of physics, is far more interesting and plau-
sible.
Existing replies to the problem of underdetermination have, for the most
part, focused on Quine or on denying unsavory relativistic implications of
underdetermination while accepting the reality of the underlying problem.
For example, Laudan (1990) distinguishes between deductive and ampliative
forms of underdetermination and argues that although it is correct that we can-
not deduce a single theory from the phenomena, we can employ inductive prin-
ciples such as parsimony to single out a unique theory as the best one.
Laudan does not intend to undermine Duhem, though, and nothing like his
approach will do the job. For this sort of reply cannot get us the certainty that
Bacon is after. Bacon, when he is at his most epistemically ambitious, aims to be
able to show not just that a single hypothesis is the best one but that a single
hypothesis is the only one compatible with the evidence. Duhem’s objections
are serious, however, and even though Bacon and Duhem have different phil-
osophical contexts and are responding to different problems, Duhem’s objec-
tions would hit the mark if Bacon were just an eliminativist. So, in short, it
is worth noticing that a noneliminativist interpretation can constitute the be-
ginning of a response to Duhem.
Let us take a closer look at that response, then. An eliminativist about crucial
instances holds that they have the following structure:
Premise 1: Either hypothesis A or hypothesis B is true.
Premise 2: But hypothesis A is not true (falsified by observation or exper-
iment).
Conclusion: Therefore, hypothesis B is true.
Duhem identifies two sources of underdetermination that plague this mode of
inference, which others have subsequently labeled holist underdetermination

140
 Schwartz | SPRING 2017

and contrastive underdetermination.5 The problem of holist underdetermi-

nation is the charge that observations do not falsify a single hypothesis but in-
stead falsify the conjunction consisting of that hypothesis and all of the aux-
iliary assumptions. The problem of contrastive underdetermination charges
that our conclusion is underdetermined because we cannot establish the dis-
junctive premise with certainty. These problems are usually seen to render cru-
cial experiments impossible. Contrastive underdetermination threatens the first
premise (the objection is in italics):
Premise 1: Either hypothesis A or hypothesis B is true.
But what about hypotheses C, D, E, and so on?
And holist underdetermination threatens the second premise:
Premise 2: But hypothesis A is not true (falsified by observation or exper-
iment).
But there is no logical contradiction between the observation statement and
the hypothesis. To derive a contradiction, auxiliary assumptions must be added,
and we may choose to deny one of those assumptions instead of denying hypoth-
esis A.
Existing interpretations of crucial instances have not resolved these problems. It
is true that Dumitru modifies the Duhemian schema somewhat so as to allow
for something like as follows:
Premise 1: Either hypothesis A is true or hypothesis B is true.
Premise 2: But hypothesis A is not true (because observation or experi-
ment reveals B to be superior).
Conclusion: Therefore, hypothesis B is true.
Dumitru’s view is that a crucial instance need not rely on falsification but can
rely instead on showing that one of the available alternatives best fits the evi-
dence. There is much to like about this interpretation, but it falls far short of
Bacon’s epistemic ambitions. Premises 1 and 2 remain vulnerable, for the same
Duhemian reasons, at least if our goal is certainty.
My analysis of Bacon’s examples of crucial instances, however, has laid the
groundwork for an interpretation of crucial instances according to which nei-
ther of these problems arises. In some cases, premise 1 (in either version above)
may be entirely absent (or at least not function as a premise). And the confir-

5. This is the terminology employed in Stanford (2013).

141
HOPOS | Bacon’s Quest for Certainty

matory step (as expressed in the conclusions above) is logically prior to any re-
sulting eliminations. Although there are limits to the extent we should try to
schematize what is at bottom an inductive inference, the following argument
captures at least part of what is going on in some of Bacon’s examples:
Hypothesis B is true (affirmed directly by the evidence).
Hypothesis A is a competitor to hypothesis B.
Conclusion: Hypothesis A is false.
Let us make another pass through Bacon’s examples and see where this mode of
argument leaves the problem of underdetermination.
With respect to holist underdetermination, it is true that Bacon relies on
quite a bit of background knowledge in his crucial instances, some of it explic-
itly and some of it implicitly. But I want to argue that this use of background
knowledge does not obviously lead to holist underdetermination or threaten
the certainty of crucial instances.
Consider some implicit assumptions in the crucial instance concerning the
tides. Background knowledge about the winds (derived from a separate inquiry
into their nature) explains why Bacon might not consider the possibility that
wind could blow water toward Europe at the same time as it blows water toward
the Americas, which would allow a progressive motion to give rise to the effects
that he expects if the motion is up and down. As he makes clear in the De fluxu,
winds can produce smaller variations in the motion of the sea (the currents), but
they cannot explain motions as sizable and as regularly pattered as the tides.
Bacon also has to rely on geographical knowledge, which itself presupposes
prior knowledge from other inquiries. For example, we might need to know that
the continents do not drift so quickly as to invalidate our maps from one day or
year to the next.
Bacon explicitly assumes in this crucial instance that the earth is motionless
and states that if the earth does move (either with its own proper motion, as the
Copernican theory states, or with a motion that participates in the general cos-
mical motion) then a progressive motion might be able to explain simultaneous
high tides at all the key points. Dumitru also notes the reliance on auxiliary as-
sumptions in this example and notes that it “might hint to a problem of holistic
underdetermination” (2013, 51). The concern is that the data would falsify not
simply the hypothesis that tides are caused by a progressive motion but rather
the conjunction consisting of that hypothesis and every other required belief.
The crucial instance thus tells us that either the tidal motion is not progressive
or the earth is not motionless or our maps are not accurate, and so on, and we
must then rely on good sense or uncertain inductive rules to go any further.

142
 Schwartz | SPRING 2017

Note that holistic underdetermination, if it indeed threatened falsifying or

negative instances, would appear to be just as much a problem for affirmative in-
stances. It does not arise in quite the same way, since we are not confronted with
a choice among different beliefs to reject. But it is still true that we must rely on
background knowledge to interpret our crucial instance, so that it could be ar-
gued that we may sometimes have to choose between affirming the hypothesis
and rejecting some auxiliary assumption. In the flame-in-a-flame case, for exam-
ple, it could be argued that Bacon draws on his belief that the stars are spherical
balls of fire akin to the shielded flame. But what if the stars are made not of fire
but of ether? A spherical flame might suddenly lose much of its significance.
But there are two more characteristics of Bacon’s approach that aim to safe-
guard the certainty of its results: (1) modularity and (2) foundationalism. By
calling Baconian crucial instances modular, I mean that they can establish con-
clusions that are conditional on the results of inquiries into other natures. The
crucial instance about the tides establishes with certainty that if the earth is
motionless, then the tidal motion is a rising-and-falling one. Bacon himself
does not think that the antecedent of this conditional holds true (for he believes
that the tidal motion is progressive and is caused by the participation of the tides
in the diurnal motion), yet he still uses this as an example of a crucial instance.
The goal, of course, is the knowledge of natures, not of conditional propositions.
But Bacon shows himself to be an opportunist, and this sort of modularity—
in which we push our inquiry into one nature to the fullest extent possible and
await further information from other related inquiries—is intended to be a means
to that goal. If, at some point during the inquiry into the nature of the diurnal
motion, one establishes with certainty that the earth is motionless, then one
can return to the inquiry concerning the tidal motion and simply remove the con-
ditional.6
Bacon’s modular approach is especially evident when we look at his practice
of linking different inquiries with each other. In his Latin natural histories,
he often uses the label “connexio” to indicate that other inquiries should be
brought to bear on the current one,7 and in the De fluxu he uses the heading
“Syzigiae,” or conjunction, for the same purpose (Bacon 1996, 91). Bacon also
describes the importance of stating such links in the Abecedarium: “I interpose
connections and links to stop the inquiries being too abrupt” (2000, 223).
“Abrupt” is Rees’s possibly overly literal translation of the Latin abruptae, which

6. Compare Bacon’s opportunism when he says he could conceivably take advantage of a comet to
prove that the annual motion is possible, even though it would not prove that the annual motion is actual.
7. See, e.g., the Historia ventorum, where this practice is so frequent that one can flip to almost any
page and find the label in bold.

143
HOPOS | Bacon’s Quest for Certainty

is probably better translated as “separated” or “disjointed” in this context. The

practice of linking inquiries also offers the beginning of a response to Dumi-
tru’s question, “In the face of an apparent refutation/confirmation, how do we
decide if we need to reach for additional auxiliary conditions or if we should
just accept the result and end the inquiry?” (2013, 51). The answer must involve
consulting linked inquiries.
But we must also interpret this modularism through the lens of Bacon’s
foundationalism—which is the second feature of Bacon’s approach relevant
to the issue of holism. For Bacon to succeed, he must be able to show that
he appeals not to “auxiliary assumptions” that are on a par with the hypothesis
he aims to affirm but to previously established foundational building blocks.
If we fall short of that, then our links might allow for us to draw probable con-
clusions, not certain ones. That is why Bacon advertises himself as wiping the
slate clean and starting over from the foundations.
Duhem himself recognized that this foundationalist defense of crucial exper-
iments was to be expected. But he did not think the problem of contrastive
underdetermination was so easily answered. It is worth quoting Duhem here
since, in this passage, he might as well be addressing Bacon directly:

But let us admit for a moment that in each of these systems [concerning
the nature of light] everything is compelled to be necessary by strict logic,
except a single hypothesis; consequently, let us admit that the facts, in
condemning one of the two systems, condemn once and for all the single
doubtful assumption it contains. Does it follow that we can find in the
‘crucial experiment’ an irrefutable procedure for transforming one of the
two hypotheses before us into a demonstrated truth? Between two con-
tradictory theorems of geometry there is no room for a third judgment; if
one is false, the other is necessarily true. Do two hypotheses in physics
ever constitute such a strict dilemma? Shall we ever dare to assert that
no other hypothesis is imaginable? Light may be a swarm of projectiles,
or it may be a vibratory motion whose waves are propagated in a me-
dium; is it forbidden to be anything else at all? (1906/1954, 189)

The analysis of Bacon’s examples has already revealed that some crucial in-
stances are not intended (as Duhem imagines) to proceed by means of falsify-
ing the incorrect hypotheses but rather by means of directly affirming the
correct hypotheses. False hypotheses, including any that have as yet been un-
conceived, are then falsified as a matter of course because they conflict with
the conclusion.

144
 Schwartz | SPRING 2017

But even though exhausting the alternatives may not always be necessary,
Bacon does believe it possible. In listing the alternatives at the start of each ex-
ample, he often says that the list is necessary—“necesse est.” The tides are one
such example. It is necessary, he says, that the motion is either progressive or
up and down. Perhaps it is no coincidence that this example appears to work
by means of falsification.
Then how does Bacon justify his certainty that the cause is among the al-
ternatives listed? He must rely on prior knowledge of the natures described in
each alternative, gathered empirically from inquiries into those other natures.
In other words, he again relies on background knowledge, and the viability of
his foundationalism will again determine whether he can succeed. The exam-
ple concerning the tides offers one of the most compelling alternatives. Are
there any alternatives that fit the phenomena other than side-to-side and
up-and-down motion? My view is that Bacon intends for there to be a linked
inquiry into the nature of motion, one that might provide an a posteriori de-
fense of this way of carving up the alternatives.
But philosophers are very skilled at coming up with still other conceivable
alternatives. Perhaps the motion is of a compound sort; or worse, perhaps it
can only be explained if we think of motion as involving more than three di-
mensions (some of them unperceived by us). Or perhaps the correct genus is
so far beyond our current comprehension that it would never cross our minds.
This is a fair objection to Bacon, and it limits the justificatory force of any
merely negative crucial instance, so I concede that Duhem is right that there
are no falsifying crucial instances that establish a particular hypothesis with cer-
tainty. But falsifying crucial instances are not the only kind, as Bacon’s examples
have shown. I have been calling this other category of crucial instances “affirma-
tive.” Next, I want to justify that terminology by showing that Bacon himself
recognizes a category of inference called “reasoning with affirmatives.” I also
provide some additional evidence that Bacon conceives of some crucial instances
as belonging to that category.

Crucial Instances and Affirmative Reasoning

Bacon first elaborates the distinction between reasoning by means of affirma-
tives and negatives in De principiis atque originibus. There, he uses Greek myths
about Cupid and Coelum as the springboard for an elaboration of “the philos-
ophy of Parmenides and Telesio and especially that of Democritus” (Bacon 1996,
197). Cupid turns out to be a symbol of the atom and the initial principle of
motion that was imparted to it by God. In the myth as Bacon tells it, Cupid

145
HOPOS | Bacon’s Quest for Certainty

“came from an egg laid by Night” (197). This symbolizes the method that we
are to use in investigating the atom and its first principle of motion:

Now that business of Night’s egg is a very happy allusion to the demon-
strations which bring this Cupid to light. For things concluded by affir-
matives seem to be the offspring of light, whereas those concluded by
negatives and exclusions are wrung out and brought forth as if from
the obscurity of night. Now this Cupid is truly an egg hatched by Night,
for knowledge of him (all that may be had) proceeds by exclusions and
negatives. But proof made by exclusion is a kind of ignorance, and as it
were Night, with regard to what is included in it. (Bacon 1996, 201)

Because first principles have no prior natural causes—God is the only prior
cause—we cannot know them, as we know everything else, by means of causes.
Their primacy also makes them hard to investigate directly. We have to use an
eliminative method to learn anything about them. The atom is not visible, not
colored, not heavy, not light, and so on. These are properties that can be excluded
from its nature.
Even when it comes to the investigation of first principles, we eventually
come to affirm things of them:

But the parable further suggests that there is some end and limit to
these exclusions of which I have spoken, for Night does not incubate
forever. Certainly it belongs to God alone, that, when His nature is in-
quired of by the sense, exclusions shall not end in affirmatives. But here
the case is otherwise, that is, that after due exclusions and negations
something is affirmed and established, and an egg hatched, as it were,
after an appropriate period of incubation and not only is it that the
egg is hatched by Night but also that the person of Cupid is hatched
from the egg, that is, not only is some notion of this matter drawn and ex-
tracted out of ignorance, but also a distinct and definite notion. (Bacon
1996, 203)

What is most interesting about this passage, for my purposes, is what it implies
about the method of investigating things that are not first principles. God is
singled out as the only subject of inquiry for which the reasoning does not
end in something affirmative, and the atom and its principle of motion is sin-
gled out (in the previous passage) as the only natural phenomenon for which the
affirmatives are to be sought through an exclusively eliminative process. This
implies that other natural phenomena can be investigated directly through

146
 Schwartz | SPRING 2017

affirmatives—that they can come from eggs hatched by Light, to use Bacon’s
metaphor.
Bacon’s occasional talk of reasoning with affirmatives in the Novum organum
is compatible with my reading, but admittedly not decisive. In II.15, for exam-
ple, Bacon says that we must finish up with affirmatives:

Now if the mind tries to proceed by affirmative instances from the start
(which it always does when left to itself ) the result will be phantasms,
mere opinions, ill-defined and notional conclusions, and axioms altered
daily, unless we want to follow the schoolmen’s fashion and fight for false-
hoods. These will no doubt be better or worse according to the capacity
and strength of the intellect working on them. But only to God (the cre-
ator and implanter of forms), or perhaps to the Angels and Intelligences,
does immediate knowledge of forms by affirmation belong as soon as they
begin thinking about them. But this is certainly beyond men who are al-
lowed only to proceed by Negatives at first, and then to finish up with
Affirmatives after making every sort of exclusion. (2004, 253)

This passage tells us only that the eliminative stage must precede the affirmative
stage, but it does not tell us how the affirmative stage works or rule out that it
works just through the affirmation of the only remaining possible cause after all
others have been eliminated. The same ambiguity is present in II.19, where Ba-
con says that “the foundations of true Induction rest on the Exclusive process, yet
the process is not finished until it ends in an affirmative” (Bacon 2004, 261).
We can make some progress, though, by inquiring into what these mysterious-
sounding “affirmatives” are. Here I turn to a famous passage in Book I of the
Novum organum that deals with one of the causes of the idols of the tribe
(a cause that we would now call confirmation bias):

The human intellect takes the conceptions which have won its approval
(by general acceptance, credit, or simple charm), and pulls everything else
into line and agreement with them, and although the abundance and
strength of the contrary instances it encounters is greater, the intellect
at enormous cost overlooks and despises them, and dismisses them and
rejects them by making distinctions to keep the conceptions just men-
tioned intact. So that man gave a good answer who, after being shown
a picture hanging in the temple of people who, having said their prayers,
escaped the perils of shipwreck, and after he was pressed to say whether he
did not now acknowledge the power of the gods, asked this question in
return: But where are they depicted who, after solemn prayers, perished? . . .

147
HOPOS | Bacon’s Quest for Certainty

But even if the charm and vanity mentioned above were absent, the hu-
man intellect would still suffer from the peculiar and permanent error of
being moved and excited more by affirmatives than negatives, when it
ought to pay heed in a proper and systematic way to both equally; indeed,
in the true setting up of every axiom, the power of the negative instance is
actually greater. (Bacon 2004, 83–85, Novum organum I.46)

In this context, a negative instance is one that counts as evidence against a hy-
pothesis, while an affirmative instance is one that counts in favor of a hypoth-
esis. In the shipwreck example, the painting of men who prayed and escaped a
shipwreck is an affirmative instance because it counts in favor of the hypothesis
that prayer is effective, while the many instances of men who prayed and then
died are all negative instances because they count against the same hypothesis.
According to Bacon, we tend to overvalue affirmative instances and overlook
negative instances. The eliminative stage of his method is meant in part to bal-
ance out that natural tendency. But affirmative instances are not to be ignored
completely. There are examples of Instances with Special Powers, such as shin-
ing instances, which play an affirmative role alongside that method of exclusion.
Shining Instances are those in which the nature under investigation especially
shines forth because it is uninhibited by opposing natures. Like some crucial
instances, their role is to confirm, not to falsify, although they cannot do so with
nearly the same degree of certainty.
There is some indirect textual support for the view that crucial instances can
be affirmative instances. One piece of evidence is Bacon’s metaphorical descrip-
tion of crucial instances as experiments of light. “Since I very often lack history
and experiments, especially experiments of light and crucial instances which can
inform the mind about the true causes of things, I give directions for new ex-
periments suitable, as far as I can tell at present, for the subject under inquiry”
(Bacon 2007, 15). Another is at the conclusion of the Novum organum dis-
cussion: “And so much for Crucial Instances which I have dealt with at some
length with the aim of getting men gradually to learn and become used to judg-
ing nature with Crucial Instances and light-bearing experiments, not with prob-
abilistic reasoning” (Bacon 2004, 339). His use of this metaphor is not entirely
consistent, as sometimes he contrasts experiments of fruit, which have an im-
mediately practical purpose, and experiments of light, whose immediate pur-
pose is further knowledge, in such a way that even negative instances shed this
metaphorical light. But the De principiis atque originibus passage we have just
examined explains the equation between crucial instances and experiments of
light. Light, in this context, is to be contrasted with night, not with fruit. (Recall
that Cupid comes from an egg laid by Night.) Light is associated with both cer-

148
 Schwartz | SPRING 2017

tainty and affirmative reasoning, while night is associated with probabilistic and
eliminative reasoning.
Bacon’s examples confirm my reading. The crucial instances that I have clas-
sified as certain and affirmative use observations that (if the “auxiliary assump-
tions” are certain) we can easily imagine Bacon thinking necessitate the causal
claim, regardless of whether other hypotheses have been falsified. In addition,
Bacon sometimes requires more observations than would strictly be necessary
to falsify the alternatives that he lists, which is evidence that he is concerned
with establishing the cause by means of positive evidence rather than through
the mere process of elimination.
In the crucial instance concerning diurnal motion, if one has information
about the motion of objects at all of the possible distances from the earth (from
the ground all the way to the fixed stars), then one can model the motion of the
universe as if the whole thing were there to be observed in a single gaze. (The
example concerning the rising-and-falling motion of the tides also seems to in-
volve modeling the flow of water around the globe.) What warrants our extrap-
olating from the limited set of data to the motion of points that have not yet
been recorded (because they are in places or times where we have not looked)
must be our background knowledge, such as background knowledge about
fluid dynamics in the case of the tides.
It is admittedly difficult to gather and hold in mind all of the necessary back-
ground knowledge and to see that only one model fits the evidence, but that is a
different problem than underdetermination. The affirmative character of this
type of crucial instance means that the impossibility of listing every alternative
fails to give rise to the problem of contrastive underdetermination. If the obser-
vations along with any genuine background knowledge require a particular
causal account, then it does not matter which other hypotheses are out there
waiting to be conceived.
Have I shown that Bacon’s crucial instances can escape the problems of
underdetermination? Not definitively. There are still many questions about
how the affirmative stage of the inference is supposed to work. But I have shown
that framing the objection to Bacon in Duhemian terms begs the question
against Bacon’s foundationalism and relies on a straw man portrayal of him
as an eliminativist. Whether further objections crop up, when Bacon is under-
stood in this new way, is a topic for further discussion.

REFERENCES
Anstey, Peter. 2012. “Francis Bacon and the Classification of Natural History.” Early Science
and Medicine 17 (1–2): 11–31.

149
HOPOS | Bacon’s Quest for Certainty

Bacon, Francis. 1861–79. The Works of Francis Bacon. Ed. James Spedding, Robert L. Ellis,
and D. D. Heath. 14 vols. London: Longman.
———. 1996. Philosophical Studies c.1611–c.1619. Trans. and ed. Graham Rees and Mi-
chael Edwards. The Oxford Francis Bacon 6. Oxford: Clarendon.
———. 2000. The Instauratio magna: Last Writings. Trans. and ed. Graham Rees. The Ox-
ford Francis Bacon 13. Oxford: Clarendon.
———. 2004. The Instauratio magna. Pt. 2, Novum organum and Associated Texts. Trans. and
ed. Graham Rees and Maria Wakely. The Oxford Francis Bacon 11. Oxford: Clarendon.
———. 2007. The Instauratio magna. Pt. 3, Historia naturalis. Trans. and ed. Graham Rees
and Maria Wakely. The Oxford Francis Bacon 12. Oxford: Clarendon.
Cohen, Laurence Jonathan. 1980. “Some Historical Remarks on the Baconian Conception
of Probability.” Journal of the History of Ideas 41 (2): 219–31.
Duhem, Pierre. 1906/1954. The Aim and Structure of Physical Theory. Repr. Princeton, NJ:
Princeton University Press.
Dumitru, Claudia. 2013. “Crucial Instances and Crucial Experiments in Bacon, Boyle, and
Hooke.” Societate si politica 7 (1): 45–61.
Gaukroger, Stephen. 2001. Francis Bacon and the Transformation of Early-Modern Philoso-
phy. Cambridge: Cambridge University Press.
Hacking, Ian. 1975. The Emergence of Probability: A Philosophical Study of Early Ideas about
Probability, Induction and Statistical Inference. Cambridge: Cambridge University Press.
———. 1983. Representing and Intervening: Introductory Topics in the Philosophy of Natural
Science. Cambridge: Cambridge University Press.
Jardine, Lisa. 1974. Francis Bacon: Discovery and the Art of Discourse. Cambridge: Cambridge
University Press.
Laudan, Larry. 1990. “Demystifying Underdetermination.” In Scientific Theories, ed. C. Wade
Savage, 267–97. Minnesota Studies in the Philosophy of Science 14. Minneapolis: Uni-
versity of Minnesota Press.
McCaskey, John. 2006. “Regula Socratis: The Rediscovery of Ancient Induction in Early
Modern England.” PhD diss., Stanford University.
Pastorino, Cesare. 2011. “Weighing Experience: Experimental Histories and Francis Bacon’s
Quantitative Program.” Early Science and Medicine 16 (6): 542–70.
Pietsch, Wolfgang. 2012. “The Underdetermination Debate: How Lack of History Leads to
Bad Philosophy.” In Integrating History and Philosophy of Science: Problems and Prospects,
ed. Seymour Mauskof and Tad Schamltz, 83–106. Boston Studies in the Philosophy of
Science 263. Dordrecht: Springer.
Rees, Graham. 1975. “Francis Bacon’s Semi-Paracelsian Cosmology.” Ambix 22 (2): 81–101.
Stanford, P. Kyle. 2006. Exceeding Our Grasp: Science, History, and the Problem of Unconceived
Alternatives. New York: Oxford University Press.
———. 2013. “Underdetermination of Scientific Theory.” In Stanford Encyclopedia of Phi-
losophy, ed. Edward N. Zalta. Stanford, CA: Stanford University. http://plato.stanford
.edu/archives/win2013/entries/scientific-underdetermination/.
Urbach, Peter. 1987. Francis Bacon’s Philosophy of Science: An Account and a Reappraisal. Chi-
cago: Open Court.

150