Erkenn (2016) 81:297–304

DOI 10.1007/s10670-015-9740-3

CRITICAL DISCUSSION

On the Epistemology of the Precautionary Principle:

Reply to Steglich-Petersen

J. Adam Carter1 • Martin Peterson2

Received: 13 December 2014 / Accepted: 18 May 2015 / Published online: 5 June 2015
 Springer Science+Business Media Dordrecht 2015

Abstract In a recent paper in this journal (2014), we proposed two novel puzzles
associated with the precautionary principle. Both are puzzles that materialise, we
argue, once we investigate the principle through an epistemological lens, and each
constitutes a philosophical hurdle for any proponent of a plausible version of the
precautionary principle. Steglich-Petersen (Erkenntnis 1–9, 2014) claims, also in
this journal, that he has resolved our puzzles. In this short note, we explain why we
remain skeptical.

1 Introduction

The core idea motivating the precautionary principle is that it is sometimes

appropriate to take preventative action rather than wait for more information to
come in.1 In a recent paper in this journal (2015), we proposed two novel puzzles
associated with the precautionary principle. Both are puzzles that materialise, we
argue, once we investigate the principle through an epistemological lens, and each
constitutes a philosophical hurdle for any proponent of a plausible version of the
precautionary principle. Steglich-Petersen (2014) claims, also in this journal, that he
has resolved our puzzles. In this note we explain why we remain skeptical.

1
See Aven (2011), Manson (2002), Munthe (2011), Sandin (1999), Sandin (2004) and Peterson (2003)
for a sample of some notable interpretations of the precautionary principle.

& Martin Peterson

martinpeterson@[Link]
1
Eidyn Research Centre, School of Philosophy, Psychology and Language Sciences, University
of Edinburgh, Dugald Stewart Building (Rm 5.04), 3 Charles Street, Edinburgh EH8 9AD,
Scotland, UK
2
Department of Philosophy, Texas A&M University, College Station, TX 77843, USA

123
298 J. A. Carter, M. Peterson

2 The First Puzzle

The precautionary principle can be understood as composed of three central

elements: a damage condition (D), an epistemological condition (E), and a
suggested remedy (R). If D and E are satisfied, then R is prescribed, or activated. On
this line of thinking, if there is sufficient epistemic confidence that an activity will
bring about a damage of a certain degree of severity, then this is sufficient grounds
for taking precautionary measures against the activity.2 In short, the rule template is:
E&D ? R.
Any version of the precautionary principle in which E and D are too easily
satisfied will be implausibly demanding. It would be absurd, for instance, to suppose
we must take regulatory action against laws that permit bicycling because there are
some (unsubstantiated) claims that bicycling causes some near-negligible bad
consequence.3 By contrast, if the conditions for satisfying E and D are too strict, the
resulting decision rule will be practically inert. The rule will prescribe action against
an activity, a, only if we have something like Cartesian certainty that some a causes
some maximally destructive damage.
The gist of the first puzzle we posed can be appreciated in the form of a dilemma
that arises once we ask: should the epistemic standards that must be met for ‘E’ in
the template to be satisfied depend on the severity of the anticipated damage (D)?
Here we envisioned two styles of response. Call invariantism the view that the
question posed be answered negatively, and contextualism the view that the question
be answered positively. Our ‘first puzzle’ for the proponent of the precautionary
principle emerges because either strategy seems to lead to a dead end.
Running a reductio against invariantism is simple. Suppose there is some body of
evidence that building a certain experimental nuclear energy reactor (call this
activity ‘a’) will melt down in a way that causes (in Case A) ten deaths, or (in Case
B) ten billion deaths. How strong must the decision maker’s epistemic position be to
rationally mandate precautionary action against a in each case? Here the invariantist
has a flimsy story: the same in both cases. But surely preventative action against a is
warranted in Case B on epistemic grounds that are not as good as they must be to
warrant preventative action against a in Case A. After all, Case B threatens to wipe
out ten billion lives.
The contextualist avoids this problem, but only by incurring another. The
contextualist describes the relationship between E and D in terms of negative
correlation: the greater the severity of the anticipated damage, the lower the
epistemic standard that needs to be met for the relevant regulatory action to be

2
See Carter and Peterson (2015) for a more detailed presentation.
3
Just suppose (for the moment) that quality of one’s epistemic position, as well as the severity of some
anticipated damage, can be represented numerically by the real numbers between 0 and 1. (This, note, is
in fact an oversimplification of things, though this assumption suffices to make the present point). Now,
for reductio, imagine a (bizarre) version of the precautionary principle that says: so long as (i) one’s
epistemic position e with respect to the causal connection between some activity, a, and an anticipated
damage, d is such that e [ 0, and (ii) the ‘badness’ of an anticipated damage (d) is such that d [ 0, then
(iii) R, where (recall) R specifies some regulatory action against a. Such a principle, set where e and d are
infinitesimally low, will rationally mandate regulatory action against nearly all actions.

123
On the Epistemology of the Precautionary Principle: Reply… 299

prescribed. In this respect, the strength of epistemic position that must be met tracks
the practical stakes.4 And this seems to get the right result in the case above.
But the proponent of the precautionary principle who takes the contextualist
route (where E is understood as depending on D) walks right into a trap, or, more
weakly, a puzzle. To make this point concrete, consider that on the contextualist
route, the rating of the severity of the anticipated damage circumscribes how strong
the epistemic position must be for the E condition to be satisfied. Though, as we
noted, in situations where an application of the precautionary principle is germane,
we have competing interests at play, and the extent to which the severity of some
anticipated damage is judged is in many cases going to vary dramatically with
respect to whose interests are at stake. Therefore, the precautionary principle can be
plausibly formulated as a decision rule only if supplemented with some additional
favouring rule, a rule which adjudicates whose interests are relevant to determining
the severity of the damage in question, a severity that for the contextualist is what
fixes the relevant epistemic standard that must be satisfied.5
At this point the contextualist has no good card to play. In our original paper we
argued that the precautionary principle fails to be an appropriately neutral decision
rule if, built into what is relevant to assessing the severity of damage (a severity which
then, on the contextualist view, determines the strength that must be met for the E
condition to be satisfied) is some particular interest—say, the interest of developers as
opposed to environmentalist groups (groups who might well rate the severity of the
anticipated damage very differently). But the contextualist cannot simply retreat to the
natural countermove—that of saying that all interests be weighted equally—call this
an egalitarian favouring rule. This move, we argued, will ultimately incur the same
problems associated with invariantism.6 We considered, finally, a third option for the
contextualist: to contextualise the favouring rule itself and argued that this merely
forestalls, rather than solves, the problem. As no option looks tractable, the proponent
of the precautionary principle qua decision rule thus has a bona fide puzzle on her
hands, one that trades importantly on how to think about just how the epistemological
condition should be satisfied.
However, according to Steglich-Petersen (2014) the puzzle sketched above
should, contrary to what we have suggested, not trouble proponents of the

4
The thought that epistemic standards should track practical stakes has obvious commonalities with
contextualist approaches in epistemology (e.g. DeRose 1992; DeRose 2009; Cohen 1998) according to
which the epistemic standards that must be satisfied for an individual to count as knowing a proposition
are sensitive to the practical stakes, in the context in which the knowledge ascription is made. While we
term the approach to the precautionary principle we are envisaging contextualism, we should note that
contextualism in epistemology is, at root, a semantic thesis about knowledge attributions. Our discussion
of the precautionary principle is situated within a semantic framework, and the general suggestion that
epistemic standards track practical stakes is germane to both contextualist and subject-sensitive
invariantist semantics for knowledge attributions (e.g. Stanley 2005; Hawthorne 2004).
5
See Carter and Peterson (2015: 7).
6
In short, given the entrenched political interests of (for example) developers and environmental
regulatory bodies, the mean of all relevant standards for damage will typically be insensitive to
differences in actual damage. But sensitivity to the severity of actual damage is precisely what
contextualist approaches can claim, as a key advantage, over invariantist approaches. Retreating to
contextualism supplemented with an egalitarian favouring rule surrenders this advantage.

123
300 J. A. Carter, M. Peterson

precautionary principle. His rationale takes as a starting point the correct

observation that, independently of any epistemic considerations, use of the
precautionary principle will involve adjudicating between competing interests
involved in the decision scenario (Steglich-Petersen 2014: 3). No arguments here.
Steglich-Petersen then observes, also correctly, that any decision rule that relies on
assessments of costs and values will be subject to some prior adjudication of
interests, especially when applied in policy making. And this fact is ‘not normally
seen as an argument against the usefulness [of the decision rule in question]’ nor for
adopting an additional favouring rule. As Steglich-Petersen writes:
Such adjudication will typically be done through the normal political
processes, sometimes democratic, which are designed to facilitate adjudication
of that exact kind … [But this is not] a problem arising from the contextualist
approach to the epistemic standards involved with the precautionary principle,
and Carter and Peterson [have not] succeeded in pointing out a new puzzle for
the precautionary principle. (Steglich-Petersen 2014: 3.)
In short, Steglich-Petersen has attempted an undercutting response to our puzzle.
Rather than to show how it can be met, he argues that it doesn’t need to be met, and
he argues for this via the familiar strategy of overgeneralization: Our puzzle should
worry a proponent of the precautionary principle only if our puzzle also constitutes a
legitimate problem for any decision rule that must weigh competing interests.
Steglich-Petersen is right that if we had drawn attention to a fact about the
precautionary principle which applies mutatis mutandis to any decision rule the use
of which involves weighing competing interests, then we will not have in doing so
highlighted any puzzle that is specific to the precautionary principle. But we reject
this characterisation of our argument. In particular, we resist the suggestion that the
problem we have posed vis-a-vis the precautionary principle is in fact one that
overgeneralizes as Steglich-Petersen suggests. We want to be clear that the puzzle
we have raised does not arise because applying the decision rule involves
adjudicating per se between competing interests involved in the decision scenario.
Consider a decision rule that says for any surplus S, give S to the most needy. The
application of this rule involves adjudicating between competing interests with
respect to whom are rightly described as the needy. But this shows nothing perverse
about the rule itself, even if there is some difficulty in applying the rule—a difficulty
that, as Steglich-Petersen observes, will typically be met democratically.
Our argument, more subtly, attempts to show that there is a special difficulty with
respect to how to formulate (and not merely apply) the precautionary principle, in a
way that satisfactorily captures how the epistemological component of the view
should be regarded as satisfied. This special difficulty arises in light of the
relationship between the ‘D’ and ‘E’ conditions, in the decision rule. As we have
suggested (both here and in our original paper), we think it is obvious that a
formulation of the precautionary principle on which the level of epistemic support
that must be attained to satisfy the ‘E’ condition does not track practical stakes (e.g.
the extent of the anticipated damage) is simply unworkable—as suggested by our
reductio. However the only natural move left for the purposes of formulating the
principle in a plausible way is to ‘go contextualist’. Though we systematically

123
On the Epistemology of the Precautionary Principle: Reply… 301

showed that each of the available ways of going contextualist looks prima facie
intractable. And this is problematical for the proponent of the precautionary
principle even if we grant with Steglich-Petersen that, in the default case where a
decision rule is applied in a way that involves the adjudication of interests, such
adjudication can be unproblematically accomplished.7

3 The Second Puzzle

Everyone agrees that if we were to take precautionary measures against every risk
that could lead to a sufficiently bad outcome, we would have to apply the
precautionary principle to far too many (perhaps all) activities. The point of
departure for the second puzzle is the plausible idea is that in order to avoid this
problem, the precautionary principle has to be formulated in a way that is
compatible with the de minimis principle, according to which extremely unlikely
threats should be ignored, i.e. threats that fall below some appropriate epistemic
threshold.
In our original paper, we argued that the best way to express this epistemic
condition is to assign some non-zero probability to the hypothesis that the risk is
below the de minimis threshold.8 A decision about how to address a risk should
therefore be based on two probability functions: the first describes the first-order
probability that the risk is de minimis, and the other describes the second-order
probability that the first-order probability is correct.
The second puzzle arises as we try to render this connection between the two
probability functions and the decision to treat a risk as de minimis more precise. We

7
We want to emphasise that simply adverting to the point that (in typical cases) where the application of
a decision rule involves weighing interests, such adjudication can be satisfactorily accomplished through
democratic political means is not a point that gains any traction for the purposes of spelling out a
plausible ‘contextualist’ route to articulating the epistemological condition of the precautionary principle.
Such a democratic process will be one that, at least in principle, will aim to weight competing interests
equally. However, this approach quickly begins to resemble in the relevant respects what we called the
egalitarian favouring rule—the favouring rule we envisioned (in our 2014 paper in this journal) on
which, in determining whose interests are relevant to assessing the severity of the anticipated damage (a
severity that on the contextualist approach to the PP will be what circumscribes the epistemic standards
that must be met for the ‘E’ condition to be satisfied), interests are given equal consideration. But the
egalitarian favouring rule, we argued, ultimately relinquished the advantage the contextualist was
originally said to have over the unworkable invariantist approach. The rationale here was that, given the
entrenched political interests of (for example) developers and environmental regulatory bodies, the mean
of the relevant standards for assessing the severity of the anticipated damage will typically remain fixed
across cases that vary along damage-relevant dimensions of evaluation such as: how many individuals
will be affected, how many carcinogens will be released into the air, etc. Given the contingent fact of
entrenched political interests, then, a democratic approach to adjudicating interests vis-a-vis assessing (on
a contextualist approach to the precautionary principle) the severity of the anticipated damage will,
mutatis mutandis, end up in many cases tracing the path of the egalitarian favouring rule. And, again, this
path did not end up retaining the advantage we claimed the contextualist approach had over the
invariantist. Thanks to an anonymous referee at Erkenntnis for suggestions that have improved the
presentation of this discussion.
8
It should be noted that the second puzzle can also be stated in a non-probabilistic vocabulary. See
[Carter and Peterson 2015: 10].

123
302 J. A. Carter, M. Peterson

face a choice between two fundamentally different ways of doing this. The first
option is to aggregate the first- and second-order probability functions into a single
measure of uncertainty; the second option is to refrain from aggregating the two
measures.
Steglich-Petersen agrees with us that the second strategy, to refrain from
aggregating the two measures, is implausible for the reason we state in the original
paper: if we do so it would become necessary to introduce two or more probabilistic
thresholds for when a risk is de minimis, but then there will always be cases in
which small changes to one of thresholds trigger a ‘‘too big’’ effect on the overall
assessment. (Carter and Peterson 2015: 11–12.)
But what about the first strategy? Why can’t we just aggregate the first- and
second order probabilities into a combined measure by, for instance, multiplying the
two numbers? The problem is that if the first order probability that, say, some
substance x is carcinogenic is one in a billion, and we are 90 % sure that this
probability is correct, then the combined probability that the risk is de minimis may
very well be less than one in a billion. This contradicts the gist of the precautionary
principle. Other aggregating principles face similar problems.
Instead of calculating the probability that x is carcinogenic, it is possible to avoid
the problem by ‘‘turning around’’ the probabilities and calculating the probability
that x is safe. Steglich-Petersen describes this proposal, briefly discussed in our
original paper, as follows:
the relevant all-things-considered probability [is] the probability that some
target activity is safe. This means that the lower the second-order probability,
the lower [is] the all-things-considered probability of the activity being safe.
This is the right result, since the second-order probability measures the
reliability of the first-order assessment, and a lower second-order probability
should therefore correspond to a reduced probability of safety. A risk would
then be deemed de minimis when the probability of the activity being safe is
sufficiently high, or, which is to say the same, when the probability of harm is
sufficiently low. (Steglich-Petersen 2014: 9)
What Steglich-Petersen and we disagree about is whether this response is ad hoc
or not. We believe it is, but he argues it is not. Although we concede that ‘‘turning
around’’ the probabilities would give us the result we want, we think that what is
lacking is a reason for claiming that this line of reasoning is correct and the
alternative one wrong.9

9
Before proceeding it is worth commenting briefly on Steglich-Petersen’s claim that a low second-order
probability does not always result in an increased overall risk. He gives the following example: Suppose
that I live in Baltimore and read in the weather forecast that there is a
0.8 probability of thunderstorms, but subsequently come to doubt that I have read the forecast for
Baltimore and not Chicago (perhaps I set my phone to sometimes report the one, sometimes the
other), and assign it a 0.5 probability that I have read the right forecast. What does this relatively
low second-order probability imply about my overall risk of encountering thunderstorms in
Baltimore? Nothing, it seems. The second-order probability of 0.5 favors an overall risk higher
than the first-order probability of 0.8, as much as it favors an overall risk lower than 0.8. (Steglich-
Petersen 2014: 6)

123
On the Epistemology of the Precautionary Principle: Reply… 303

Steglich-Petersen believes he has found a reason of the sort we believe to be

lacking. He correctly observes that many formulations of the precautionary
principle involve some claim about a reversed burden of proof, according to which
it is not the party who believes that an activity is dangerous who has to prove this,
but rather the party who thinks that the activity is safe. A clear example can be
found in the Wingspread statement, according to which ‘‘the proponent of an
activity, rather than the public, should bear the burden of proof’’.10 If it is the
proponent of an activity who has the burden of proof to show that an activity is safe,
then it is—according to Steglich-Petersen—not ad hoc to claim that we should
calculate the probability that an activity is safe rather than calculating the
probability that it is dangerous. We remain skeptical, for two reasons.
Firstly, although we agree that that many formulations of the precautionary
principle do actually contain some claim about a reversed burden of proof, it is
debatable whether one ought to accept this element of the principle. The mere fact
that many defenders of the precautionary principle seem to think that it is the
proponent of an activity who has the burden of proof to show that an activity is safe
does not entail that they are right, or have some non-arbitrary reason for making this
claim. No purely exegetical study of what defenders of the precautionary principle
say or do not say will help us to solve the second puzzle. What we need is a reason.
Secondly, we would like to stress that even if we were to accept Steglich-
Petersen’s claim about a reversed burden of proof, nothing follows from this about
how we are to aggregate first- and second-order probabilities. The burden-of-proof
claim is a point about who should provide information about an activity: ‘‘the
proponent of an activity, rather than the public, should bear the burden of proof’’.11
This is not a claim about how the relevant information should be aggregated.
Therefore, it will not suffice to just point out that some formulations of the
precautionary principle state that ‘‘the proponent of an activity, rather than the
public’’ has the burden of proof.12
To be more precise, even if we think it is the proponent of an activity rather than
the public who should bear the burden of proof, it seems clear that the proponent of
the activity could fulfill this obligation in at least two different ways. Firstly, the
proponent could calculate the probability that the proposed activity is safe.
Secondly, he or she could calculate the probability that the activity is dangerous.
Under normal circumstances, in which the only relevant probability is the first-order
probability, it doesn’t matter which approach is taken. If ‘‘safe’’ is the opposite of
‘‘dangerous’’, the probability that x is safe equals one minus the probability that x is

Footnote 9 continued
We agree that in this example the low second-order probability of 0.5 does not increase the overall risk.
This is because the probability of encountering a thunderstorm in Baltimore and Chicago is (to the best of
our knowledge) roughly the same. However, in nearly all cases in which the precautionary principle is
discussed, the probability that any two risks are de minimis is not the same. The overwhelming majority of
all risks that reach the public discourse are not de minimis. This is, of course, an empirical assumption–
but to be fairly uncontroversial.
10
Wingspread Conference on the Precautionary Principle, January 26, 1998.
11
Ibid.
12
Ibid.

123
304 J. A. Carter, M. Peterson

dangerous.13 However, as we pointed out in our original paper, this type of

straightforward reasoning is not applicable when we attempt to aggregate first- and
second-order probabilities into a single measure. Not very much hinges on which
aggregation mechanism we choose. The multiplicative rule is simple and fairly
attractive, but other aggregation rules will run into similar problems: the influence
of the second-order probability of the overall-conclusion will depend of whether we
calculate the probability that something is risky or safe. What we need is not a
technical solution to this problem (that would be easy to construct), but a reason for
favoring the calculation that gives the intuitively most plausible answer. The upshot
of all this is that our two puzzles remain unresolved.

References
Aven, T. (2011). On different types of uncertainties in the context of the precautionary principle. Risk
Analysis, 31(10), 1515–1525.
Carter, J. A., & Peterson, M. (2015). On the epistemology of the precautionary principle. Erkenntnis,
80(1), 1–13.
Cohen, S. (1998). Contextualist solutions to epistemological problems: Scepticism, Gettier, and the
lottery. Australasian Journal of Philosophy, 76(2), 289–306.
DeRose, K. (1992). Contextualism and knowledge attributions. Philosophy and Phenomenological
Research, 52, 913–929.
DeRose, K. (2009). The case for contextualism: Knowledge, skepticism, and context: Knowledge,
Skepticism, and context. Oxford: Oxford University Press.
Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Oxford University Press.
Manson, N. (2002). Formulating the precautionary principle. Environmental Ethics, 24(3), 263–274.
Munthe, C. (2011). The price of precaution and the ethics of risk. Berlin: Springer.
Peterson, M. (2003). Transformative decision rules. Erkenntnis, 58, 71–85.
Sandin, P. (1999). Dimensions of the precautionary principle. Human and Ecological Risk Assessment:
An International Journal, 5(5), 889–907.
Sandin, P. (2004). The precautionary principle and the concept of precaution. Environmental Values,
13(4), 461–475.
Stanley, J. (2005). Knowledge and practical interests. Oxford: Oxford University Press.
Steglich-Petersen, A. (2014). The epistemology of the precautionary principle: Two puzzles resolved.
Erkenntnis: 1–9.

13
We are aware that this is an oversimplification, but in order to keep the discussion simple we can
ignore the possibility of a third option.

123