The Status of Determinism

Jonathan Bennett

from: British Journal for the Philosophy of Science 14 (1963), pp. 106–19.

Among the principles which Kant describes as ‘regulative’ is
the determinist principle that every empirical state of affairs
follows in accordance with causal laws from some earlier
state of affairs. In calling this ‘ regulative’, Kant is making a
five-fold claim.

(1) The principle cannot be conclusively proved. Because
the future is or may be endless, it is always possible that
new states of affairs are going to come about in ways which
do not accord with the principle.

(2) The principle cannot be conclusively disproved. Be-
cause logical space is limitless, it is always possible, however
persistent our failure to find a causal explanation for a state
of affairs, that there is an explanation which we have not yet
found.

(3) Unlike some statements which can be neither settled
nor ruled out by experience, the determinist principle is at
least relevant to experience. Although we cannot reach the
goal of showing it to be true, or that of showing it to be false,
we can and do take small steps towards each goal. We move
towards verifying the principle whenever we explain a state of
affairs by means of a well-tested and unrefuted hypothesis;
and we move towards falsifying it whenever we try but fail
to find such an explanation. This sense of direction marks

off the determinist principle sharply from such unverifiable
and unfalsifiable statements as that all change results from
the exercise of the divine will, or that there are intangible,
invisible, and impotent goblins.

(4) Sense can therefore be given to the notion of assenting
to the determinist principle. For it can be construed as a
piece of advice; urging us always to move as far as possible in
the direction of verifying the principle, and never to despair
of moving further. Assent to the principle is just acceptance
of this advice.

(5) The principle gives good advice to anyone who wishes
to understand the empirical world, for it advises him to
pursue such understanding optimistically and therefore
energetically; furthermore, precisely because the determinist
principle cannot be falsified, it constitutes good advice with
no risks attached. Advising someone to assume there is an
explanation for every state of affairs is not like advising to
assume that there is gold in every river-bed. Acceptance of
that advice might engender optimism and therefore industry,
but it would also certainly lead to error, for some river-beds
are known contain no gold.

I want now to hold in one hand Kant’s treatment of the
determinist principle as ‘ regulative’, while holding in the



The Status of Determinism Jonathan Bennett

other his claim that he has proved the so-called ‘second
analogy of experience’ which says: ‘All alterations take place
in conformity with the law of the connection of cause and
effect’ (B 232). He makes it abundantly clear that he does
not take this to mean merely that it is safe and sensible
to say that all alterations take place. . . etc. The claim of
the second analogy of experience is that to say there are
uncaused events is to say something demonstrably false;
and this is in conflict with the description of the determinist
principle as ‘ regulative’, for that description implies that to
say there are uncaused events is to say something which
is objectionable because it may induce despondency in the
scientific enquirer, but which cannot be proved to be false.

In fact, Kant does not succeed in proving the second
analogy of experience; but he does (in my view) succeed in
proving the more limited conclusion that there could not be
intellectually graspable experience which was not experience
of a world which manifested a fairly high degree of causal
order. (I cannot present his proof of this here: it includes,
among much else, all that is valid in Wittgenstein’s argument
against the possibility of a private language.) This limited
conclusion implies that there is no possibility of our finding
that the world has become altogether resistant to science; but
it leaves open the possibility that some causally inexplicable
states of affairs should come about, and so it is consistent
with claiming a purely regulative virtue for the statement
that there is a causal explanation to be found every state of
affairs.

There is, then, a double problem about Kant’s handling of
the second analogy. Why does he fail to see the gap between
the weaker version which he does prove and the stronger
version which he claims to have proved? And why does
he fail to see that the claim to have proved the stronger
version is inconsistent with the attribution, a few hundred

pages further on in the Critique, of a regulative status to
the determinist principle? This is a problem about Kant
rather than about philosophy, but I hope to show that it
shares with many other problems about Kant the merit that
in discussing it we can learn something in philosophy. The
real objection to the protective Kantian commentators whose
main concern is to defend Kant from charges of error and
inconsistency is that they take a real, human, confused,
fertile philosopher from whom much can be learned and
present him as a mythical, superhuman, infallible, sterile
bore. For example, G. Bird (Kant’s Theory of Knowledge,
p. 71) has rebutted the accusation that Kant tells two dif-
ferent and mutually inconsistent stories about the status of
the determinist principle, on the grounds that ‘Kant ‘takes
the trouble to distinguish between’ the determinist principle
which he calls ‘regulative’ and that of the second analogy
for which he claims a more than regulative status. This
might be plausible if it were accompanied by an account of
what these two principles are, and of how two principles, of
which one is regulative and the other not, can both express.
some sort of determinism. But Bird tells us nothing about
this. Furthermore, his only justification for saying that Kant
‘takes the trouble to distinguish’ the determinism which is
called regulative from the determinism of the second analogy
which is not called regulative is that Kant says things about
the one which are inconsistent with things he says about the
other. But this can be taken as the making of a distinction
only if one takes it as literally axiomatic that Kant never
contradicts himself. I find this axiom unacceptable, but my
main objection to this smoothing-over approach is that it
teaches us nothing about what is true in philosophy.

I hope it will not prove too exasperating if I set about
solving (and learning from) this problem about what Kant
is up to by raising first the different but related problem of

2



The Status of Determinism Jonathan Bennett

what he is up to when he says that there is a radical error
in Hume’s analysis of the concept of cause. Of a group of
concepts which includes that of cause, Kant says (A 112):

All attempts to derive these. . . concepts. . . from expe-
rience, and so to ascribe to them a merely empirical
origin, are entirely vain and useless. I need not insist
upon the fact that, for instance, the concept of a cause
involves the character of necessity, which no experi-
ence can yield. Experience does indeed show that one
appearance customarily follows upon another, but not
that this sequence is necessary. . .

Kant denies that the concept of cause can be derived from
experience because derivation requires thought, and one can
think about one’s experience only if one already has a use
for the concept of cause. But his concern here is less with
the Humean account of how we acquire the concept of cause
than with the Humean analysis of this concept. Because ‘the
concept of cause involves the character of necessity, which
no experience can yield’, no analysis of the concept in purely
experiential terms can be correct.

Some writers have tried to fault Hume’s analysis by
producing statements which, they say, Hume is committed
to calling causal laws but which we know perfectly well not
to be causal laws: every time the 5 o’clock whistle blows at
Dagenham, the workers at Coventry their tools and go home.
Such counter-examples may tell against specific empiricist
analyses of causality, but they cannot show the wrongness
of the empiricist programme for such an analysis. If we know
perfectly well that this whistle’s blowing does not cause those
workers to down tools, the empiricist has only to ask how
we know this, and to amend his analysis in the light of the
answer. Kant makes mistakes, but he does not make the
mistake of trying to show, by the production of empirical
counter-examples to specific empiricist analyses, that no

empiricist analysis can succeed. His position is accurately
stated by Körner (Kant, p. 74): ‘While Kant insists that
the notion of causality is not equivalent to that of regular
succession, he holds that unless it implied this notion it
could not refer to anything in perception.’ But although Kant
thus seems to claim the true notion causal necessity to be
narrower than Hume says it is, he rightly refrains from any
attempt to prove or illustrate the over-generosity of Hume’s
concept by pointing to universal statements which Hume
would have to call causal but which are obviously not so.

Just because Kant does not attack Hume in this mis-
guided way, there a problem over what attack he does think
he is launching. The solution to this problem is twofold.
One part of it is well known to commentators; in respect of
the other part—which alone is relevant to regulativeness—I
claim a certain mild originality.

The first way in which Kant thinks that Hume cannot do
justice to the necessity of causes brings us back to the part
of the second analogy which Kant does succeed in proving.
Hume presents us as discovering moment by moment that
our world continues to be fairly orderly, and gives the
impression that the onset of total chaos is a possibility which
might at any moment be realised in our experience; Kant, on
the other hand, knows that causal order is not dispensable
to this extent. It is often said that the burden of Kant’s
complaint against Hume is just this: Kant is accusing Hume
of implicitly denying that causal laws are necessary in the
sense that there must be (known) causal laws if there is to
be graspable experience at all. But this is different from
saying that causal laws are necessary in the sense that
every causal law expresses or involves some non-empirical
element of necessity. What looks like a complaint against
Hume’s analysis of the concept of cause is here presented as
a complaint which is compatible with any analysis, namely

3



The Status of Determinism Jonathan Bennett

that Hume has exaggerated the intellectual dispensability of
causal order. Kant does sometimes seem to conflate these
two complaints, but that is no excuse for his commentators’
doing so. To take an example from one of the best of
them, W. H. Walsh (Reason and Experience, p. 153) says:
‘Kant is emphatically not saying that some kind of inner
necessity binds cause to effect. . . The necessity which marks
the causal relation is. . . derived wholly from the necessity of
the principle of causality itself.’ This says that Kant brings
only one charge against Hume, namely that concerning the
dispensability of causal order; yet Walsh expresses this in
terms of the phrase ‘the necessity which marks the causal
relation’, which is appropriate only to the other anti-Humean
charge which Walsh explicitly (and rightly) denies Kant to be
making.

We could leave it at that, and conclude that Kant has
really nothing to say about Hume’s analysis of cause but
only about his failure to see how far our ability to grasp and
be aware of our experience depends upon the obedience of
experience to causal laws. I was myself inclined to leave it at
that, until I began to pay close attention to a passage (A 91)
in the Critique from which emerges a new line of thought
altogether. I number its sentences for purposes of reference;
and, since I am taking the passage rather seriously, I should
mention that the translation is Kemp Smith’s and—as almost
always with this translator—is beyond reproach:

(1) If we [said] that experience continually presents
examples of . . . regularity among appearances and
so affords abundant opportunity of abstracting the
concept of cause, and at the same time of verifying the
objective validity of such a concept, we should be over-
looking the fact that the concept of cause can never
arise in this manner. . . (2) For this concept makes
strict demand that something, A, should be such

that something else, B, follows from it necessarily
and in accordance with an absolutely universal rule.
(3) Appearances do indeed present cases from which
a rule can be obtained according to which something
usually happens, but they never prove the sequence to
be necessary. (4) To the synthesis of cause and effect
there belongs a dignity which cannot be empirically
expressed, namely, that the effect not only succeeds
upon the cause, but that it is posited through it
and arises out of it. (5) This strict universality of
the rule is never a characteristic of empirical rules;
they can acquire through induction only comparative
universality. . .

The first two sentences seem to voice the old accusation—for
which I have sought and failed to find a justification in
Kant’s pages—that Hume’s kind of analysis omits some
essential element of necessity from the concept of cause.
To take these sentences in this way would involve taking
the stressed phrase ‘ necessarily and in accordance with an
absolutely universal rule’ to mean ‘ not only in accordance
with a universal rule (for which Hume can allow) but also
necessarily (for which Hume cannot allow)’. But in sentence
(3) there is an indication that Kant believes that there is
also an issue about universality between himself and Hume,
for he there seems to accuse Hume of fobbing us off with
statements about regularities which ‘usually’ hold. Also,
in sentence (3) there is a suggestion that the issue over
universality is the issue over necessity; for Kant seems to
treat the idea of a regularity which usually holds as though it
were the obvious antithesis of the idea of a regularity which
necessarily holds, thus suggesting that the only strictly true
universal statements are those which express some sort of
necessity. This suggestion is reinforced by the transition
from sentence (4) to sentence (5). The subject-matter of

4



The Status of Determinism Jonathan Bennett

(4) is the idea of something’s not just succeeding upon but
arising out of something else, which Kant clearly thinks to
involve the necessity which escapes Hume. Sentence (5)
purports to continue discussing this notion of necessity, yet
its opening words are ‘This strict universality of the rule is
never a characteristic of empirical rules. . . ’

[Added in 2012: When I wrote that Kemp Smith translates this passage
perfectly, I must have been guessing and intending to check the guess be-
fore publishing. Careless! In fact there’s a serious flaw in the translation.
The start of (5):

’This strict universality of the rule is never. . . ’
is a mistranslation of

Die strenge Allgemeinheit der Regel ist auch gar keine. . .
which means

‘Strict universality of the rule is also never. . . ’.

By adding ‘This’ and suppressing auch = also, Kemp Smith presents two

wrong indications that Kant is identifying necessity with strict univer-

sality. This seems to be a case where the commentator borrowed the

translator’s pen.]
If Kant is saying that the only true universal statements

are those which are in some way necessary, then he is wrong.
If he is saying more modestly that until we know that a
universal statement is necessary we cannot know that it is
true, then he is still in trouble because, on his own showing,
the necessity in question cannot be discovered to obtain in
any particular case and therefore it cannot be used as a
guarantee of truth or of anything else. There is, however, a
different way of associating Hume’s position with a certain
kind of permissiveness about ‘strict universality’; and I am
confident that it is part of what is expressed in the passage
I have been discussing as well as in some others, though
Kant seems not to have isolated and crystallised it in his
own mind sufficiently to be able to spell it out clearly, or to
see how misleading is its formulation in terms of ‘necessity’.
The point is as follows:

While Hume thinks of causal laws as universal statements
which have no exceptions at all, his account of the sort of
thing a causal law is and of the ways in which causal laws are
useful to us is such as to allow that we could make almost as
much use of statements which were like causal laws except
that they were known to have just a few counter-instances.
He shares the common belief that when we are inclined to
accord a causal status to a lawlike statement which then lets
us down by proving false in a single instance, the statement
ought to be dropped entirely because (a) it has, or could be
made to have, an unlimited number of counter-instances,
and (b) the facts which it covers are also covered by some
other (still unformulated) universal statement which has
no counter-instances at all. But just suppose that we
believed—for no matter what reason: perhaps because God
told us—that the nearest we could come to exceptionless
universal statements was to establish universal statements
which were true in almost every instance, what should we say
then? To put it in another way, what should we say about the
possibility that we might have to settle for a weakly quantified
science—i.e. one whose hypotheses were all stated in terms
of a quantifier which had the force of ‘ For very nearly every
value of x. . . ’? (Like Kant and Hume, I say nothing about
hypotheses which say that all but a certain small proportion
of the members of a certain class belong to a certain other
class. That raises entirely different issues, on which Kant
said nothing because they never occurred to him.)

Hume would answer that such a science could be useful
in the same kind of way, and to nearly the same extent, as
a science consisting of genuinely universal statements. He
would have to concede a modest loss of utility: predictions
would be made a little less confidently, and there might
also be some attenuation of our sense of having explained,
or understood, an event by relating it to earlier events in

5



The Status of Determinism Jonathan Bennett

accordance with the statements of our science. But Hume
has no reason for denying that predictions and explanations
in terms of a weakly quantified science would still be genuine
predictions and explanations. For him, all this must be a
matter of degree: if God tells us that the best laws we can find
will have exactly fifty known exceptions each, then things
are moderately bad; and if He tells us that the best laws we
can find will have exactly one known exception each, then
things are only one-fiftieth as bad. Strict universality is the
ideal, but only as the maximum of something which it is
desirable to have in as high a degree as possible.

In contrast with this, it is Kant’s view that there is
a great gulf between exceptionless universal statements
and universal statements which have one counter-instance
each; but that the difference between ‘one counter-instance’
and ‘two counter-instances’ or ‘one hundred and seventeen
counter-instances’ is of no theoretical importance at all,
because the real damage is done by the admission of a single
counter instance. This is because an ‘explanation’ of an
event which appeals to a statement of the form ‘Very nearly
every time A occurs, B follows’ is not an explanation at all.
The production of such a statement might soothe someone
into saying ‘You have explained the event to me’ or ‘Now I
know why it occurred’ or ‘Now I understand’; but the event
would not in fact have been explained; the hearer would not
know why; he would not understand. It is of the essence of
the notions of explanation, of understanding, of something’s
being so because something else is so, of an event’s being
made to occur, that these are all expressible in terms of ‘an
absolutely universal rule’. The flaw in Hume’s position is
that he cannot do justice to this fact. Hume’s only reason
for connecting ‘explain’, ‘understand’, ‘because’ and the rest
with exceptionless universal statements is his contingent
belief that there are exceptionless rules to be found and

that any rule which has one exception can be made to have
indefinitely many exceptions. Just because he thinks that
the concept of causality can be analysed in purely empirical
terms—and crucially in terms of regularity—Hume does not
and cannot attach a fundamental philosophical importance
to the difference between a rule which always holds and one
which nearly always holds.

I wish now to bring out two connections between the
notion of regulativeness and Kant’s view that genuine expla-
nations demand a strongly quantified science. One of them
involves using Kant’s view about strong quantification to
clear up a difficulty about his handling of regulativeness; the
other reverses the order and uses the notion of regulativeness
as an aid to adjudicating between Kant and Hume in their
conflict over strong quantification.

Firstly, let us return to the question of why, having
proved that all graspable experience must have a fairly high
degree of causal order, Kant claims to have proved that total
determinism is strictly true.

There are two ways in which it is prima facie possible
that a fairly orderly world should fall short of being totally
causally determined. (i) It might manifest a causal order
which is fully expressible in strongly quantified statements
which do not claim all the territory—that is, which give no
answers at all to some questions about what, in given fully
described situations, will happen next. (ii) It might manifest
a causal order which is expressible in statements which give
answers to every question about what will happen next, but
only weakly quantified statements of this sort. If we think
only in terms of strongly quantified science, then these two
prima facie possibilities are the possibilities (i) that a world
might admit only of an uncomprehensive science and (ii) that
a world might admit only of a slightly inaccurate science.

6



The Status of Determinism Jonathan Bennett

I have no reason to think that the first of these ever
occurred to Kant as a possibility, but in the present context
this oversight is not a damaging one. There could of course
be an accurate but uncomprehensive science; but I doubt if
there could be a partially ordered world the whole of whose
order was expressible in an accurate, strongly quantified
science—so that events which were not predictable by this
science could not be brought under any generalisations at
all, even weakly quantified ones. If I am right in thinking
that this is not possible, then any world which was not
totally determined would have to owe at least part of the
incompleteness of its causal order to the second of the
two possibilities, that is, to its admitting only of a slightly
inaccurate science or a weakly quantified one.

Now, Kant’s arguments for the second analogy all turn
on such notions as those of explanation, retrodiction, rea-
sons for believing, the way in which something must have
happened, and so on. What his arguments show is that
we cannot have graspable experience at all unless we have
experience to which we can apply notions like these. But his
complaints against Hume include (I have argued) the claim
that we are not using these notions properly unless we are
using them in connection with a true, strongly quantified
science; and I have suggested that a fairly orderly world can
escape being totally determined only if its science is either
not-quite-true or weakly quantified. And this, I submit, is
why Kant thinks that there is no gap between the notion
of a world which is orderly enough for the notions of expla-
nation etc. to be applicable to it and that of a world which
is totally determined; it therefore provides the reason for
Kant’s exaggerated description of what he has achieved in
his arguments for the second analogy of experience. Qed.

By speaking of ‘exaggeration’, I commit myself to taking
Hume’s side against Kant on the issue of weakly versus

strongly quantified science. I believe that most people find
Kant’s view here the initially attractive and plausible one,
and I hope that this is so because I do not wish to take
up space here trying to increase its attractiveness before
showing it to be false.

Let K be a scientist whose working assumption about
what kinds of causal laws can be found is the strictly
determinist one which Kant thinks to be a sine qua non
of scientific respectability. Let H be another scientist whose
working assumption is that the world admits of a weakly,
but not of a strongly, quantified science. I wish to examine
the ways in which their respective assumptions affect the
behaviour of K and H as scientists, in order to see whether
Kant is right in withholding from H the cachet of intellectual
respectability which he confers upon K.

H thinks that some events are causally inexplicable, while
K denies this. Suppose them to be confronted by an event
for whose occurrence they have no explanation. It is not an
event upon which received science is silent, but one whose
occurrence is in conflict with some hypothesis which has
hitherto been a part of received science. A short description
of what then happens would be as follows: K says that the
hypothesis in question must be dropped and a replacement
for it sought, and he maintains this in face of persistent
failure to replace the hypothesis; H on the other hand
regards it as possible that the challenged hypothesis is the
best hypothesis of its kind of which the world admits, and
therefore that the event whose occurrence conflicts with
it is simply an illegality which cannot be coped with in
science at all.

K stands a chance of discomfiting H—by finding an
explanation for the event which H has conjectured to be
inexplicable—while H has no comparable way of discomfiting
K. Those of us who were brought up to think of vulnerability

7



The Status of Determinism Jonathan Bennett

as a virtue may well regard this as a point in favour of H’s
position. But there is a reason which might be given for
awarding the honours the other way, namely that although
H and K both make acts of faith, it is K who makes the
optimistic, industry-urging act of faith. To give this reason
for favouring K’s strict determinism is to appeal to its alleged
merit as salutary advice, that is to its alleged acceptability
as a regulative principle. More than this is needed to justify
Kant’s position, for he says that a belief in illegalities is not
just mildly bad for the scientist who holds it but is radically
incompatible with the rational understanding of the world.
Nevertheless, I wish to discuss the weaker claim for the
regulative virtue of strict determinism, as a discussion of
this will give us all we need for a refutation of Kant’s stronger
claim that strict determinism is not only salutary for the
scientist but is mandatory upon him.

The suggestion before us, then, is that when K insists
that there is an explanation to be found for the as-yet-
unexplained event, he is saying something which is healthier,
because less conducive to complacency, than H’s conjecture
that the event is an illegality. The short reply to this is
that it is a libel on H to assume that he is not prepared to
test his conjecture; and to test it he will to have do exactly
what K does, namely to search as hard as he can for an
explanation for the event in question. At any level of science,
the behaviour of the honest friends of an hypothesis is the
same as that of its earnest enemies.

‘Still,’ it may be said, ‘H will eventually give up the search
for an explanation. He lacks K’s ultimate commitment to a
belief in the total intelligibility of the world, and this failure of
faith will sooner or later sap his will and lead him to abandon
the pursuit.’ This is correct; but it is not a difference between
H and K.

It must be remembered that the problem concerns a
single event, not a large class of events of a certain kind.
H and K have adopted a hypothesis about (say) sulphates
and then discovered not that it fails for (all samples of)
copper sulphate, but that it failed for one particular sample
of copper sulphate on one particular Wednesday morning. If
their difficulty arises from an experiment which can be
reproduced at will, and always with the same result, then
H and K do have a pair of strictly universal statements
which cover the known facts: one about sulphates other
than copper-sulphate-under-conditions-C, and one about
copper-sulphate-under-conditions-C. Much time may be
spent in trying to derive both hypotheses from some more
general hypothesis; but even if they are not as general as H
and K might wish they are nevertheless ‘strictly universal’
and so suffice to ward off our Kantian problem. Our problem
arises only when a hypothesis is challenged by a single
datable, locatable, unrepeatable experiment, not when it is
challenged by a kind of experiment of which there are many
instances.

There is, I think, universal agreement amongst scientists
and philosophers of science. that unrepeatable experiments
do not count against well-corroborated hypotheses. So, when
H—faced with the stubborn failure of his attempts to repeat
the experiment—says that the challenged hypothesis ought
to be retained and the challenging experiment relegated to
limbo, he is only saying what K too will say if he does not
wish to make an ass of himself in the eyes of the scientific
community, The only difference between them is that H
can give a better account than K can of why he refuses to
give up a hypothesis which is challenged by an unrepeatable
experiment. For H can say that he is acting on the conjecture
that the experiment was an illegality and that the challenged
hypothesis has that ‘true with very few exceptions’ status

8



The Status of Determinism Jonathan Bennett

which is the most that we can hope for in science. K of
course cannot say this; but what can he say?

He may say that he does regard the hypothesis as refuted,
but that he continues to treat it as though it were unrefuted
because he has nothing to put in its place. But if he says this,
he will find himself estranged from the rest of the scientific
community: not because he is so lax as to treat a hypothesis
as true which he believes to have been refuted, but because
he is so silly as to believe a hypothesis to have been refuted
by an unrepeatable experiment.

I shall suppose him, then, to say whatever the other
scientists and the philosophers of science say about the
negligibility of unrepeatable experiments. The trouble is that
most of them say nothing; and those who do pronounce
on the subject are quite unclear about the particular point
which I am raising. However, if we can see clearly what
are the possible positions on this matter, we can afford to
ignore the question of which philosophers have in fact taken
which position. Stripping the issue to its bare essentials,
we can say that if a well-corroborated hypothesis of the
form ‘(∀x) (Fx ⊃ Gx)’ is challenged by the production of an
experimental report of the form ‘Fa & Ha & ¬Ga’, and all
attempts to repeat the experiment described therein meet
with failure, there are just three possibilities: (1) To say that
the hypothesis is refuted by the single experiment; (2) To
say that the experimental report is not true; and (3) To say
that the experimental report is true but the hypothesis is
unrefuted—this is what H says, and it commits anyone who
says it to reading the hypothesis as weakly quantified.

Option (1) is eccentric and uninteresting, and I shall
discuss it no further; the important and interesting clash is
that between options (2) and (3). I do not think it has been
generally recognised that the orthodox view about unrepeat-
able experiments confronts us with the disjunction of (2) and

(3), but there are signs in the literature that this has been
half-recognised: some writers speak of unrepeatable experi-
ments in terms appropriate to the adoption of (2), and the
only reason I can think of for adopting (2) as necessarily the
right account is a Kantian belief that to adopt (3) is in some
way to let the side down. The trouble with option (2)—that is,
with saying that the experimental report is untrue in some
relevant respect—is that there could be cases in which it was
entirely implausible. It does not matter whether there have
been such cases: their mere possibility is sufficient to show
that the insistence upon a strongly quantified science, which
debars one from ever taking option (3) when dealing with an
unrepeatable experiment, is irrational and unsatisfactory.

In saying that there could be cases in respect of which
option (2) was implausible, I am not calling attention merely
to those cases where we have no reason to suspect the
integrity and basic professional competence of the author
of the experimental report. A report may be deemed untrue
without the honesty or basic competence of its author being
called into question, if the doubts do not concern those
parts of the report which say what apparatus was used,
what readings were taken, what kind of graft was made,
etc., but rather those parts which say that the sample used
was pure, that the meteorological conditions did not vary
during the experiment, that adequate safeguards were taken
to prevent cross-pollination. Claims of this sort will always
be needed if the reported experiment is to be made relevant
to a received hypothesis, and a doubt which is cast upon any
such claim in an experimental report is a doubt as to the
truth of the report in so far as the report says that something
happened which was in conflict with a received hypothesis.
It is in connection with this sort of doubt, I think, that it has
become customary to dismiss unrepeatable experiments as
involving some sort of ‘experimental error’.

9



The Status of Determinism Jonathan Bennett

But it is just dogmatism to say that wherever an experi-
ment turns out to be unrepeatable it is ipso facto reasonable
to say that it involved some ‘experimental error’ of this sort.
If the claim that there was an experimental error has no
backing except for the fact that the experiment has turned
out to be unrepeatable, then the imputation of experimental
error is just a face-saving, word-spinning device which has
much less to recommend it than has the alternative of
saying that the hypothesis under challenge must be taken
as weakly quantified. The impression is sometimes given
in the literature that it is always reasonable to take option
(2) in connection with an unrepeatable experiment, because
repeatability is a necessary condition of the experiment’s
counting as an ‘objective’ happening, or of its not counting
as ‘occult’ or ‘chimerical’. The picture which this evokes
is that of a scientist ending his experimental report with
something like ‘. . . and then I saw a blue flash; no-one else
was in the laboratory at the time, but I swear that I saw it’.
But of course the circumstances of the experiment do not
have to be like this at all: there can be witnesses, research
assistants, photographs, recorded running commentaries—a
whole host of sources of evidence that the experiment did
take place as reported. In these circumstances, it would
be merely arbitrary to insist that nevertheless there was
an ‘ experimental error’ in the relevant sense, or that what
happened was ‘ chimerical’; but it would not be arbitrary to
dismiss the experiment as negligible on the grounds of its
unrepeatability. In such a case, the only person who could

give a sensible account of why the experiment was negligible
would be the one who—like my scientist H—does not insist
upon a strongly quantified science.

In conclusion, then, I maintain that a preparedness to
accept a weakly quantified science is not only permissible
but is mandatory upon any scientist who wishes to be able
to cope sensibly with a really well-attested but unrepeatable
experiment, if one should occur. Kant’s insistence upon
‘absolutely universal rules’ is not just the contradictory, but
a contrary, of the truth.

* * * * * *

[Added in 2012:]
At a meeting of the British Society for the Philosophy of

Science where I presented this paper its conclusion was met
with scepticism unsupported by arguments. As evidence that
the conclusion isn’t merely weird, I remarked that Popper at
least agreed with it. This produced a storm of disagreement;
but Imre Lakatos produced a copy of The Logic of Scientific
Discovery, in section 22 of which this is said:

We say that a theory is falsified only if we have
accepted basic statements which contradict it. This
condition is necessary, but not sufficient; for we have
seen that non-reproducible single occurrences are of
no significance to science. Thus a few stray basic
statements contradicting a theory will hardly induce
us to reject it as falsified.

10