DEDUCTIVE UNIFICATION AND IDEALISATION*

S. K6RNER

APART from a sprinkling of Hegelians, modem philosophers of science

are on the whole either moderate Humeans or moderate Kantians.
The moderation of the Humeans shows itself in their rejection of
Hume's doctrine that all belief can be reduced to impressions and ideas ;
that of the Kantians in the rejection of transcendental arguments to the
effect that a categorial scheme, which is in fact accepted by somebody,
must—in an interesting sense of ' must'—be accepted by everybody.
The Humeans and Kantians of today agree with each other and their
orthodox predecessors that theoretical reason has no constitutive func-
tion or, in more modern terminology, that the deductive unification
of a field of empirical enquiry is merely an ordering, and not a modi-
fication, of the concepts and propositions describing this field.
I shall try to show that on the contrary the deductive unification of
a field of experience ipso facto involves the replacement of empirical
predicates and propositions by non-empirical ones, and is thus not a
mere ordering but an idealisation. The talk is divided into three parts.
First, the classical account of the function of hypothetico-deductive
systems will be criticised. Next, some logical questions, which arise
from the criticism, will be considered. In conclusion, I shall outline
an alternative account of the function of hypothetico-deductive
theories and some of its implications for the understanding of the rela-
tions between science and extra-scientific thinking.

The classical account of scientific prediction and explanation by

means of a hypothetico-deductive theory T, e.g. classical mechanics, is
very briefly this : Starting with a basic proposition, say bu which
describes what has been found by experiment or observation, one
deduces from bx and T another basic proposition bit which can again
be compared with experiment or observation. Schematically,
(&x and T) h b+
* This paper was read to the British Society for the Philosophy of Science 20.V.63.
274
 DEDUCTIVE UNIFICATION AND IDEALISATION
In this formula ' ^- ' expresses deducibility with respect to an underlying
logic L, comprising as a rule the two-valued logic of propositions,
quantification theory, and the theory of equality, and often extended
so as to accommodate the theory of real numbers.
The predicates of T are empirical or non-empirical (' theoretical').
The constituent predicates of the basic propositions are included in the
empirical predicates of T. There exist different variants of the classical
account, according to the way in which the basic propositions are dis-
tinguished from other empirical propositions, and the way in which
empirical and non-empirical predicates and propositions are distin-
guished from each other. There is no need to discuss these differences
here.
From the point of view of a working scientist and a philosopher
of science who is interested only in the structure of scientific theories,
and not also in their relations to extra-scientific thinking, the classical
account has many merits. However, from the point of view of a
philosophy which aims at an understanding of these relations— one
would almost say, from the point of view of a philosophical philosophy
of science—it is open to objections. These are all connected with re-
strictions imposed by the underlying logic of scientific theories upon
their predicates and propositions. Recognition of the restrictions
counteracts illegitimate tendencies to extrapolate the theories beyond
the range of their applicability.
A hypothetico-deductive theory can, as is well known, be regarded
as an extension of its underlying logic. This extension consists in
adding to the logical terms and predicates the substantive terms and
predicates of the field of empirical enquiry which is unified by the
theory ; and in adding to the logical postulates the substantive postu-
lates which govern the use of the substantive terms and predicates.
The possible logical relations between propositions are not affected by
the extension. Thus adjunction, negation, class-inclusion, etc., are
determined by the underlying logic.
The first, and most obvious, restriction imposed by L on T concerns
theories the underlying logic of which is sufficiently rich to contain the
classical mathematics of real numbers. If the basic propositions are
about any kind of continuous expanse or transition, they will be formu-
lated in terms of a mathematical notion of continuity, conforming to
Dedekind's or some similar mathematical postulates. These include,
among others, the so-called postulate of density according to which
between any two members of a continuous series there exists at least
275
 S. KdRNER
one further member and therefore an infinite number of members.
The class of these members has to be thought of as a given totality
which as follows from the postulates of the theory, is non-denumerable.
It is clear that the mathematical notion of continuity is an ideali-
sation of an empirical notion. I have argued elsewhere1 that the notion
of empirical continuity is neither, as is sometimes asserted, unanalysable
nor self-contradictory, but that it can be analysed in terms of a finite
number of inexact classes. The classical logic which recognises only
exact classes, which can in no way be regarded as merging into each
other, counterbalances, as it were, this idealisation by the further ideali-
sing assumption of non-denumerable infinite totalities. Without going
into a more detailed analysis of either empirically or mathematically
continuous series, the latter are certainly not descriptions, but at best,
idealisations of what is observed.
The second, and hardly less obvious, restriction imposed by L on T
concerns an even wider class of theories, namely all those, the under-
lying logic of which contains the theory of equality. The difference
between the logico-mathematical and the empirical notion of equality
has often been pointed out. To avoid misunderstandings, it might be
better to speak of perceptual or operational indistinguishabihty instead
of equality. Poincare" and many others after him have shown that,
whereas mathematical equality is transitive, perceptual equality is not.
To give an often quoted example : If we consider empirical equality
or indistinguishability with respect to weight—weight being deter-
mined by a pair of scales—then, since no pair of scales is infinitely sensi-
tive, it is always possible to find three objects A, B, C, such that A
balances B, B balances C, but A does not balance C.
Similar considerations apply to empirical equality in length and in
temporal duration. All these notions are non-transitive and the empir-
ical propositions involving them are on incorporation into a theory, the
underlying logic of which contains the theory of equality, idealised into
propositions involving transitive relations of equality. The ideali-
sation affects not only the fundamental quantities of physics but clearly
also the derived quantities, which are defined in terms of them. In-
deed, if in a very wide sense of the term we call ' quantitative' all
theories embedded in a logic with a transitive equality-relation, then
all quantitative theories contain basic propositions which are, strictly
speaking, idealisations of experience.

1
Monist, 1962, 47, 1-19
276
 DEDUCTIVE UNIFICATION AND IDEALISATION
I now turn to an idealisation which is imposed on all theories, even
those that are not quantitative, by the structure of their underlying
logic. About this logic I assume only that it comprises the two-valued
logic of propositions and the predicate logic of first-order or quanti-
fication-theory. In order to clear the ground I shall first of all very
briefly consider the manner in which experience of the world around
us is differentiated into individuals and classes. Many schemata of em-
pirical differentiation, i.e. of individuation and classification, are con-
ceivable and there is no reason to suppose that one of them is logically
prior to any other. I should be prepared to argue that arguments to
the contrary (so far given and known to me) rest either on a petitio
prindpii or on a drculus vitiosus the radius of which is usually too small
to make it either interesting or excusable.
The most familiar, and perhaps the oldest, schema of empirical
differentiation divides the world around us into individual things
moving, or capable of being moved, in empty space and time. Others
are the schema of things in fields and the schema of thingless fields in
which changing patterns are discerned. One cannot predict which as
yet unconceived schemata of empirical differentiation will arise from
the ever-changing language of ordinary discourse, from science, meta-
physical speculation and their interaction.
As far as these schemata are concerned—and I can see no reason
why what holds of the known ones should not also hold for those of
the future—the definition of empirical predicates and the classes corres-
ponding to them, depends on the discernment of similarities and dis-
similarities. Their definition in terms of other empirical predicates
requires in the end the exhibition of examples of a predicate's applic-
ability and inapplicability, i.e. of standard members and non-members
of the corresponding classes. This is to say that among the qualifi-
cations and disqualifications of an individual for falling under a predi-
cate, or for its membership of the corresponding class, there is the
requirement that it should be similar to standard members and dis-
similar from standard non-members—there being a method available
for producing standard objects. I shall call any predicate or class which
is characterised in the manner mentioned a ' resemblance-predicate'
or ' resemblance-class'.
The way in which resemblance-predicates and classes are defined,
or introduced into language, accounts for their inexactness, i.e. their
admitting of borderline cases. Apart from having standard members
and non-members, and ordinary members and non-members (i.e.
277
 S. KORNER
objects which are sufficiently similar to standard members and dis-
similar from standard non-members to qualify for membership ; and
objects which on the basis of the comparison with standard members
and non-members are clearly disqualified for membership), a resem-
blance-class also admits of neutral candidates for membership of the
class, i.e. of objects which can with equal correctness be assigned or
refused membership of it.
Just as some empirical predicates and classes are inexact, so some
empirical individuals are indefinite in the sense that whereas some in-
dividuals are clearly their parts and others clearly not, there are yet
other individuals which with equal correctness can be judged to be or
not to be their parts. This may be so, whether the parts in question are
spatial or temporal. If examples are needed, one might instance the
gradual separation of one or more parts from a complex individual, or
the difficulties of determining the beginning or end of an individual's,
especially an animal's, history.
It may be worth emphasising that an individual's indefiniteness in
some aspect or phase is compatible with its definiteness elsewhere.
Some individuals are as clearly parts or non-parts of others, as they are
members or non-members of resemblance-classes. Although for
some purposes it is useful to treat the indefiniteness of individuals
independently of the inexactness of predicates, it should be noted
that the former notion can be explained in terms of the latter.
The empirical predicate ' x is a spatio-temporal part of y ' is clearly
a resemblance-predicate ; and every neutral instance of this pre-
dicate is an ordered couple, the first member of which can with equal
correctness be judged to be or not to be a part of the second. I
shall, therefore, say no more of the indefiniteness of empirical in-
dividuals.
After these remarks it is not difficult to see that every hypothetico-
deductive theory, which unifies a field of experience described in terms
of resemblance-predicates, and is embedded in the classical two-valued
logic, enforces the replacement of these predicates by idealisations of
them. Whereas in the classical logic, if O is an individual and P is a
predicate, only two kinds of predication are possible namely, (i) oeP, i.e.
' o is an instance of P' and (ii) oe'P, i.e.' o is not an instance of P ', resem-
blance-predicates admit also of a third possibility, namely (say) (iii)
0€*P, i.e. ' o is a neutral candidate for P'. To these three analysed
propositions, there correspond three unanalysed atomic propositions,
namely (i) p, i.e. ' p is true ', (ii) 7p, i.e. ' p is false ' and (iii) *p, i.e. ' p
278
 DEDUCTIVE UNIFICATION AND IDEALIZATION
is neutral'. It is clear that classical two-valued logic cannot accom-
modate the third possibility.

In looking for a logical system which will do justice to the possi-

bility that in addition to atomic propositions of form (oeP) or p and
negated atomic propositions (oe'P) or 7p, there are to be considered
also neutral propositions (oe*P) or *p, we must be guided by the
following considerations. Firstly we must observe the principle of
permanence. Since not all propositions are neutral and not all predi-
cates inexact, the rules governing their composition and interrelations
must be so chosen that for propositions that are not neutral and classes
that are exact, they reduce to the rules of classical logic (or, if we prefer,
of intuitionist logic, which I shall not consider here). Secondly, it must
be observed that neutrality is not an independent third truth-value.
It may be turned into either of the other two, i.e. truth or falsehood.
Neutrality is neutrality between truth-values. This rules out the
Eukasiewicz-Tarski type of three-valued logic (and also, as I am inclined
to think, modal logic).
It is fortunate that a logic which is suitable for our purpose is—on
the whole, apart from slight modifications—already available. It has
been constructed by Kleene1 in order to do justice to reasoning in-
volving partial recursive predicates (i.e. predicates which within their
range of definition are algorithmically decidable, but for which so far
no algorithm is available, by which to decide, for any given object,
whedier it does or does not fall within the predicate's range of defini-
tion). Kleene distinguishes the following possibilities : (i) an algor-
ithm is available for deciding whether a given object o falls under a
given predicate or not, i.e. (a) (oeP) (b) {oe'P) and (ii) the possibility that
so far no such algorithm is available, a situation which may be indicated
by (pW). However, tins situation need not be final since an algorithm
might be found which would turn (o?P) into either (oeP) or (oe'P).
Now the important similarity between (o?P) and (oe*P) lies in the fact
that, as Kleene puts it, the third truth-value is in both these cases ' not
on a par with the other two '. Both (o?P) and (oe*P) are provisional,
but whereas in the case of (o?P) we may never find an algorithm which
will turn it into (oeP) or (oe'P), in the case of a neutral candidate o, we
1
Sec especially §64 of his Introduction to Metamathematics, Amsterdam, 1952
279
 S. K6RNER
may always decide to turn (oe*P) into (oeP) or {oe'P) and the corres-
ponding neutral proposition *p into p or 7p. In the first, the Kleene-
case, the provisional evaluation may, but need not, be followed by a
final evaluation. In the other case, the case of neutrality, the provi-
sional can always be followed by a final evaluation, since it is a matter
of decision only.
I shall now, following Kleene, outline the first order predicate logic
prior to thefinalevaluation of neutral propositions. The relevant truth-
tables are :
7P pAq pMq
* •k
P fp t f q t t
1 f f
t f P ' t t t P f t f •k

f t / t * f t t t
*
/
t
f
•k
t •k •
* *
Quantification theory can be developed on the usual lines by
treating universal quantification and existential quantification on the
analogy of conjunction and adjunction. Thus the proposition {x)P(x)
is true if, and only if, all substitution-instances of P are true, false if at
least one is false, and neutral if at least one of them is neutral and the
others neutral or true. Similarly (3x)P(x) is false, if and only if, all the
substitution-instances of P are false, true if at least one of them is true,
and neutral if at least one of them is neutral and the others false. Classes
are also introduced in the usual way by stipulating that two predicates
P(x) and Q(x) determine the same class : (Ax)P(x) = (A^Qfy) if, and
only if, for any object o the substitution-instances P(p) and Q(p) are
either both true, both false or both neutral.
Since neutral propositions can always be turned by our decision into
true or false ones, we can always distinguish between a provisional and
a final stage of evaluation.1 In the provisional stage we ignore all
neutral components together with their connectives, whereas in the
final stage we evaluate them as true or false. Writing square brackets
for the provisional and braces for the final stage we can distinguish
between the following cases :
I-(«)[].{'J>
In the provisional stage the proposition is neither true nor false ; in the
final stage we have truth or falsehood.
{b) [ ] {t}

1
For an earlier formulation of this distinction see ch. VIE of my Philosophy
of Mathematics, London, i960, -where it is applied to the logic of classes.
280
 DEDUCTIVE UNIFICATION AND IDEALISATION
In the provisional stage we have neither truth nor falsehood, but in
the final stage we have either only truth or only falsehood. For
example, if p0 is neutral and q0 is false, then p 0 illustrates (a), (p0 v 7p0)
illustrates (b) and (p0 A 7p0) illustrates (c).
n . (a) [t], {t}
(b) \f\, If}
The value at the provisional stage is always preserved in the final
stage. For example (7 q0 vp 0 ) illustrates (a) and (q0 Ap0) illustrates (b).
IE. (a) [t], {t,J)

The value of the provisional stage is not always preserved but may
change in the final stage. For example, (7 q0 A p0) illustrates (a) and
(q0 v p0) illustrates (b). There is little point in giving special names to
these possibilities, which hold also if the propositions under consider-
ation contain quantifiers.
Since the provisional stage of evaluation is always succeeded by a
final stage in which neutrality has been eliminated, I shall call the
system here developed in outline, the modified two-valued logic. As
regards the meaning of a logical identity in this logic we may choose
between a weaker and a stronger definition. According to the stronger
definition a proposition of the modified two-valued logic is an identity
if, and only if, both its provisional and its final evaluation yield a
classical identity ; according to the weaker definition a proposition of
the modified two-valued logic is an identity if, and only if, the final
evaluation yields a classical identity. If we accept the weaker definition
then the identities of classical logic coincide with the identities of the
modified logic.
It would, however, be a mistake to conclude that there are no im-
portant differences between the two systems. From the point of view
of this talk, the most important difference is that the modified two-
valued logic cannot in general serve as an instrument for the deduction
from neutral statements, and statements with inexact predicates as con-
stituents, of other such statements—in conformity with the usual rules of
inference, especially modus ponens, which though dispensable in predi-
cate logic is not dispensable in an axiomatic theory embedded in it.
Although the form : ((p A (p-*~i))~+ 5) is an identity, we are
allowed to deduce a proposition q0 from the premisses p0 and (p0 -> q0)
only if these premisses are either true or, in case they are neutral, evalu-
ated as true. This means that premisses which in their final evaluation
can be true or false are treated as if they could be evaluated as true only.
u 281
 s. KSRNER
In other words we treat statements characterised by [ ], {t,J) ; or [t]
{t,f} ; or [f], {t,f} as if the second member of each pair were {t} and not
{f, t}. In applying the modus ponens we thus ignore the neutrality of
propositions.
A more radical modification is required if one of our contingent
premisses is a universally quantified proposition involving an inexact
predicate e.g. (x)P0(x), with Po inexact. For it is then not sufficient to
turn a single neutral substitution-instance of P0(x), e.g. P0(x0), where
x0 is a neutral candidate for Po, into a true proposition. To avail one-
sen0 of (x)P0(x) as a premiss, all neutral substitution-instances must be
evaluated as true, whether or not they are specified, unspecified, or
indeed specifiable. We treat P o as if it had no neutral candidates, i.e. as
if it were exact. In applying modus ponens to universally quantified
propositions, we thus not only ignore the neutrality of propositions, but
also the inexactness of predicates. The same applies also for other rules
of inference, e.g. to
C -*• P(x)
— . . „. . (provided x is not free in C).
C -*• {x)P(x) ^ '
To sum up : In so far as we restrict ourselves to logical identities
and logically true propositions, there is hardly any difference between
the classical two-valued logic and the modified two-valued logic,
which unlike the former can accommodate neutral propositions and
inexact predicates, and thus resemblance-predicates and propositions
with such predicates as constituents. As an instrument of deduction of
contingent conclusions from contingent premisses, the classical logic
requires that neutral propositions be replaced by definite propositions
and inexact predicates by exact predicates. This replacement, which
is enforced by a transition from the modified two-valued to the class-
ical two-valued logic, shows that deductive unification of an empirical
field by means of classical logic is not a mere ordering but an idealisation.

Since the basic propositions of a hypothetico-deductive theory func-

tion as premisses from which together with the postulates of the theory
other basic propositions are deducible, the constituent predicates of the
basic propositions are not descriptive, but idealisations, of experience.
A basic statement of a continuous transition involves the replacement
of an empirical notion of continuity, definable in terms of a finite set of
inexact classes, by a mathematical notion of continuity, defined in
282
 DEDUCTIVE UNIFICATION AND IDEALISATION
terms of a non-denumerably infinite set of exact classes. A basic
statement of an equality involves the replacement of an empirical non-
transitive relation by the mathematical relation of equality, which is
transitive. Lastly, the incorporation into the theory of any statement
with an inexact, in particular a resemblance-predicate, among its con-
stituents, involves its replacement by an exact predicate. Of this re-
placement the other two are special cases since empirical continuity and
empirical equality in a certain respect, are resemblance-predicates.
It is thus, strictly speaking, false to assert the identity of the basic
propositions of a theory with the empirical propositions of which they
are idealisations. It is, however, still possible and advisable to distin-
guish between the basic and other propositions of the theory, and to
define as basic those propositions which are identified (though not
identical) with empirical propositions. The identification is per-
missible only for some purposes and within some contexts. Which
these purposes and contexts are, is in any case an empirical question.
If we write ' b <=» e ' for the relation which holds between b and e
when b is an ideahsation of e due to the replacement of empirical by
idealised predicates in the way explained above, then we can replace
the classical schema
(&x and T) ^ fc8
by the more adequate schema
(et «s b,) ; (&! and T) H b2 ; (b2 «a e2)
It indicates and emphasises that whereas b2 is deducible from bx and T,
none of the e-propositions is deducible from any of the t-propositions
or vice versa. Neither is e2 deducible from the conjunction of elt blt T
and b2.
Put into words the schema may be read : If within specified con-
texts and for specified purposes—such as the construction of machines
or the prediction of physical events—an empirical proposition elt
describing a certain type of situation, is (in the manner explained) ideal-
ised into a basic theoretical proposition blt if b2 is deducible from bt
and T, and if b2 is ' de-idealised ' into e2, then if ex is true, e2 is also true.
The following comments on the statement are appropriate. First, it
expresses an empirical proposition. Second, one does not, strictly
speaking, verify, falsify, confirm or disconfirm the theory, but the
whole empirical proposition. In it the theory is only one constituent
—as it were, an empirically incomplete symbol. Third, the scientific
proposition does not differ from an ' ordinary' empirical proposition
283
 S. KORNER
such as ' If e1 is true, e, is also true' by being more or less empirical.
The difference lies in that the scientific empirical proposition is con-
nected through the theory with other empirical propositions, the ideal-
ised counterparts of which are deductively unified by it. Fourth, the
identification of empirical with non-empirical predicates and proposi-
tions allows for great precision and facility of reasoning and calcul-
ation. Last, since the identification is limited to special purposes and
contexts, its extension beyond its original limits, though of heuristic
value, requires fresh evidence.
I conclude with a brief remark on an example of unwarranted
philosophical extrapolation of a scientific theory beyond the Hmits of
legitimate identification, namely the problem of the compatibility
or incompatibility of physics and morality.1 If our bodies are with-
out qualification or limitation subject to the laws of physics, then since
we cannot act without our bodies, those of our actions which seem to
us determined by free decisions are in fact—with certainty or prob-
ability—determined by physical antecedents alone. But since the
applicability of theoretical physics to experience, via the identification
of empirical with idealised propositions, is limited to contexts in which
decisions play no part, it is not theoretical physics, but an unwarranted
philosophical extrapolation of it, which is incompatible with the
assumption that decisions, for which we are responsible, at least some-
times determine our actions. The compatibility between scientific
theories (embedded in classical logic) and moral responsibility does not
imply that moral responsibility is, or is not, an empty notion. But it
leaves room for the belief that it is not empty—a belief or faith, which
is deeply rooted in Western tradition.
1
For a fuller statement see ' Science and Moral Responsibility' to be published
in Mind., April 1964, 73
Dept of Philosophy,
The University,
Bristol, 8

284