You are on page 1of 11

Determinism, the Remote past, and the Causal or Determinational Structure of the Universe

Author(s): David Sapire


Source: Philosophy of Science, Vol. 56, No. 3 (Sep., 1989), pp. 474-483
Published by: The University of Chicago Press on behalf of the Philosophy of Science Association
Stable URL: http://www.jstor.org/stable/187996 .
Accessed: 11/04/2013 08:25
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press and Philosophy of Science Association are collaborating with JSTOR to
digitize, preserve and extend access to Philosophy of Science.

http://www.jstor.org

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINISM, THE REMOTE PAST, AND THE CAUSAL OR


DETERMINATIONAL STRUCTURE OF THE UNIVERSE*
DAVID SAPIREt
Department of Philosophy
University of the Witwatersrand, Johannesburg

Lukasiewicz and, more recently, other philosophers have cast doubts on arguments from one version of determinism to another: roughly, from the view
that every event (condition, state) has a cause or is determined, to the view that
the remotest possible past determines the present and future. This paper defends
a special class of such arguments. It identifies constraints on the relation of
determination under which the arguments concerned are valid. And, by reference
to the overall causal or determinational structure of the universe, it argues that
the constraints themselves are highly plausible.

1. Introduction. It may seem to be undeniable that if every event (condition, state) is caused or determined by something earlier than it, then
the present and future must be determined by the remotest possible past.
But such "remote past arguments" (RPAs), as I shall call them, which
argue from one version of determinism to another, are invalid. As
Lukasiewicz (1967, pp. 30-32) and others have noted, even an infinite
sequence of causes reaching back in time from an event might, like an
infinite sequence of real numbers, have a lower limit or be bounded below, and so not reach back into the remote past at all. And there are other
grounds on which RPAs may fail. Furthermore, some philosophers have
claimed that the extra premises to be added to RPAs in order to make
them valid are themselves doubtful or questionable (van Inwagen 1983;
Sobel 1975).
Are we therefore to move to the opposite extreme, and conclude that
RPAs in general, despite their superficial cogency, are actually quite unreliable? The points I have mentioned might well suggest such a move;
it would nonetheless be a mistake to make it. There is another side to the
RPA story, and I want to present it here. Some RPAs, which I call "good
RPAs" (section 2), are very good arguments; the extra premises required
to render them valid (section 3), are highly plausible (section 4). Indeed,
*Received February 1987; revised May 1987.
tI am grateful to Michael Pendlebury, Flint Schier, and an anonymous referee for Philosophy of Science for helpful comments on earlier versions of this paper. I also acknowledge financial support by the Human Sciences Research Council for research on which I
have drawn here.
Philosophy of Science, 56 (1989) pp. 474-483.
Copyright ?) 1989 by the Philosophy of Science Association.

474

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINATIONAL STRUCTURE OF THE UNIVERSE

475

it is only plausible that these premises are false if it is plausible that the
causal or determ-iinational
structureof the universe as a whole is not merely
peculiarly complicated, but, perhaps, utterly bizarre (section 4).
2. The Form of Good RPAs. The premises and conclusions of RPAs
vary appreciably, though within certain limits. The premises I take to be
variants of the following broad claim:
Every event (condition, state) has a cause or is determined.

(P)

And the conclusions of the specific class of RPAs to be considered


here-there are others-I take to be variants of another broad claim:
Every (overall, instantaneous) state of the universe
determines every later state of the universe.

(C)

Different interpretations of most of the terms in (P) and (C) are possible. Thus there are many different RPAs, and some of them are better
arguments than others. Because a main interest here is in logical relationships, it is best to begin by considering the form of RPAs. A few
closely related forms might be suggested for (P), but the following is one
of the more general and so more appropriateof them at this point:
(u) (3 v)[D'(v,

u)]1.

(PI)

D'(v, u), which I shall call a "determinationpredicate", occurs in (P')


because (P) claims that every event (condition, state) is related by what
may be called a "determinationrelation" to some event (condition, state).
As long as the concern remains with form, let us leave aside the content
of determination predicates. But it remains convenient to read D' (v, u)
as "v determines u" or "u is determined by v".
As for the conclusions of RPAs, (C) leaves open the nature of the
states, as well as that of the determination relation involved. (C) varies
as these vary. One clear indication of the form of (C) is given by:
(x)(y)[T(x)

*T(y) - x < y

D(s(x),s(y))].

(C')

(C') could be viewed purely formally. And D(s(x,s(y))) should be so


viewed for awhile, although like D'(v, u) it is a determination predicate
that may be read for convenience as "s(x) determines s(y)". But reference
to time, as well as to overall instantaneous states of the universe, is distinctive of the RPAs whose plausibility I want to defend.
Let us call such RPAs "good RPAs". (This use of the word "good"
does not prejudice anything. In later sections I clarify the sense in which,
and defend the view that, good RPAs are indeed good arguments. In this
section the task is to identify the class of "good" RPAs.) Now if (C') is

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DAVID SAPIRE

476

to capture the import of good RPAs, then T(x) must be an abbreviation


of the predicate "x is a moment of time". Also, s(x) must be a functional
term referring to the overall and instantaneous state of the universe at the
time x. And x < y must be a representation of the predicate "x is earlier
than y".
These specifications in turn require that the premises of good RPAs
take a certain form; versions of (P) that imply (C') must involve the same
terms as (C'). Otherwise, for example, an RPA might fail because its
premise concerns events and its conclusion, say, states. Or it might fail
because of differences in the determination predicates in its premise and
conclusion. For instance, suppose that D'(u,v) is the following kind of
non-necessitating causal predicate: (roughly) u causally produces but does
not necessitate v. And suppose that D(u,v) is a predicate expressing what
might be termed "law-based determination":the occurrence of u together
with the laws of nature ensures that v cannot but occur. Then, arguably,
D'(u,v) could be true, but D(u,v) false. Incidentally, van Inwagen, though
he mentions the point made by Lukasiewicz, bases his doubts about RPAs
mainly on just such an argument (1983, p. 5).
Failures like these are inessential to good RPAs as such; they can occur
in other arguments, and can be precluded from good RPAs. In order to
preclude them, I shall restrict the class of good RPAs to RPAs with a
main premise of the following form:
(x)[T(x) -- (3y)(T(y)

*y < x D(s(y), s(x)))].

(P")

(Every state of the universe is caused or determinedby a previous state.)


The restrictions so far placed on good RPAs still leave too much open.
Most importantly, they leave us largely in the dark about the nature of
instantaneous states, and about the properties of time as it features in (P")
and (C'). The nature of the states involved can be left open. By contrast,
good RPAs make some quite specific assumptions about time and about
the temporal ordering of states. They involve a classical notion of absolute time-as a single or universal and totally ordered set of instants
or moments of time. Correspondingly, the instantaneous states of the universe as a whole which they concern involve a notion of absolute simultaneity. Good RPAs, moreover, assume that these states inherit the
total temporal ordering.
More precisely, and more generally, good RPAs conform with the following assumptions:
(A) First, the relation < totally orders the set of instants of time, T.
That is, < is irreflexive, transitive, and connected on T. Second,
the notion of an overall and instantaneous state of the universe
can be defined in such a way that it makes sense to say that s(x)

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINATIONAL STRUCTURE OF THE UNIVERSE

477

maps elements of T to states; s(x) associates the time x with the


state of the universe at that time. I shall call the sequence H =
{s,}, with s, = s(x) for all x in T, "the history of the universe".
H inherits a total ordering from T: a state s(x) either precedes,
or occurs at the same time as, or succeeds, a state s(y) iff x <
y, or x = y, or y > x respectively.
Though assumptions like those in (A) are usually made when RPAs
are put forward, they conflict with relativity theory. But I shall ignore
such conflicts here, and confine myself to an investigation of RPAs under
a classical understanding of time.
So far, then, good RPAs have premises of the form (P"), the further
premises (A), and conclusions of the form (C').
3. The Extra Premises in Valid RPAs. As yet there is no reason to
think that good RPAs are really good arguments. Indeed, they are invalid.
They can fail in three main ways, all of them concerned with the determination predicate. It is worth considering these failures because they
provide the best guide to the kind of extra premises that must be added
to good RPAs so that they become valid arguments. In each case (P")
would be true but (C') false:
(a) It might be that some state s(x) is determined by some earlier
state s(y), that is, D(s(y),s(x)), that s(y) in turn is determined by
an earlier state s(z), but that s(z) does not determine s(x).
(b) It might be that some state s(x) determines some later state s(y),
that s(z) is between s(x) and s(y), but that s(x) does not determine
s(z) nor s(z) determine s(y).

(c) If time is continuous, then it might be that s(x2) determines s(x1),


that S(X3) in turn determines s(x2), and so on, as required by (P"),
but that no term of the infinite sequence of determiners . . . s(xn),
. . S(X2), s(x1) stretching back in time from x1 occurs
earlier than or as early as some time t. The sequence of determiners would be bounded below by the state s(t). This case,
stressed by Lukasiewicz and Sobel, reflects the fact that any point
t on a continuous line can be approached as closely as one likes
from points "above" or "below" t without t ever being reached.
(It may be noted that if time is assumed to be discrete ratherthan
continuous, and something like (P") is true, then every sequence
of determiners stretches back into the remotest possible past or
lacks a lower bound.)

S(X,,I)g.

I now state three conditions on the determination relation D(s(x),s(y))


which ensure that (P") and assumptions (A) imply (C'):

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DAVID SAPIRE

478

(I) The relation is transitive: If D(s(x),s(y)) and D(s(y),s(z)), then


D(s(x),s(z)) for all times x, y, and z.
(II) The relation is contiguous: If D(s(x),s(y)), then for all times z
such that x < z < y, D(s(x),s(z))

and D(s(z),s(y)).

(III) The relation is unbounded below: If xo is any moment of time


and x1 any moment of time such that x1 < xo and D(s(x1),s(xo)),
then for all times t < x1 there are states s(xn), s(xn_-), . . ., S(Xl),
s(xo), with xi < xi-1 (for i = 1,2, . . ., n), such that xn < t and
D(s(x,)

vs(x,-

1)) v . . . .

D(s(xl),s(xO)).

It is easy to show that (P") and (A), together with (I), (II), and (III),
imply (C').
Proof outline. Let s(x) and s(y) be any two states, with x < y. We
want to show that D(s(x),s(y)). By (P"), there is a time z such that
z < y and D(s(z),s(y)).

By (A), either z = x, or z < x, or x < z.

In the first case the result follows immediately. If z < x, then by


(II), D(s(x),s(y)). If x < z, then by (III) there is a sequence
D(s(z) s(z- 1)),... , D(s(z1),s(z)) with zn < x. By (I), D(s(z,),s(y)).
So by (II), D(s(x),s(y)).
4. The Plausibility of the Extra Premises. The argument form having
(A), (I), (II), (III) and (P") as its premises, and (C') as its conclusion,
only becomes an actual RPA, rather than an RPA form or schema, when
its relational predicate letter is replaced by what I have called a "determination predicate". And then the extra premises (I), (II), and (III) make
substantive claims about the determination relation concerned-in particular, about the structuralproperties of that relation. So given a particular determination relation, such claims can be assessed as true or false,
plausible or implausible, and so on. In addition, (I), (II), and (III) induce
structure on the sequence of states H-structure that I shall call "determinational structure". H, it may be recalled, does not lack structure entirely; it inherits the assumed temporal order on the moments of time
themselves. But the furtherstructure, the determinational structure, which
H acquires because of the constraints that (I), (II), and (III) place on
determination relations, is crucial here; it provides us with an invaluable
way of assessing the plausibility of (I), (II), and (III).
In the rest of this section I argue that if the determination predicate is,
roughly speaking, a causal predicate, of any one of a number of varieties,
then (I), (II), and (III) are extremely plausible.
Causation being a notoriously slippery and contentious topic, however,
it might be thought that there are insuperable problems at this point. But
without prejudicing the argument it is possible to avoid such problems
by adopting a broad view of causal or determinational predicates. In par-

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINATIONAL STRUCTURE OF THE UNIVERSE

479

ticular, I allow that (D(u, v) can be defined along at least the following
two broadly indicated lines:
(i) D(u,v) iff the occurrence of u, by way of some inherent connection with v, necessitates the occurrence of v;
(ii) D(u,v) iff a proposition to the effect that u occurs or obtains,
together with the laws of nature, implies a proposition to the
effect that v occurs or obtains.
(i) may be thought of as classical, (ii) as law-based causation or determination. Both are conceptions of sufficient causation, which features
both in common and in good RPAs.
The question of the plausibility of the extra premises now becomes the
question whether it is plausible that classical or law-based determination
relations on instantaneous states of the universe as a whole are transitive,
contiguous, and unbounded below. In answering it, I assume that the
version of determinism captured in (P"), interpreted in the light of (A),
and with the determinationpredicate taken as either classical or law-based
causation of determination, is true. So interpreted, (P") might be false;
quantum theory suggests that it is in fact false. But the question here is
what is in addition plausibly the case if it is true.
One of the least contentious claims about relations of sufficient causation in general is that they are transitive. There are conceptions of causal
relations that, like Lewis's relation of causal dependence (Lewis 1973),
are not transitive, or that, like Suppes' notion of probabilistic causality
(Suppes 1980), are only transitive under special conditions (Eells and
Sober 1983). However, it is not notions like these, but ones of sufficient
causation that feature in good RPAs. And sufficient causation, whether
of a classical or law-based type, is clearly transitive. I shall therefore
assume that (I) is not merely plausible, but true.
Now consider (II). The contiguity expressed in (II) excludes causal
action or determination at a distance in time. It allows one state s(x) to
cause or determine another state s(y), with x # y, only by the "action"
of the first being transmitted to the second through the medium of the
intervening states and times, or by the determinational connection between the two states passing through the intervening states. (II) is compatible with time being discrete ratherthan continuous. It requires in general not the continuity but the contiguity or adjacency of causal action or
of law-based determination.
That causes and effects are spatially contiguous is regarded widely as
plausible. When the causes and effects are overall states of the universe,
and the contiguity temporal, as in the present case, the plausibility increases enormously-especially when, as will be done here, the truth of
(I) is assumed for the purpose of assessing the plausibility of (II). It may

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

480

DAVID SAPIRE

seem that (II) is more plausible for classical than for law-based causation
or determination. But that is not so, as we shall see. The basic problem
is that if (II) is false then a variety of bizarre possibilities for the causal
or determinational structureof the universe, classical or law-based, arise.
None of the possibilities is appreciably more implausible than the others.
The best way to exclude any of them is to exclude all of them: by regarding (II) as more likely to be true than its negation.
If (II) is false, one state s(x) might cause or determine a later state s(y)
entirely independently of any of the states between s(x) and x(y) in the
history of the universe. The history of the universe might have had an
entirely different "filling" between s(x) and s(y). Even more problematically, it could now happen that all the states in some initial segment of
the history of the universe I
[s(x) Ix < y], or in some final segment F
-[s(x)lz < x], with y < z, are causally or determinationally related to
each other, but no state in I or F is so related to any state not in I or F.
From the point of view of its causal or determinational structure, such a
history has two quite separate parts.
Time must be continuous for this to occur, for if it is discrete then (P")
and (I) will ensure that any state in F is causally or determinationally
related to some state in I. But similar possibilities nonetheless arise if
time is discrete. The main problem is that the denial of (II) allows the
possibility of the fragmentation of the causal or determinational structure
of the universe. Consider another example that works if time is continuous or discrete. Suppose each state is determined by the state occurring,
say, five minutes before it. (P") would be true, but (II) would be false.
Then, unless other assumptions are made, the universe would break up
into infinitely many distinct causal or determinational chains-indeed,
into a nondenumerable infinity of them if time is continuous. The states
that are the components of each chain would be causally or determinationally related to each other, but no state in any chain would be so related
to any state in any other chain. Or there could be just one entirely separable chain, with all the states not in that chain related to each other.
Or there could be two separable chains, or three, or any number of them.
To be sure, other assumptions, ones compatible with the truth of (P")
and falsity of (II), could be introduced to exclude such possibilities. For
instance, one could simply claim that there can be no causally or determinationally separable chains; some state in any chain must be causally
or determinationallyrelated to some state in any other chain. But I suspect
that all such assumptions would be more ad hoc, less well motivated, and
more complicated than (II). It should be noted in this connection that the
temporal gaps between consecutive states in chains need not be five-minute ones. They can last an hour, or a millennium, or any other period.
And they need not even be regular gaps. Not only could they be irregular,

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINATIONAL STRUCTURE OF THE UNIVERSE

481

they could be entirely without rhyme or rhythm.


What is wrong with all these possibilities? Why can't the universe consist of two, or three, or any number of causally or determinationally distinct parts, unrelated to each other either by inherent causal connections
or by laws? Similarly, why can't the time gaps between consecutive causally
or determinationally related states in a causal chain be as irregular as one
wants? Let us take it that these are indeed possibilities. To adapt a wellknown remark, let us allow that the universe is quite possibly not merely
more complicated than we do suppose, but more complicated than we
can suppose. What is possible, however, is one thing; what is plausible,
quite another. The main problem with the possibilities I have mentioned,
and with the many others like them, is that they are all exceedingly implausible. And they are implausible largely because they conflict radically
with that elusive yet apparently widely shared sense of simplicity that
guides the development and choice of scientific theories. For instance,
the simplest, the most harmonious, time gaps between consecutive causally
or determinationally related states in a causal chain are the regular onesno matter if the states are related by real connections or by laws. Similarly, the simplest of the infinitely many possible different regular gaps
are those required by (II): the shortest possible ones. Whether time is
discrete or continuous, simplicity suggests that states only determine distant states by determining nearer ones. Similarly too, the simplest number
of causally or determinationally distinct parts that the universe can have
is just one.
There is another way of supporting (II), but one that I shall only touch
on. As noted, if (II) is false, the history of the universe might lack any
overall causal or determinational unity. Consequently, its linear, regular,
and very orderly temporal structure, which I have assumed, and its causal
or determinational structure, which is induced by causal or determinational relations among states, can turn out to be radically different. But
whatever one's precise view about the difficult and disputed relationship
between time and causality, it is to be expected that the overall temporal
structure of the universe, and its overall causal structure, mirror each
other closely. Hence, (II) rather than its negation is likely to be true. For
it is the negation of (II) that allows temporal and causal structureto come
apart so radically.
(III), unboundedness below, is of a type with (II). Some plausibility
may attach to the idea that causal or determinational chains are bounded
below when the relata entering into them are localized facts or events.
That is the kind of case that Lukasiewicz and Sobel concentrate on. But
all such plausibility vanishes when the relata are overall states of the
universe, as in the present case. Suppose that the determination relation
is bounded below. And suppose that s(t) is the lower limit of causal or

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

482

DAVID SAPIRE

determinational sequences reaching back in time from some state s(x).


Then s(x) has no causal or determinationalconnections with any state s(z),
where z is earlier than or the same time as t. The result is a singularity
in the causal or determinational structure of the universe at t. Indeed, if
the determination relation is not only transitive but also contiguous, then
the history of the universe splits into two causally or determinationally
quite distinct parts at t. For s(t) has to be a lower bound on the causal
sequences reaching back in time from any state s(y), where t < y. Moreover, under the same assumptions of the transitivity and contiguity, it is
quite possible that the universe breaks up into three, or four, or into any
number of distinct parts, each one a causally or determinationally united
whole, but none of them united either by real causal connections or by
laws to any other part.
Dropping the assumption that the relation is contiguous does not help
much. A species of causal or determinational unity can be restored in this
way, but only at the expense of a convoluted pattern of causal or determinational relations. And if the contiguity assumption is retained, then
once again temporal and causal structure can differ radically. How can
the parts, however many of them there may be, be in a single time series,
or even in related ones? And why, say, two parts rather than, say, seven,
or thirteen, or forty, or some other magical number?
None of these possibilities-and there are in fact infinitely many of
them-is remotely plausible. None of them is remotely more plausible
than any of the others. If there is a universe, so that we do not have no
parts at all, and if every state of the universe is caused or determined by
somneearlier state, then there is only one plausible possibility for the total
number of causally or determinationally separate parts of the universe.
Similarly, there is only one plausible possibility for the overall causal or
determinational structure of the universe. The universe is one, and it is
very simply structured.
If simplicity considerations are anything to go by, and if each state of
the universe is caused or determined by some earlier state, then, in all
likelihood, the entire history of the universe is caused or determined by
states in its remotest past. In other words: RPAs with a main premise of
the form (P") interpreted in the light of (A), the extra premises (I), (II),
and (III), the conclusion (C'), and the determination predicate interpreted
appropriately, are good arguments; they are valid and their additional
premises (I), (II), and (III) are highly plausible.

REFERENCES

Eells, E., and Sober, E. (1983), "Probabilistic Causality and the Question of Transitivity",
Philosophy of Science 50: 35-75.

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

DETERMINATIONAL STRUCTURE OF THE UNIVERSE

483

Lewis, D. (1973), "Causation", The Journal of Philosophy 70: 556-567.


Lukasiewicz, J. [1961] (1967) "On Determinism", in S. McCall (ed.), Polish Logic 19201939. Oxford: Clarendon Press, pp. 19-39.
Sobel, J. H. (1975), "Determinism: A Small Point", Dialogue 14: 617-621.
Suppes, P. (1970), A Probabilistic Theory of Causality. Amsterdam: North-Holland.
van Inwagen, P. (1983), An Essay on Free Will. Oxford: Clarendon Press.

This content downloaded from 111.68.96.57 on Thu, 11 Apr 2013 08:25:25 AM


All use subject to JSTOR Terms and Conditions

You might also like