=Paper=
{{Paper
|id=None
|storemode=property
|title=The Shape of Absolute Coincidences. Salmon's Interactive Fork Model as Shape of Coincidental Processes
|pdfUrl=https://ceur-ws.org/Vol-1007/paper8.pdf
|volume=Vol-1007
|dblpUrl=https://dblp.org/rec/conf/shapes/Melas13
}}
==The Shape of Absolute Coincidences. Salmon's Interactive Fork Model as Shape of Coincidental Processes==
https://ceur-ws.org/Vol-1007/paper8.pdf
The Shape of Absolute Coincidences.
Salmon's Interactive Fork Model as Shape
of Coincidental Processes.
Alessandra MELAS
University of Sassari, Italy
Abstract. According to a particular view, chance events are not uncaused but they
are simply the result of intersecting causal lines. More precisely, the intersections
between different processes that belong to independent causal chains are the
origin of accidental events, called absolute coincidences.
This paper provides a new account devoted to showing the strong relation
between absolute coincidences and Salmon's interactive fork criterion, in an
attempt to endorse the idea that coincidences can be shaped in terms of a causal
model.
Keywords. Absolute coincidences, causal, interactive forks, shape
Introduction
As for the word “chance”, the word “coincidence” is used to indicate many different
things. The present study, however, considers only a particular type of coincidences,
namely what is known as absolute coincidences.
According to Jacques Monod, absolute coincidences are the result of intersections
between different processes that belong to totally independent causal chains1:
Mais dans d’autres situations, la notion de hasard prend une signification essentielle et non plus
simplement opérationnelle. C’est le cas, par exemple, de ce que l’on peut appeler les “coïncidences
absolues”, c’est-à-dire celles qui résultent de l’intersection de deux chaînes causales totalement
indépendantes l’une de l’autre.2
I will call that type of coincidences “causal absolute coincidences”.
The first part of the present work provides a precise definition of causal absolute
coincidences. The second one presents the strong relation between causal absolute
coincidences and Salmon's common cause model. As we will see, causal absolute
coincidences are events that can be divided into intersecting causal components, and
1
In Metaphysics, Aristotle already maintains the fact the existence of per accidence causes is a sign of the
existence of per se causes. In commenting on Aristotle's Metaphysics, Saint Thomas also says that if we treat
accidental beings as things produced by per se causes, many things may be by accident, such us the meeting
of independent causal lines. Although very important for the philosophical historiography, I leave out from
this discussion the Aristotelian-Thomistic notion of chance. For an extended enquiry see [1].
2
[2], p. 128.
107
that intersection gives origin to consequences. Equally, Salmon's interactive forks are
characterized by two, or more, intersecting processes and two, or more, ensuing
processes. It seems that causal absolute coincidences can be represented by an x-shape,
as with Salmon's interactive forks.
The core idea of this paper is that representing and reasoning with the shape of
coincidental phenomena is essential to understand those phenomena.
The fact that the DNA molecule has the shape of a double helix is crucial to
understand how it functions. The same is true of coincidences: the fact that
coincidences have a particular shape is crucial to understand how they work.
1. Causal Absolute Coincidences: a Definition
As we have already seen, causal absolute coincidences are not uncaused, but they are
simply the effect of the intersection of independent causal processes.
We can explain the independence between different processes, A and B, that belong
to independent causal chains in the following terms:
1) A and B are independent if they are statistically independent, so that:
P(A/¬B) = P(A/B) = P(A) (1)
and
P(B/¬A) = P(B/A) = P(B) (2)
2) The statistical independence between A and B is not due to a common cause in
their past3.
To clarify this point let us consider Monod's example as it is represented in Figure
1. Doctor Dupont is going to visit a patient for the first time. In the meanwhile, Mr
Dubois is fixing a roof in the same area. When doctor Dupont comes across Dubois'
work site, Dubois' hammer falls inadvertently down and the trajectory of the hammer
intersects the trajectory of doctor Dupont, who dies4.
In Figure 1:
P(A/¬B) = P(A/B) = P(A) (1)
and
P(B/¬A) = P(B/A) = P(B) (2)
3
I want to avoid the case in which an event X is a screening-off common cause of the two processes A and B.
In that case, according to Reichenbach's screening-off condition, given X, A and B are independent of each
other. For a more extended discussion on Reichenbach's screening-off condition see [3].
4
[2], p. 128.
108
Figure 1. Monod’s example of absolute coincidences
That is, the fact that doctor Dupont goes to visit his patient is statistically
independent of the fact that the hammer falls down, and the fact that the hammer falls
down is statistically independent of the fact that doctor Dupont goes to visit his patient.
The two dotted lines in Figure 1 represent the two independent causal histories of A
and B.
To sum up, causal absolute coincidences are events that can be divided into
components independently produced by some causal factor and those components join
together.
Let us consider another example. Suppose I am watching a TV programme on
Boris Pasternak. In the meanwhile my best friend, without knowing what I am doing
and without knowing anything about that TV programme, is reading doctor Zhivago.
We would say that it is a coincidence that at the same time (but in different places)
my friend and I are doing something that concerns Boris Pasternak.
However, since there is no physical direct interaction between the two coincidental
processes, someone may conclude from this example that something is a coincidence
only in the eye of the beholder.
Hence, we need a more precise definition of what a causal absolute coincidence is:
109
We speak of a causal absolute coincidence whenever there is an intersection, which
is also a physical interaction, of two or more statistically independent causal processes
in exactly the same space and at exactly the same time.
A good example could be Monod’s one: in that case the two independent causal
processes intersect, and physically interact, in exactly the same space and at exactly the
same time. No one may conclude that the intersection is a coincidence only in the eye
of the beholder5.
Moreover, in such cases the intersection gives origin to some consequence, as it is
illustrated in Figure 2.
2. The Connection Between Salmon's Interactive Fork Criterion and Causal
Absolute Coincidences
Causal absolute coincidences are events that can be divided into components which
intersect (and also interact) in certain spaces and at certain times and, as we have
already seen, that intersection gives origin to consequences.
This section is devoted to showing the strong relation between causal absolute
coincidences and Salmon's common cause model. More precisely, after an overview of
Salmon's model, I will show that coincidences can be entirely described in terms of
interactive forks6.
2.1 Salmon's Interactive Fork Model
As Salmon says:
[...] Consider a simple example. Two pools balls, the cue ball and the 8-ball, lie upon a pool
table. A relative novice attempts a shot that is intended to put the 8-ball into one of the far corner
pockets, but given the positions of the balls, if the 8-ball falls into one corner pocket, the cue ball is
almost certain to go into the other far corner pocket, resulting in a ``scratch''. Let A stand for the 8-
ball dropping into the one corner pocket, let B stand for the cue ball dropping into the other corner
pocket, and let C stand for the collision between the cue ball and the 8-ball that occurs when the
player executes the shot. We may reasonably assume that the probability of the 8-ball going into the
pocket is also about 1/2 if the player tries the shot, and the probability of the cue ball going into the
pocket is also about 1/2. It is immediately evident that A, B, and C do not constitute a conjunctive
fork, for C does not screen off A and B from one another. Given that the shot is attempted, the
probability that the cue ball will fall into the pocket (approximately 1/2) is not equal to the
probability that the cue ball will go into the pocket, given that the shot has been attempted and that
the 8-ball has dropped into the other far corner pocket (approximately 1).7
5
One may say that coincidences are also unexpected events, so that one could ask whether a non-unexpected
intersection, between processes that belong to independent causal chains, is still a coincidence. I find this
point very interesting, but I will leave it out of this discussion.
6
Although a very similar work can be done using Bayesian networks, I will leave that discussion for another
paper.
7
[4], pp. 168-169.
110
Figure 2. Consequences of coincidences
Salmon's interactive forks are considered as spatio-temporal intersections between
two processes. The space-time diagram of the interactive forks has the shape of an x, as
illustrated in Figure 3.
In Figure 3, P1 and P2 are two processes which intersect in C. E1 and E2 are the two
emerging processes from the intersection.
The intersection produces an interaction and the interaction, according to
Reichenbach's mark criterion8, has the capacity to produce changes in the properties of
the two separate processes that come from the intersection.
In order to characterize Salmon's forks we need the following condition:
P(E1 ∧ E2/C) > P(E1/C) × P(E2/C) (3)
The intersection in C makes the two effects statistically dependent on each other.
The interactive forks are considered as spatio-temporal intersections that in general
violate Reichenbach's screening-off condition9.
8
For a more extended discussion on Reichenbach's mark criterion see [3].
9
For a more extended discussion on Reichenbach's screening-off condition see [3].
111
Figure 3. Salmon’s interactive fork
However, the screening-off condition may represent a limit case for interactive
forks. There seems to be a kind of screening-off that is valid for some macroscopic
interactive forks10 and it can be represented by the following condition:
P(E1 ∧ E2/C) = P(E1/C) × P(E2/C) = 1 (4)
which violates the relation (3). Salmon calls this sort of forks “perfect forks”.
According to Salmon, we can say:
The main point to be made concerning perfect forks is that when the probabilities take on the
limiting values, it is impossible to tell from the statistical relationships alone whether the fork
should be considered interactive or conjunctive. 11
Which is, then, the most important difference between conjunctive12 and interactive
forks?
The answer is: the shape of the fork, as we can read from the following quotation:
10
[4], pp. 177-178.
11
[4], p. 178.
12
For a detailed discussion on conjunctive forks see [3].
112
[…] For it seems essential to have two processes going in and two processes coming out in
order to explain the idea of mutual modification. 13
In recognizing interactive forks, the shape of the phenomena is more relevant then
statistical relationships.
Moreover, according to the conjunctive fork model, the common cause can produce
one of its effects without producing the other one, and vice versa. According to the
interactive fork model, the common cause (the intersection) cannot produce one of its
effects without producing the other one, and vice versa.
Furthermore, as a result of the physical interaction, in the case of interactive forks,
the two emerging processes have something (a mark) of each other and this is not true
for the case of conjunctive forks.
2.2 The Coincidences x-Shape and Interactive Forks
Interactive forks are characterized by two, or more, intersecting (and interacting)
processes and two, or more, ensuing processes. Equally, absolute coincidences are
events that can be divided into components which intersect (and interact) in certain
spaces and at certain times and that intersection gives origin to consequences. It seems
that causal absolute coincidences can be represented by an x-shape, like interactive
forks. I will call this x-shape coincidences x-shape, and I will try to investigate its
relation with x-shapes which characterize interactive forks.
2.2.1 The First Part of the Coincidences x-Shape and Interactive Forks
In Salmon we find some instances in which the two intersecting processes, P1 and P2,
are statistically independent of each other, as in our definition of causal absolute
coincidences. We can quote the following example:
In every day life, when we talk about cause-effect relations, we think typically (though not
necessarily invariably) of situations in which one event (which we call the cause) is linked to
another event (which we call the effect) by means of a causal process. Each of the two events in this
relation is an interaction between two (or more) intersecting processes. We say for example, that the
window was broken by boys playing baseball. In this situation there is a collision of a bat with the
ball (an interactive fork), the motion of the ball through space (a causal process), and a collision of
the ball with the window (an interactive fork).14
According to this example, the ball travelling towards the window and the window
are two intersecting processes; the collision between them represents the intersection.
In this case, the two intersecting processes belong to independent causal chains, since
the fact that the window is there is statistically independent of the fact that the ball
travels towards the window, and vice versa.
Salmon does not care about the distinction between intersecting processes that
belong to independent causal chains and intersecting processes that belong to non-
independent causal chains. This is not a relevant point for a good definition of
interactive forks.
Interactive forks are not a sufficient requirement to say that an interaction happens
by coincidence, since a fork can always be found in which the interacting processes do
13
[4], p. 182.
14
[4], p. 178.
113
not belong to independent causal chains15. However, cases of intersecting processes
that belong to independent causal chains are not ruled out by Salmon's account, and
Salmon's model can be used to describe such cases.
The first part of the coincidences x-shape can be described with Salmon's model.
2.2.2 The Second Part of the Coincidences x-Shape and Interactive Forks
Let us consider again Monod's example, as it is represented in Figure 2. In that case we
may reasonably assume that the probability of doctor Dupont dying D is of 1/2 if there
is a collision between the hammer and doctor Dupont's head C, and that the probability
of the hammer having some bits of brain E is also of about 1/2. In this case, it is
evident that D and E, given C, are statistically dependent on each other:
P(D ∧ E/C) > P(D/C) × P(E/C) (5)
Given that the hammer collides with doctor Dupont's head, the probability that the
hammer will have some bits of brain (approximately 1/2) is not equal to the probability
that the hammer will have bits of brain, given that the hammer has collided with doctor
Dupont's head and doctor Dupont has died [P(E/C ∧ D) ≈ 1]. That is:
P(E/C ∧ D) > P(E/C) (6)
Moreover, given the collision between the hammer and doctor Dupont's head, the
probability that doctor Dupont will die (approximately 1/2) is not equal to the
probability that the doctor will die, given that the hammer has collided with doctor
Dupont's head and the hammer has some bits of brain [P(D/C ∧ E) ≈ 1]. That is:
P(D/C ∧ E) > P(D/C) (7)
The intersection in C makes the two effects, D and E, statistically dependent, like in
Salmon's example of the two billiard balls.
However, as we have already seen, the condition of screening-off represents a limit
case for interactive forks. There seems to be a kind of screening-off which is valid for
some macroscopic interactive forks. That limit case is represented by the following
condition:
P(E1 ∧ E2/C) = P(E1/C) × P(E2/C) = 1 (4)
Consider the two billiard balls example once more. Suppose that our novice returns
to attempt another shot from time to time. Since practice helps improve one’s skills to
perfection, the novice becomes so good that he can invariably make the cue ball and
the 8-ball collide (C) in the manner that the 8-ball drops into one of the far corner
pockets (E1) and the cue ball goes into the other far corner pocket (E2). We may
reasonably assume that the probability of the 8-ball going into the pocket is about 1 if
the player tries the shot, and the probability of the cue ball going into the pocket is also
15
The aim of this paper is to show that interactive forks are a necessary condition for coincidental
phenomena.
114
about 1. It is immediately evident that C does screen E1 and E2 off from one another.
Up until the moment when our player has perfected his technique, the results of his
shots exemplified interactive forks, and it would be absurd to claim that when he
achieves perfection, the collision of the two balls no longer constitutes a causal
interaction, but it must now be considered as a conjunctive fork. It is an arithmetical
accident that, when perfection occurs, the equation (4) is fulfilled while the inequality
(3) must be violated.
Let us go back to Monod's example once again. We may reasonably assume that
the probability of doctor Dupont dying (D) is of about 1, if there is a collision between
the hammer and doctor Dupont's head (C), and that the probability of the hammer
having some bits of brain (E) is also of about 1. In this case, it is evident that C screens
D and E off from one another:
P(D ∧ E/C) = P(D/C) × P(E/C) = 1 (8)
However, it would be absurd to claim that the collision between the hammer and
doctor Dupont's head no longer constitutes a causal interaction, but must now be
considered as a conjunctive fork. According to this last example, the common cause
(C) cannot produce one of its effect without producing also the other one, and vice
versa. Moreover, the two emerging processes (D and E) have something of each other.
It is an arithmetical accident that, when perfection occurs, the equation (4) is fulfilled
while the inequality (3) is violated.
According to what I said in this section, even the last part of the coincidences x-
shape can be described using Salmon's interactive fork model.
3. Conclusion
To conclude, the primary reasons for saying that interactive forks can describe causal
absolute coincidences are:
1. Interactive forks are characterized by two, or more, intersecting (and interacting)
processes and two, or more, ensuing processes. Equally, absolute coincidences are
events that can be divided into components which intersect (and interact) in a given
space and at a given time, and that intersection gives origin to consequences. Causal
absolute coincidences can be represented by an x-shape, as with interactive forks.
2. Cases of intersecting processes that belong to independent causal chains are not
ruled out by Salmon's account and Salmon's model can be used to describe such cases.
The first part of the coincidences x-shape can be described by Salmon's criterion.
3. Salmon's interactive fork model can be easily used to describe the last part of the
coincidences x-shape.
Finally, we can conclude not only there is a connection between causal absolute
coincidences and the Principle of Causality, according to which whatever begins to
exist has a cause, but also the possibility to describe them in terms of some causal
model.
115
References
[1] Julienne Junkersfeld, M. (1945) The Aristotelian-Thomistic Concept of Chance. University of Notre
Dame.
[2] Monod, J. (1970) Le Hasard et La Nécessité. Essai sur la philosophie naturelle de la biologie moderne.
Éditions du Seuil.
[3] Reichenbach, H. (1956) The Direction of Time. University of California Press.
[4] Salmon, W.C. (1984) Scientific Explanation and the Causal Structure of the World. Princeton
University Press.
116