Newsletter No. 35, July 4, 2000
CAUSATION: A REALISTIC ACCOUNT*
Wesley C. Salmon
Paper presented at the Kyoto Colloquium for the Philosophy of Science
Causation: A Realistic Account
This talk began with a reminder that causality plays an important role in most areas of philosophy, e.g., metaphysics, epistemology, history of philosophy, ethics, aesthetics, philosophy of language, and philosophy of science. The only exceptions I can think of are pure deductive logic and philosophy of pure mathematics.
In previous publications (esp. "A New Look at Causality" in Salmon, 1998), I offered an analysis of causality in which cause and effect are not fundamental concepts. Instead, I recommend temporarily abandoning the corresponding terms. In their place, I introduce the concepts of causal process, causal interaction, and causal transmission as basic.
The term "process" is a primitive term introduced ostensively by examples, e.g., a shadow moving across a wall, a pulse of light traveling from a star to us, a billiard ball (moving or stationary). Causal processes are distinguished from pseudo-processes by the fact that the former transmit some conserved quantity (such as momentum, energy, or electric charge) while the latter do not. A causal interaction occurs when two processes intersect and exchange a conserved quantity. Hume's famous billiard balls exchange momentum when they collide and exhibit changed states of motion after the collision. A process transmits a conserved quantity if and only if it possesses that quantity through a stretch of its history without replenishment from external sources (i.e., in the absence of other interactions). I take these concepts to provide a basis for a fully objective and realistic theory of causation.
My attention was recently called to the fact that I had never explained how the concepts of cause and effect are to be reintroduced following the above-mentioned analysis. It turned out to be a much bigger task than I had imagined. I worked on it in Kyoto and presented the tentative results to the Kyoto Colloquium in May. What follows is a condensed version of the new material. Among other things, it involves introduction of a new concept, namely, a complete causal structure.
1. Considering the Reintroduction
At the beginning of this discussion, I suggested that we temporarily abandon the terms "cause" and "effect." The time has come to consider reintroducing them. Although the reintroduction may seem simple, we shall see that it takes a bit of doing. Given two events, spatiotemporally separated from one another, under what circumstances can we say that they are directly connected as cause and effect? Although it is rather easy to identify some necessary conditions for a direct cause-effect relation--e.g., that there exist one or more causal processes connecting them--it is much more difficult to find sufficient conditions. It seems to me that there are at least two basic patterns that we naturally take to qualify as direct cause-effect relations.
The first and simplest is Hume's famous billiard-ball collision. He called it "as perfect an instance of the [first type of] relation of cause and effect as any which we know either by sensation or reflection." This is just a causal interaction between two processes. The interaction produces changes in both processes, but because Hume was more interested in the change from rest to movement on the part of the second ball, he said that the collision caused the motion of the second ball. In a slightly different context, we might be more interested in the subsequent motion of the first ball. Suppose the game is pool rather than billiards; the balls used in the two games are the same, except for their markings. The pool table, unlike the billiard table, has pockets. The cue-ball is initially put in motion by the player; the aim is to make it strike one of the object-balls in such a way that the object-ball will drop into a pocket. If, however, the cue-ball drops into a pocket, that is called a "scratch," and the shooter is subject to a penalty. If a scratch occurs, it would be natural to say that it was caused by the collision with the object-ball--i.e., the interaction caused a certain kind of motion on the part of the first ball, the one that was initially in motion rather than the one originally at rest. In interaction of this kind both of the incoming processes are modified, and both of these modifications are effects of the interaction.
The second type is a bit more complex. This cause-effect pattern consists of a causal interaction between two processes, a causal process that emerges from the interaction, and a subsequent interaction of that process with another process. The interactions can be considered events, so this pattern consists of two events connected by a causal process. For example, a child hits a pitched baseball with a bat, the ball changes its direction and flies toward a window, and then the ball strikes the window causing it to shatter. To quote Hume, "This is as perfect an instance of the [second type of] relation of cause and effect as any which we know either by sensation or reflection." In common parlance, the kid hitting the ball with the bat causes the breaking of the window (the effect). In this case, the bat and the ball are two separate processes that intersect and interact causally. The ball is a causal process that moves from the bat to the window. The causal process constituted by the flying baseball intersects and interacts causally with the process constituted by the window pane. Less formally, why did the window break? Because the kids were playing baseball in the vacant lot next door, and Kim hit a fly ball that crashed through the window. The traveling baseball is the causal connection between the striking of the ball and the shattering of the window. This causal process transmits mass, momentum, and energy from one event to the other. It is the causal connection Hume sought in vain.
My fundamental thesis is that every instance of a cause-effect relation involves a more or less complicated pattern of causal processes and interactions. However, realization of such patterns is patently insufficient for an instance of either type of cause-effect relation. Recall the baseball example. Suppose that, just as the ball is about to strike the window, the batter, anticipating what is about to happen, shouts "Oh, no!" (or words to that effect), so that the sound waves from the exclamation reach the window just when the ball does. We recognize immediately that, although both the sound wave and the ball carry linear momentum, the momentum of the sound wave is insufficient to break the window. A clear constraint here is the conservation of linear momentum. When we think that a conservation condition might be violated, we withhold the judgment that a cause-effect relation obtains. It appears that something essential is being left out of the story. Nonviolation of conservation relations is another necessary condition. If there is any doubt about it in this case, empirical experiments can confirm that baseballs break windows, whereas normal sounds do not.
2. The Problem of Context
The major obstacle to the creation of a fully objective and realistic theory of cause-effect relations is the fact that the instances we tend to select are highly context dependent. Take Hume's billiard-ball example again. I have already remarked on the difference between the viewpoint of the player of billiards and the player of pool. This suggests that our interests are a function of the rules of the game, and these are human constructions. Moreover, they depend on the proficiency of the player. In either game, the spin on the cue-ball is a matter of serious concern, because it largely determines what will happen to the cue-ball after the collision. This means that angular momentum, in addition to linear momentum, must be taken into account. The tyro, like Hume's player, tends not to look ahead.
Context also figures conspicuously in the baseball example. As the ball travels from the bat to the window, it undergoes a great number of causal interactions with the molecules along its path through the air. Since the people involved are interested in the general form of the interaction between the ball and the window, not with the precise trajectory of the ball, they will ignore these interactions. If, however, they were thinking about the curve thrown by the pitcher, it would be essential to take interactions with the air into account.
For another example, consider a barn that burns down. Here we have only a more complicated structure involving processes and interactions. A lighted cigarette falls from the fingers of a passing tramp onto some dry straw on the floor. In the interaction between these two processes, energy is exchanged, raising the temperature of the straw to its ignition temperature. The burning piece of straw releases energy that ignites neighboring pieces of straw. A conflagration ensues in which the burning straw ignites neighboring pieces of wood; in this fashion the entire barn is consumed in flames.
This example illustrates J. L. Mackie's (1974) well-known concept of the INUS condition (an insufficient, but non-redundant part of a condition that is unnecessary but sufficient condition of the effect). An INUS condition is a cause in ordinary usage. In this case, the dropping of the burning cigarette is the cause of the barn burning down. There is no question that the selection of one INUS condition or another is highly pragmatic and context dependent. In some cases, such as a spark from a workman's torch, the presence of combustible material might be singled out as the cause. The full cause, according to Mackie, is a disjunction involving such terms as the dropping of a burning cigarette, a spark from workmanﾕs torch, being struck by lightning, spontaneous combustion of hay stored in the barn, and a deliberate act of arson. Each of the disjuncts contains enough other items to constitute a sufficient condition for the effect. This complex formula states a necessary and sufficient condition for the effect, i.e., the burning of the barn.
Mackie's account requires two comments. First, in many cases, and this is no exception, we are not given complete sets of disjuncts. For instance, burning material blown onto the roof from a nearby forest fire was not mentioned. Mackie explicitly admits that the formulations of full causes typically contain "elliptical or gappy universal propositions" (ibid., p. 66); the gaps are the reflection of sufficient conditions that are unknown at any particular time. It is difficult to see how such propositions can constitute objective descriptions of full causes; they are epistemically relativized to the knowledge situations of the persons involved. Second, Mackie explicitly maintains that the statement of the full cause formulates conditions both necessary and sufficient for the effect against the background of a causal field. Items in the field--standing conditions--cannot qualify as causes. Among other things, the field might cover such conditions as the presence of oxygen. Again, there are special circumstances--a particular laboratory experiment, for example--in which the presence of oxygen would be singled out as the cause. It seems obvious that the selection of the causal field is guided by pragmatic considerations, and is, therefore, context dependent.
In view of the foregoing examples, as well as many others, I conclude that cause-effect statements are almost always--if not always--context dependent. (It is easy to find arguments to support context-dependency; the hard part is to locate the context-independent factors.) If this conclusion is correct, it means that most cause-effect statements lack full objectivity. It seems clear that Mackie has not succeeded in finding causation "in the objects." Given this conclusion, it appears that we face a serious problem in the attempt to provide a realistic account of causality. Nevertheless, I think it can be done, but only by going to a different level.
3. Complete Causal Structure
The concept of a direct cause-effect relation involves many subtleties. In order to facilitate its further clarification, I shall introduce the concept of a complete causal structure. Having the concepts of causal process, causal transmission, and causal interaction at our disposal, we can do so quite straightforwardly. The complete causal structure of any convex chunk of spacetime--i.e., of the universe--is given by the entire network of causal processes and causal interactions contained in this selected region. It must include an account of the conserved quantities transmitted by the processes and of those exchanged in the interactions. Assuming that we are dealing with parts smaller than the entire universe, we will have to take account of the processes entering or leaving that section, and with the conserved quantities they bring in or take out.
My notion of a complete causal structure is closely related to ideas of Peter Railton (1981) and Christopher Hitchcock (1993). In his discussions of scientific explanation, Railton formulated the concept of an ideal explanatory text, which would include all facts, no matter how insignificant they might seem, that are in any way relevant to the explanandum under consideration. In the case of causal explanation, the ideal explanatory text would be essentially the same as the complete causal structure. The fundamental difference between them is that Railton's construction is a text (a linguistic entity) whereas my complete causal structure is a complex physical entity. Railton stated emphatically that seldom--if ever--do we attempt to write out the ideal text; rather, we seek to illuminate particular parts in which we happen to be interested. In order to do so, we attempt to furnish explanatory information about some part or aspect of the ideal text. Selection of the part to be illuminated is clearly a pragmatic matter, and different investigators will choose different aspects to examine and different levels of detail.
In his work on probabilistic causality, Hitchcock regards probabilistic causal statements as revealing features of particular interest in an underlying probability space. This underlying probability space is, of course, an abstract mathematical construction, but, like many other mathematical abstractions, it is applied to the physical world. Again, the decision regarding the parts to which causal language is to be applied is a highly pragmatic matter.
One basic characteristic shared by Railton's ideal text, Hitchcock's probability space, and my complete causal structure is their objectivity. They are correct or incorrect without regard to context or other pragmatic considerations. This is a fundamental aspect of my claim to furnish a realistic account of causality. The complete causal structure is a fact of nature that exists quite independently of our knowledge or interests; it is not epistemically relativized. It is an extremely complex entity, but that is because the world is extremely complex. Statements about the relations between causes and effects are usually highly selective, and they are typically context-dependent. For example, when we discuss Kim's breaking of the window in the course of the baseball game, we ignore the many collisions of the ball with molecules in the air; it is sufficient to take account of the momentum of the ball after it has interacted with the bat. The collisions with the molecules are part of the complete causal structure, but they are not germane to the story.
From the outset, Mackie's goal was to find causation "in the objects"--i.e., to provide a fully objective, non-mind-dependent, account of causation (1974, pp. 1-2). His effort produced three items: (1) the concept of causal priority, specified without reference to temporal priority, (2) the concept of an INUS condition, and (3) the concept of the full cause. I have shown that his analysis of causal priority in terms of causal fixity is entirely untenable. However, Reichenbachﾕs conjunctive forks might be imported to determine causal priority. That is the method I adopt in this paper. The concept of an INUS condition, it seems to me, provides a useful tool for the analysis of cause-effect relations, but it is doubtful that such causes can be fully objective and context-independent. Mackie was entirely clear on this point. He hoped to find objectivity in the full cause, but his invocation of "elliptical or gappy universal propositions" (ibid., p. 66) gives the strong appearance of epistemic relativity or context-dependence. In the pages that follow immediately (pp. 67-75), he makes a good case for the practical utility of such universal propositions; this shows that they have significant pragmatic virtues, but it does not make a convincing case for full objectivity.
Throughout The Cement of the Universe, Mackie makes many casual references--more or less in passing--to causal processes, causal mechanisms, interactions, and common-cause configurations. What he failed to see, in my opinion, is the fact that these are precisely the kinds of entities that form the objective basis for our causal claims. Moreover, they manifest another of Mackieﾕs main desiderata, namely, the capacity to distinguish between causal and noncausal sequences of events (ibid., p. 29). In the foregoing discussion, I have taken care to show how we can make objective distinctions between causal processes and pseudo-processes and between causal interactions and mere spatiotemporal intersections. My conclusion is that Mackie conducted his analysis on a relatively superficial level, where pragmatic considerations and context-dependence play legitimate roles, but that he virtually ignored the objective noncontextual grounding which underlies his level of analysis.
In practical situations, there is great latitude in deciding what constitutes a single process or single interaction. To the astrophysicist, Earth, traveling in its orbit, is a single process for most purposes. To the geophysicist, the history of our planet is a complex set of interactions and processes. To a casual observer, a meeting of two people might be a single interaction; to those involved, the meeting might have considerable social and psychological complexity. In the complete causal structure, all of the details are present, and the complexity of some processes that we often regard as single processes is exhibited. In fact, the complete causal structure reveals the extent to which it is permissible to treat complex processes and interactions as simple entities for various purposes. As I see it, the complete causal structure is limited to situations that do not involve quantum phenomena, where it is well known that causal stories encounter severe difficulties. Below the complete causal structure, there is, so to speak, a quantum mechanical substructure. I do not profess to understand this substructure, beyond being strongly convinced that quantum mechanisms do not conform to the specifications of normal causality. This happens when the wave-particle duality of light or matter enters the context in any significant way.
4. Cause-effect Relations
I do not think it is profitable to try to define such terms as "cause" and "effect" in any precise way. They are part of the common idiom; they are used quite loosely and are highly context-dependent. Nevertheless, statements about cause-effect relations furnish valuable objective information about the world with respect to our wide range of purposes, interests, and background knowledge. I cannot give an exhaustive list of such applications, but a few examples should convey the main point.
The most obvious context is a situation in which we hope to exercise control by furnishing causes that bring about desirable results or by eliminating conditions that lead to undesirable situations. We search for the causes of airplane crashes and for the means to prevent or cure diseases. The common-cause configuration is crucial in ascertaining what measures should be taken to overcome medical, psychological, or social problems. Treating the symptom rather than the disease frustrates attempts at amelioration. Causal knowledge of this sort is so much a part of daily life that some philosophers (e.g., Gasking, 1955; von Wright, 1971) have tried to characterize causality in terms of manipulability. While manipulability is undeniably an important aspect of causality, I agree with Mackie in viewing a general manipulability account as excessively anthropocentric.
Cause-effect relations figure prominently in the assignment of moral or legal responsibility. My only experience as a juror was on a case in which spoiled food was the alleged cause of the illness of several members of a family. The case was terminated by a settlement between the parties outside of the courtroom. This outcome was somewhat frustrating because the terms of the settlement are confidential. It seemed to me to be an almost perfect real-life case for application of Mill's methods. Everyone recognizes the importance of ascertaining the cause of death when murder is involved. Cause-effect relations pertain crucially to the transmission of information. If the sender is human, this is just one form of manipulation. We understand how to produce and send messages. However, the message need not be the result of a voluntary act of any intelligent being. Astronomers who analyze the spectra of light from celestial sources receive information about the chemical constitution of these bodies.
The objective basis for the use of cause-effect relations is that, no matter how complex the case, it is fundamentally reducible to a network of causal processes and interactions. Usually, however, much of the complexity is dispensable. Even such an ordinary act as starting your car is quite complex. After the key has been inserted into the ignition switch, your hand turns the key, which results in the closure of an electrical circuit. That, in turn, permits an electric current to flow from the battery to the starter and to the ignition system (which yields sparks to ignite the fuel in the cylinders). In addition, the fuel injector must supply the fuel. The sparks and fuel injections must be timed in a proper sequence. Even this description presupposes a rich causal field of background conditions. This combination of processes and interactions is embedded in a real--and much more complex --network of processes and interactions. One of the most significant uses of our knowledge of cause-effect relations is to separate useful from useless information in a given context.
This kind of separation enables us to make practical use of such techniques as Mill's methods and controlled experiments to investigate cause-effect relations. It enables us to treat complex combinations of causal interactions and processes as single processes in appropriate contexts. It enables us to treat large assemblages, such as the gas in a container, as a single unit for some purposes; e.g., heating causes an increase of pressure in a container of fixed volume. The cause-effect approach often enables us to avoid innumerable useless details. These are pragmatic virtues; it is no surprise that they are context dependent.
In a different context, such as the study of Brownian movement, the individual interactions between small numbers of molecules and individual pollen particles are essential. This illustrates a common research strategy, namely, to look at phenomena on a smaller scale when the larger scale view is unsuccessful. The details are present in the complete causal structure; they can be exposed when it is advantageous to do so. This particular example had crucial importance, around the turn of the twentieth century, in the debate between energeticists and kinetic theorists in thermodynamics.
At the beginning of this paper, I stated the aim of providing a realistic account of causation in the sense that causal relations should have the same status as ordinary middle-size material objects. Obviously, we observe such things as billiard balls, baseballs, bats, windows, and children. These are causal processes. There are, of course, causal processes, such as electromagnetic radiation in the radio/television frequency range, that we cannot observe directly. There are also ordinary material objects, such as viruses and molecules, that we cannot observe directly. We observe such causal interactions as collisions of billiard balls and shattering of windows. There are also collisions of individual molecules that we cannot observe directly. Causal processes and interactions thus seem to be on a par with ordinary material objects--both observed and unobserved.
It should be noted that, in observing objects and events of the types just mentioned, we may not be observing them as causal processes or as causal interactions. Nevertheless, the discussion has amply shown, I believe, the kinds of empirical experiments that can be performed to ascertain whether a process is causal or an intersection is an interaction. Performing such experiments does not violate the principles of Humean empiricism.
Although my 'official analysis' appeals to conserved quantities at the fundamental level, I do not mean to suggest that the richness and complexity of the world are exhausted in these properties. I have taken causal transmission to require that some conserved quantity or other be present in a causal process, but causal processes transmit many other kinds of properties as well. Hume's billiard balls transmit color, solidity, elasticity, and shape. They transmit blue chalk marks that come from the cue stick. Baseballs transmit standard patterns of stitching in their covers and the trademarks of their manufacturers. Bullets transmit characteristic marks of the guns from which they were fired. Causal processes are, after all, the channels of communication that we find in the physical world quite apart from human thoughts, intentions, and desires. Electromagnetic radiation exists in nature; we have learned to use it to transmit information, music, entertainment, and commercial messages. We have learned to make paper, to write on it, and to convey our thoughts and feelings. The term "communication" is not meant to include only transmission of information by intentional acts. Fossils inform us of types of organisms that lived on Earth at much earlier times.
There is one final caveat. In no sense have I been attempting to provide a conceptual analysis of causation that would apply to all possible worlds; I have no idea what such a thing would be or how it could be done. Indeed, I do not mean to suggest that causation, as here characterized, applies to all domains in our actual world. I do not believe that what might be called "normal causation" applies in the domain of quantum mechanics--the domain in which wave/particle duality manifests itself.
Moreover, I make no claims to possess an account of mental causation, if there is such a thing. I have speculated--and it is no more than speculation --that all of the causal processes and interactions related to our conscious experience occur in the brain, and that consciousness is a collection of pseudo-processes. This form of epiphenomenalism makes consciousness an analogue of what we see on the screen of the cinema or television set. The main problem is that I have no idea what constitutes the 'screen' during conscious awareness.
I realize that the theory I am proposing has a highly reductionistic flavor. It seems to me that my account should hold in the natural sciences--including biology, but not quantum mechanics. I am not confident that it is suitable for psychology and the social sciences. In the preceding paragraph, I mentioned my reservations with respect to psychological phenomena. Where interpersonal relations are concerned, I would claim only that causal processes, transmission, and interaction are necessary for recognition of other people and communication among them. The importance of such causal mechanisms should not be overlooked. Words, spoken or written, can break someone's heart, gladden someoneﾕs day, or incite someone to violence. Printed messages can amuse, educate, or mislead. Whether other kinds of causation are involved in social intercourse I leave as an open question for philosophers of psychology and of the social sciences.
My aim has been to examine causality at what might be characterized as the 'deepest metaphysical level'. The account that has emerged removes this concept from the field of metaphysics and transports it to physics. If this goal has actually been achieved, I count it as philosophical progress.
Gasking, Douglas, 1955, "Causation and Recipes," Mind 64, pp. 479-487.
Hitchcock, Christopher Read, 1993, "A Generalized Probabilistic Theory of Causal Relevance," Synthese 97, pp. 335-364.
Mackie, J. L., 1974, The Cement of the Universe. Oxford: Clarendon Press.
Railton, Peter, 1981, "Probability, Explanation, and Information," Synthese 48, pp. 233-256.
Reichenbach, Hans, 1956, The Direction of Time. Berkeley and Los Angeles: University of California Press.
Russell, Bertrand, 1948, Human Knowledge: Its Scope and Limits. New York: Simon and Schuster.
Salmon, Wesley C., 1998, Causality and Explanation. New York: Oxford University Press.
Von Wright, G. H., 1971, Explanation and Understanding. Ithaca, NY: Cornell University Press.
(c) Wesley C. Salmon
Editor's Note: Wes Salmon has been teaching a course on causality here at Kyoto; and he kindly contributed one of his up-to-date results to our Newsletter. The editor, on behalf of all those who attended his course, and who listened to his talk at the Kyoto Colloquium, wishes to express warm thanks to Wes.
Last modified Nov. 30, 2008. Soshichi Uchii