## Sunday, September 28, 2014

### In defense of causes

Students in Early Modern Philosophy seem shocked to learn that the scientific revolution beginning with Galileo was in effect an attack on the notion of causality. They assume that scientific explanations are causal explanations. Appeals to causality, however, did not figure in early modern science. Nor (while common in ordinary affairs) do they figure in our contemporary explanatory practices, whether inside science or out.

Scientific laws make no mention of causes, for instance. Galileo’s law for a freely-falling body

d = 1/2gt2

tells us how far the object has fallen in a given number of seconds. The value for ‘g’ – 32.2 ft/sec2 – cannot be said to ‘cause’ the object to fall. (I propose a way to say that it does.)

In the ‘Deductive-Nomological’ model logical deduction takes the place of causality. You derive the phenomenon in question as a logical conclusion from general laws plus a statement of initial conditions. Indeed, to its advocates it was a virtue that it avoided issues of causality.

Increasingly, scientific explanations take the form of statistical correlations, leaving the question of causality entirely aside. The belief seems to be that, once you grasp the patterns of statistical variation, you have access to everything that it is possible – or necessary – to know.

So, are causal relations real features of the world or are they not? If they are, then explanations omitting them are spurious.

On the other hand, if they are not real, some difficult philosophical questions arise. A formulation of a law of nature is logically contingent. So if we take it to express a natural necessity then (barring the postulation of a Lawgiver) there will be no explanation why the particular law in question is the law. On the other hand if we accept a statistical correlation as the explaining formula, then the fact of the correlation itself cries out for explanation.

To be fair, three important considerations made denying a place for causality in explanation seem a reasonable thing to do – one historical, one epistemic, the third conceptual.

First, causes formed a central feature in Aristotelian natural philosophy. It is easier now to see that the apparent incompatibility between Aristotelian and Early Modern forms of explanation arose from features of a particular historical situation; it isn’t logical or metaphysical. 16th century Aristotelian natural philosophy was routinized and degenerate, but in the 14th century it was still very much alive and fruitful in results. Natural philosophers were mathematicizing Aristotle’s principles of moving bodies. William of Heytesbury, a member of the ‘Mertonian Calculators,’ derived the mean speed theorem usually attributed to Galileo. Jean Buridan improved on Aristotle by postulating that a moving body possessed an impetus. This impetus was proportional to the object’s weight, not identical to it as in Aristotle. It was an enduring property, thus it did not require continued action to maintain it. More importantly, in Buridan’s application of impetus to freely-falling bodies it causes a change the momentum of the body: that is, like Galileo’s factor g, it was an acceleration. To be sure, on Buridan’s account objects with more mass should fall faster. But this was also true in Galileo’s earlier Pisan dynamical theory (1589), which was not a significant improvement on Buridan. It’s now reasonable to claim that the resources necessary to have produced the Scientific Revolution were available to thinkers within the Aristotelian synthesis. Thus it is merely a contingent historical fact that the New Science makes no appeal to causes.

Second, it is reasonable to suppose that, even if there are causal relations, we have no independent cognitive access to them. All we can hope for are empirically discoverable natural laws or statistical correlations. However, it is also reasonable to suppose that we do have such access. The view that we don’t was of course codified in philosophy by Hume and has become one of the deep prejudices in philosophy. Pace Hume, we observe causes quite frequently. As John Searle points out, when a car backfiring makes you jump, you experience the causal relation: you don’t need to experience two backfires to get the connection. Wittgenstein’s advice to philosophers is particularly helpful here: “Don’t think, but look!”

Third, the prevailing debate on causality concerns whether it is a relation between events or states of affairs. However, this is a symptom rather than a cause of the modern avoidance of appeal to the relation. It is a logicizing of the relation, reducing it to a species of entailment. It leaves us unsure about such fundamental issues as whether it is even a temporal relation at all.

Are these considerations a sufficient excuse for continuing to avoid causality? I think not. If there is no fundamental conflict between two fundamental styles of explanation, then causal explanations and the whole panoply of contemporary science can work together. Indeed they should.

The concept of causality I favor would make it not just a relation between events or states of affairs, but between individuals in a number of categories – including events and states of affairs. Abstractly, the properties of an individual N give it causal powers to affect, and to be affected by, other individuals. Those powers would be described dispositionally and functionally. Sometimes N’s causal powers result in effects on individual J and sometimes they don’t, depending on J’s own powers as well as features of the environment.

So Galileo’s g does ‘cause’ an object to fall. It is a measure of an object’s disposition to accelerate. Accordingly, causes don’t ‘necessitate’. Air resistance could affect the distance the object falls in a given time. Unlike Hume, we could still say that N is being affected, still has the disposition, even though it is not manifesting it. Dispositions and functions can be said to be ‘realized’ by the micro-entities of standard science.

The advantage to this explanatory move is that the temptations toward instrumentalism and eliminativism so common in our present explanatory practices would be much diminished, if they do not vanish entirely.

Who’s with me?

Thomas Pyne
Department of Philosophy
Sacramento State

1. Tom, thanks for this interesting post.

I am reluctant to agree that the scientific revolution involved a wholesale attack on causality. I believe that the scientific revolution was a rejection of teleological causation, but the dominant metaphor of the Newtonian age was that of mechanism. Bodies in the Newtonian worldview are understood to exert forces, or otherwise act upon other bodies. Newton's first and third laws of motion make explicit reference to this. As I understand him, Hume was challenging the widely held view that Newtonian science had laid bare the causal structure of the world.

I am also reluctant to agree that scientific explanations that make no mention of causes are not causal explanations. On the traditional account, scientific explanations make essential reference to laws of nature. The concept of causation is what we have traditionally used to distinguish a law of nature from an accidental generalization.

I think you are right that Hempel's D-N model of explanation originally attempted to substitute logical relations for causal relations. But it is easy to get the wrong idea about how this worked. (I do not accuse you of this wrong idea.) Hempel was trying to develop an epistemic model of explanation, and he located explanatory power, which for him was essentially predictive power, in the deductive apparatus rather than in the laws. But Hempel's model did not essentially involve the rejection of causation. The covering laws were generally understood to be causal or "law-like", not accidental.

Even scientists and philosophers of an instrumentalist persuasion needn't reject the view that the world has a causal structure (i.e., that there are laws of nature.) What they collectively doubt is whether we are ever in a position to claim that we have discovered such laws. They endorse a non causal conception of explanation because they think it is evident that science has improved our understanding of the world (which is what explanation aims at) even though it is not evident that we have produced theories that have gotten us closer to knowing the underlying causal structure of the world.

Put differently, many would simply disagree when you you say:

"So, are causal relations real features of the world or are they not? If they are, then explanations omitting them are spurious."

They endorse a concept of explanation that is agnostic with respect to the degree to which we have penetrated the causal structure of the world.

I also have a question for you. If you are right that the scientific revolution involved the rejection of causation, and you are also right that this is an error that needs to be corrected, how do you understand the fact that the scientific revolution provided the greatest qualitative leap in human understanding that the world has ever witnessed? Do you claim that the revolution would have been even more powerful if causation had not been abandoned? Do you believe that the failure to make explicit reference to causation is currently impeding fruitful scientific research?

2. Hi Tom, thanks for this provocative piece! You said: " Increasingly, scientific explanations take the form of statistical correlations, leaving the question of causality entirely aside. The belief seems to be that, once you grasp the patterns of statistical variation, you have access to everything that it is possible – or necessary – to know." It was not totally clear to me if you were talking about contemporary science or not. If you were talking about the science of today, then I think you've gone too far here.

I agree with this from Randy: " Even scientists and philosophers of an instrumentalist persuasion needn't reject the view that the world has a causal structure (i.e., that there are laws of nature.) What they collectively doubt is whether we are ever in a position to claim that we have discovered such laws. They endorse a non causal conception of explanation because they think it is evident that science has improved our understanding of the world (which is what explanation aims at) even though it is not evident that we have produced theories that have gotten us closer to knowing the underlying causal structure of the world."
If I just presented the results of statistical correlations in my social scientific papers,  then the editor would reject my paper without review. "Where is the discussion of the results? ", she would say. Sure, I'm not supposed to claim likely casual relations unless I can back them up, but it is expected that I mention which casual theories appear to be consistent with my correlations, and those which appear not to be. This expectation is likely based on the idea that we want to know how the world works as well as how it is at a point in time. In modern science we do want to uncover causes, we are just (justifiably) cautious about claiming casual laws when our methods of investigation are not exact.
Having said all this, scientific investigation by data mining (using big data) might be a useful example if you wanted to argue back.

1. Dan, this point about big data is an interesting one. In their book Big Data: A Revolution that Will Transform How We Live Work and Think, Mayer-Schonberger and Kukier argue explicitly that generating causal models is just a kind of heuristic that we have had to use until now because we had so little data at our disposal. In other words, a causal model just helps us to generate new data with a slightly higher likelihood of being valuable to us. In the age of Big Data we can dispense with the intrinsic limitations of sample size and work with "n = all."

Regarding Tom's thesis, two points are worth making. (1) This is, in fact, a highly controversial thesis in the scientific community. (2) In the second half of the book they actually make it clear that they are not at all rejecting the importance of causal modeling per se, but more pointing out that we need to let computers do the work of sorting through the data to find the interesting correlations first.

I appreciate your point above. My impression has been that it is nearly impossible to get scientific papers published when all people have done is identified statistically significant correlations, but have no causal model to propose. This actually strikes me as pretty dumb.

2. Yeah, I read that book. Very interesting!

Science journals could publish "bare" results... and I suppose they do in rare circumstances. But usually, and reasonably I think, the editor might expect that the people who found the result might be in the best position to retest it or otherwise explain it. Furthermore, inability to say what causal theories it is consistent with betrays a lack of knowledge of the field.
But, if scientific publishing was completely open access (including open reviews) and much more collaborative, then publishing just results would work well. Other scientists could propose and debate competing explanations in the open, further evidence would be wielded, and (closest to) truth will win out! There is at least one journal that is truly open like this, and Nature trilled it http://www.nature.com/nature/peerreview/debate/nature05535.html

3. Yeah, excellent point. When I said that the current policy was dumb, that's actually what was in the back of my mind. Not just open access, but reforming the whole industry so that all experimental results end up being published, not just "interesting ones," which, I think, largely explains Ioannidis' point that most new published findings are wrong.

3. Randy and Dan, thank you for your comments. It’s gratifying to have a piece I was, quite frankly, fairly tentative about given a serious, critical hearing.

I’m operating with Nancy Cartwright’s distinction between two general kinds of laws: laws of association and causal laws. My point was that laws of association have displaced causal laws more or less completely since the early modern period.
Laws of association, after all, are the way we generally state physical laws. They assert how qualities or quantities are associated. Cartwright’s example: “Whenever the force on a classical particle of mass m is f, the acceleration is f/m.”

I take these not to involve appeal to any causal relation. To cite Cartwright, “…(T)hey provide no account of what makes things happen.”

I also take it that Bertrand Russell was merely defending prevailing doctrine by arguing that laws of association are all the laws there are, and that causal principles cannot be derived from the causally symmetric laws of association.
And I’m agreeing with Cartwright that causal principles cannot be reduced to laws of association and that they cannot be dispensed with.

To Dan’s point, to say that a given causal relation is ‘consistent’ with a set of results means (am I getting this right?) that (i) the relation being claimed for the causal relation is heuristic, a way of interpreting the results, and that it can be one way among others, because (ii) causal relations are inaccessible except via the associations manifested in the results – or in others relevantly like them. In short the claim is a version of the second reason I gave for avoiding appeal to causality.

Dan indicates that we are “(justifiably) cautious” about claiming causal (non-associative) laws. That would be true on my view, but not on Dan’s. If we have no independent cognitive access to causality except via associations, then caution would be out of place. There’s nothing to be cautious about. However, if as I claimed, we sometimes directly experience causal relations, or can ascertain the presence of a causal relation on other grounds, then caution becomes an admirable precept in the making of causal judgments. Experience, after all, is fallible. (Am I getting your point right, Dan?)

And Randy would be right that the demand for a causal model in addition to the results is dumb. Associative methods of sufficient power, as promised by Big Data, would show that the causal relations, construed thus, do no real work.
Consider that associative law regarding force. It’s a transform of the classical ‘f = ma’: force equals mass times acceleration. Suppose I hold a book in my hand with my arm bent 90 degrees at the elbow. Why isn’t the book falling? Because the force of gravity is being exactly compensated for by the upward force exerted by my arm. Which is beginning to wobble…

Where are the compensating forces, however, on the associative law? The acceleration up is zero; the acceleration down is zero. Thus the value for the mass times acceleration is zero. But the forces are surely real, and I am experiencing them, as the pained grimace on my face reveals.

1. I'm understanding this better now. I can see how my point seems like it falls under Tom's second reason to ditch casual claims in science. I don't think my caution is unwarranted though. And I don't think causal theories are spurious on my view. If we take the associations found in the perfect Big Data data set for exactly what they are worth, then they can tell us an awful lot about what has happened and what happenings have happened together. But, all those associations cannot tell us anything at all about future events unless we assume that the associations of the past will closely resemble the associations of the future. The best reason to make this assumption is that there is a real casual structure underlying the associations.

Regarding Randy’s Success of Science argument: “…(H)ow do you (Tom) understand the fact that the scientific revolution provided the greatest qualitative leap in human understanding that the world has ever witnessed.” Since this is not a thesis in dispute, I am obliged to give my understanding. Here it is.

Agnosticism is a justified in the absence of any reasonable way of gaining access to the disputed reality. If we have reason to think that causes are operating in the world, then an explanation omitting them will not tell the story of the world correctly.

Additionally, postulating a causal relation is probably an indispensable step in determining what associations to pursue in research. If we have other means of access to causal relations than the associative laws themselves, this would yield a nice mix of roles for the two kinds of laws. The now-traditional associative laws become the heuristic tools. While they are, as Cartwright points out, all false, they do reveal not-otherwise-easily detected associations between entities precisely through their radical, falsifying idealizations.

My proposal, I’m sorry to say, is even more radical than this. The sort of causality I advocate reviving is formal causality: doing science by attributing causal capacities and powers to objects in virtue of their kind and constitution. For one thing, this avoids the tradeoff Cartwright sees in associative laws between accuracy and explanatory power.

Thank you both. (Tom keeps digging….)

5. Tom, thanks for this reply. I think I am seeing more clearly what you are trying to work out now. Here are just a few follow-up thoughts for the mill.

(1) It seems to me that the distinction between causal processes and causal relations is useful here. Russell, at one point anyway, disparaged causal relations as an Aristotelian confusion that philosophers of the 20th century had not yet unlearned, but he fully endorsed causal processes and causal laws, and I think that may be a reasonable characterization of science today. Anyway, I think I now understand that you are advocating for the reality of causal relations, not just causal processes.

(2) As you mention in your original post, causation is what we appeal to in explaining why an empirically established association is in fact a natural necessity. (Or, if you are an instrumentalist, why, even though you doubt that any particular association expresses a natural necessity, why you think the order we find in the world is not an accident.) However, in agreeing with that, I don't think we are thereby committed to a relational notion of causation between individuals. I don't think we need causal relations to explain causal processes. I speculate that causal processes might be useful in explaining the appearance of causal relations, though.

(3) Tell me if I am wrong, but I find myself interpreting this post in the light of some of your earlier ones, as part of a general defense of the reality of the properties, entities and relations that populate what Sellars called the Manifest Image. I personally think there is a huge amount of value to working out the logic and ontology of the MI, not just because it is intrinsically interesting, but because it captures the way ordinary folks think, and therefore can help to clarify and improve thinking done within it. It is also an essential project in artificial intelligence. But none of that commits us to the fundamental reality of the MI. To connect with my post, immediately preceding this one on Explanation and Illusion, these may be accepted facts that are better explained as illusions. For me, the MI is not the operating system of the world, but a highly simplified user interface, which uses a bunch of simple but incredibly effective heuristics to allow us to intervene successfully in the world.

(4) Cartwright is certainly a worthy ally. As my last point suggests, I'm inclined to subscribe to an interventionist notion of causality, as defended by the likes of Woodward and Glymour (and strongly criticized by Cartwright). I think one of the virtues of the interventionist approach is that it explains very well what you say about postulating causal relations being an essential step in research, which is all about trying to figure out what will happen if you manipulate some variable this way or that.

6. Tom, some further noodling regarding the two concrete examples you appeal to above:

(1) Searle's backfire example.

Searle says that when you hear a backfire, you experience the causal relation, in support of which he says you don't need two backfires to get the connection. I don't know if I understand this, but here is how I see it.

When a person experiences a backfire for the first time, s/he is clueless about the cause. It is just a loud noise that makes her jump- F---, what the hell was that??-, and inquiry is required to determine its source. Once she has experienced a backfire, and assimilated the theory, she will, I agree, more or less instantly interpret that kind of report as a backfire. I think it is fine- accurate within the conceptual scheme of the MI- to say (once we have the theory) that we hear a backfire. I also agree that to experience a backfire is to experience it as having been caused by the car. That's what a backfire is. But I experience it this way only because I have assimilated a simplified relational account of what is actually an incredibly complicated causal process. It is fine for the purposes of ordering the salient events of every day life, but only the crudest of guides to the causal structure of the world.

(2) The gravity example

You point out that when you hold a book motionless above the ground you experience a force.( In Newtonian theory this would be the force predicted by the universal law of gravitation or, alternatively, by Galileo's g for earth according to the second law F = ma, where g = a.) And you suggest that this is inexplicable on the these associative laws because the acceleration up and down is 0, implying that the force is 0.

I think the answer to this is to run the argument in reverse. The forces are real, both on the book and on you, therefore it follows that both you and the book are in fact accelerating. This seems like an absurdity, but it is not, because in Newtonian mechanics, force, acceleration and velocity are vectors. So in fact, if the force downward is exactly matched by the force upwards (which, of course, is what we always experience standing motionless on earth) the vectors will sum to 0, and consequently there will be no observed motion, even though both objects are in fact accelerating.

I think this makes a little more intuitive sense in General Relativity, where the equivalence principle makes it explicit that there is no difference between an accelerating reference frame and a motionless one in a gravitational field.

For me, examples like this help to underscore the fact that leaps in scientific understanding typically do a lot of violence to ordinary intuitions and concepts. The most common example in this area is, of course, Aristotle's intuition that a cause is required to keep a body in constant motion.