These notes and drabbles represent summaries and responses to readings, as well as some of my very unpolished thoughts on causation.
20090117
We do not use the word "cause" ("causation", etc.) to mean only one thing. Part of the reason it is difficult for philosophers to come up with a simple, coherent account of causation is because multiple ideas are conflated by a single term. (Perhaps Aristotle got this right.) This becomes clear when considering our motivation for determing causes and the kinds of things we appeal to in causal statements. Causal statements of physics (e.g., one billiard ball hit another and caused it to move) are not the same kind of thing as statements of moral responsibility (e.g., he caused the window to break and should be held accountable) nor are either the same thing as statements of efficiency (e.g., I slammed on my brakes because the car in front of me stopped) nor telos (e.g., I bought beer because you like it).
A distinction can be made between singular particular causes (A caused B) and general universal causes (C causes D). Our motivation for determining causes of the one is different than our motivation of the other, and correspondingly our explanations take a different form (e.g., whether they point to universal regularities).
While distinct, singular causation and general causation are related. In particular, we find general causes by exploring singular causes and are useful because they enable predictions of future singular causes. Not surprisingly then, explanations of singular causes often make reference to statements about universal causes. (On one level, this is nothing more profound than saying that if G={g1,...,gn}, then G and gi are distinct but related.)
Correlation captures a lot of what we want out of general causation: it allows for predictions and manipulations. But there is a difference between causation and correlation (both metaphysically and epistemologically). We think of correlation as the lesser of the two, and once we have established correlation between two things, we are not satisfied, but instead attempt to determine causation (either by determining which caused the other or by looking for a third cause that underlies them both). Still, it's hard to say precisely wherein lies the difference. Additional conditions which seem obvious at first glance (e.g., causes precede effects) are hard to pin down. (And of course there's the question of whether we can ever know we've determined general causes, since appeal to general causes rests on inductive/abductive reasoning, but that's a question which can be decided separately.)
We want to discover general causes, in part, because we want to be able to make predictions (for similar scenarios in the future), and we want a fairly high degree of confidence that our predictions will obtain. We think of causal statements as not merely predictive, but also explanatory. We are looking for explanations of why and how things are, not just that they are.
When considering specific causes, it becomes clear that part of our motivation for identifying causes is to assign (moral or legal) responsibility. Cases are never as simple as "A alone caused B", but rather take the form of "{a1,...,an} caused B", and then we look to single out a particular ai which is relevant in some particular way. For purposes of responsibility, it is crucial that we single out causal factors where both the moral agent had reason to expect the outcome (i.e., should have known otherwise) and where the moral agent had the ability to avoid the outcome (i.e., could have done otherwise). This highlights another reason why we look for causes: by identifying causes, we (as free agents), can manipulate our environment to do otherwise.
In short, we want many things out of causation, including predictability, explanability, responsibility, and manipulability.
20090125
The following is based on chapter 3 of Hart and Honoré's Causation in the Law (2nd edition). Hart and Honoré use three scenarios to illustrate the difficulties courts have in tracing out chains of causation that lead to particular outcomes. I found comparing and contrasting these scenarios to be a good starting place for clarifying our notions of, and the relationship between, causation and responsibility.
Disclaimer: Although the following agrees with Hart and Honoré in many important regards, I have restated their scenarios in modern terms and much of the discussion of the scenarios is my own work. In other words, this is deliberately not a faithful summary of their work. Further, any discrepencies or sloppy thinking should be attributed to me and not them.
Scenario 1:
Suppose this scenario is brought to a (Anglo-American) court, whose job it is to decide the cause of the forest fire. What does this mean? For legal purposes, this means the court is determining whether Joe (a person, not an action) caused the forest fire and whether he can reasonably be held responsible for the outcome of his actions.
In his defense, Joe says "I just disposed a cigarette near the forest, not in the forest. And then this wind came along! The wind caused the forest fire. If it weren't for the wind, the cigarette would have just burnt out." Well, it does seem that Joe had a bit of bad luck. We can easily conceive of the scenario Joe wishes had happened, that in which the cigarette extinguished itself without conflagration. It might even be the case that in the overwhelming number of similar cases, no fire ensues. For the sake of argument, suppose we know that this forest fire would not have been started without the wind. So the wind is definitely a necessary part of the causal description of how this fire came to be.
And yet, a jury (at least a jury in California) would probably think, "Well, that was pretty stupid of Joe. Was he really surprised that the forest caught fire?" After all, it's common knowledge that lit cigarettes can catch (flammable) things on fire and that fires spread rapidly in forests, and as such, people should make sure to put out their ciagrettes before disposing of them near forests. Duh! We think Joe took a thoughtless risk, and that he should have known better and acted more responsibily. A jury would probably hold him responsible for causing the forest fire and explain his culpability in terms of ideas like carelessness.
Scenario 2:
Here the jury most likely holds Bob, not Alice responsible for the fire.
But wait ... what's going on here? These cases are nearly identical, with leading to a forest fire. Alice's behavior is the same in this scenario as Joe's behavior in the previous scenario. Surely, then she is just as careless.
The real difference between these scenarios must lie in the third sentence. Why do we blame Bob here, when we didn't blame the wind before? One reason is that the wind, being non-human, is not the sort of thing that can be held responsible by a court of law. Another reason is that wind is a common occurrence while pouring lighter fluid on trees is not. We believe Joe could have reasonably anticipated wind, wheareas Alice could not have been expected to anticipate Bob and his lighter fluid. A third reason is that Bob's actions were deliberate, with the intent of setting the forest on fire. The jury focuses on Bob's intent, rather than Alice's accident, and shifts all culpability to Bob (after giving Alice a stern "don't do that again").
Scenario 3:
After checking on his student in the hospital, Professor Snape went immediately to the headmaster's office, where Headmaster Dumbledore, Hagrid, and a distraught Hermione had gathered. "I demand Hermione be expelled at once, and that 5000 points be taken from Gryffindor House for Hermione's actions!" Hagrid spoke up at once in Hermione's defense. "That's ridiculous! The usual punishment for punching another student is only 100 points and the loss of Hogsmeade privileges." "Irrelevant", said Snape. "This is no ordinary situation. Just look at the extent of Draco's injuries!"
Indeed, Draco's injuries were quite severe, and Dumbledore conceded that had Hermione deliberately set out to inflict such harm, she would deserve a more harsh punishment. However, he believe it was clear that this was not her intent. Not surprinsgly, Snape was unsatisfied. "If Hermione hadn't hit Draco, he wouldn't have been there on the ground at that exact place at the exact time the tree fell. She meant to cause him harm. You may not think the outcome was Hermione's deliberate intent, but there's no denying that Draco got injured and his injuries are due to her actions. It's her fault, and she should be punished accordingly." Despite Snape's persistent arguments, Dumbledore stood firm in his position: Like it or not, Hogwarts students got in minor fights fairly often, and the initial damage Draco sustained was typically not taken very seriously. The tree caused the majority of Draco's injuries. Draco being there was just an unfortunate coincidence, and Hermione was not to be held responsible for more than the single immediate bruise caused by the impact of her fist with Draco's face. It's unfortunate, but these things just happen.
This scenario looks similar to the first one. Someone takes an action. A separate event of non-human agency occurs. And yet unlike in the first scenario, where we held Joe responsible anyway, here we think Dumbledore is correct to exonerate Hermione. Again, the difference seems to be a matter of what possible outcomes the actor could be reasonably expected to anticipate. Had Hermione punched Draco under the Whomping Willow tree (a tree which is known to attack those who disturb it), Dumbledore probably would have held her responsible for all of the Draco's injuries. But Hermione punched Draco under an ordinary tree, one that she couldn't possibly have known or even suspected would fall then. And that makes all the difference (at least to Dumbledore).
Conclusion:
The analysis of these scenarios suggests that courts use terms like "causation" and "the cause" in an attempt to identify who can be held responsible and should be blamed/punished for particular actions.
Courts (and headmasters) are faced with three related questions, only the first of which is discussed at any length in traditional philosophical literature on causation. (1) What set of events caused the outcome? (2) Which of these causal events can best be singled out as the primary cause, and is that cause associated with a human agent who can be said to have caused the outcome? (3) Can that person be justly held culpable for the outcome? Judges and philosophers want to be able to answer these questions not just for individual cases, but to establish criteria which can be used as guidelines for assessing individual cases.
I am not prepared to answer the first two questions, but as to the third, I suggest that based on the scenarios considered here, it makes sense to hold someone accountable for the outcome of their actions only if they can reasonably be expected to have been able to have both foreseen and prevented the outcome.
20090202
On causal analysis as a means to avoidance, and the value of postmortems in computer systems administration:
The claim that there are many different things meant by the term "cause" (or "aitia") is nothing new. But (to riff on a phrase of Dennett's) what is "the kind of causation worth wanting"?
The kind of causation worth wanting varies with context. Similarly, the reason we perform causal analysis -- seek out causes -- varies with context as well.
Reasons for seeking out causes may include (note this list is not exhaustive):
In my non-philosophy life as a sysadmin, it is most often this last reason which motivates causal analysis. It is a truth universally acknowledged that systems must eventually fail. And when they do (after the immediate situation has been addressed), it's valuable to write a postmortem.
Why is it valuable? What is the point of a postmortem, and what material belongs in a postmortem? Put simply, the point of a postmortem is to learn from our mistakes so we don't make them again. A postmortem clearly spell out how an (undesirable) event happened. But a postmortem is not just about the past; rather it is about the future and future possibilities.
When a postmortem describes "causes" (using the term loosely) that contributed to the outcome, it suggests:
A good postmortem reads like a checklist of things to improve. It outlines ways to modify the environment to prevent whole classes of possible future failures. And herein lies the value of causal analysis to the sysadmin: by understanding the causal relations between events, we can take appropriate actions to avoid similar events in the future.