The other day I “wrote about”:http://tar.weatherson.org/2007/02/22/hall-on-causation/ Ned Hall’s recent theory of causation. I should have mentioned the reference for the paper. It’s “Structrural Equations and Causation”, _Philosophical Studies_ 132: 109-136. What I was really interested in in that paper was what Ned says about ‘short-circuits’. This is his name for causal structures like the following.
We have a causal pathway leading from A that will, if unchecked, result in E. C launches two causal pathways. The first of these, T, is a Threat to E. If unchecked it will cause D, and that will prevent E. In the examples to follow, E is the survival of Victim sometime after the events described, and D is (or would be if it occured) a particular death of Victim’s. Victim is merely the victim of misfortune in these stories; he always however survives. That’s because C also causes P, which Protects Victim from the threat, and Prevents D from happening.
Question: Does C cause E?
Answer 1 (Lewis): Yes. Causation is the ancestral of counterfactual dependence. And E is counterfactually dependent on P, which in turn is counterfactually dependent on C.
Answer 2 (Structural Equations Theory): Yes. It is sufficient for C to cause E that there is a related model in which E is counterfactually dependent on C. The model where we determine exogenously that T occurs is a model where E is counterfactually dependent on C, and is related in the right way to the actual model. So C causes E.
Answer 3 (Hall): No. C never produces a real threat to E, because it cancels itself out. So C does not cause E.
My answer: We just haven’t been given enough information. Some cases with this structure are cases where C causes E, and some are not. So everyone is wrong. Figuring out what separates the causal cases from the non-causal cases, or figuring out which features of a case and its presentation make us judge a case as causal or not, is really a very hard puzzle. Below the fold I’ll work through some hypotheses and theories.
First a case that I think pretty clearly supports Ned’s take on matters.
_Arrested Assassin_
Enemy hires Assassin1 to kill Victim. (This is C.) Assassin1 starts planning the kill. (This is T.) Happily, Agent has a wire inside Enemy’s building, and hears of the plan. Assassin1’s accepting the contract to kill Victim is obviously a criminal offence, so this leads Agent to arrest Assassin1, thus preventing Assassin1 carrying out his mission. (This is P.) As always, D is Victim’s death (which doesn’t happen) and E is Victim’s Survival (which does).
Here is seems just bizarre to say that C caused E. Perhaps I could be talked into accepting this – in philosophy we often have to accept bizarre claims. But let me just say that my intuition that C does not cause E in this case is stronger than my intuition that Gettier cases are not cases of knowledge. So this makes it look like Hall is right. But we should look at some more cases before we are so sure.
_Bomb Detector_
Victim is now a little nervous. He is particularly worried about bombs on his car. So he hires Mechanic to install a bomb detector on his car, one that blows up the car when a bomb is detected, as long as it detects no one is actually in the car. Let C be the installation of the detector. Assassin2 sees Mechanic installing the detector. Assassin2 doesn’t want to kill Victim; he wants to kill Bill Clinton. Unfortunately, Assassin2 believes that only Bill Clinton would be having this kind of bomb detector installed, so this is the car he should attack. Fortunately, Assassin2 always assassinates with car bombs, bombs that he (mistakenly) believes can’t be detected by bomb detectors. So Assassin2 puts a bomb on Victim’s car. (This is T.) The bomb detector is programmed to regularly scan the car for bombs. As T is happening, one of the regularly scheduled scans starts. (This is P.) The scan detects the bomb, and blows up the car. Victim is safely inside his house, so D does not occur.
I imagine people will have differing reactions to this case, but I have both an intuition that C causes E, and a small argument that C causes E. The argument is that in the case Mechanic saves Victim’s life. At least if I were Victim, I would feel that Mechanic had saved my life. But you can’t save someone’s life without causing their continued survival. And there’s nothing that Mechanic does other than install the bomb detector to save Victim’s life, so it must be C that causes E.
This suggests that there is something in the fine details of the case that distinguishes between various ‘short-circuits’. Some short-circuits are causes and some are not. What might separate the two?
My first hypothesis turns on a suggestion that Christopher Hitchcock and Ned Hall have made. They have both argued recently that we should distinguish between ‘default’ and ‘deviant’ states of affairs. The distinction is meant to be somewhat intuitive, and I won’t try to say much to explicate it here.
Now all of the events (except D) in the above diagram are deviant happenings. But C doesn’t produce T and P all by itself. There are background conditions at play as well. Let’s put these in to the diagram and see what happens. So let T and P be ‘recalcitrant’ neurons that only fire if doubly stimulated. (That’s what we represent in the diagram by giving them a thick border.) And let BT and BP be ‘background’ events that combined with C are sufficient for T and P respectively.
Here is a distinction between our two cases. In _Arrested Assassin_, BT is basically Assassin1’s disposition to accept contracts and ability to carry them out. That’s a relatively stable state of affairs, and we can think of it as happening by default. In _Bomb Detector_, BT is that Assassin2 is watching Mechanic’s shop, and has formed the belief that only Bill Clinton would be having a bomb detector installed. That is not the way the world is by default. Here then is a hypothesis.
Hypothesis 1: C causes E in a short-circuit case iff BT is a deviant state of affairs.
This covers a lot of cases, but I think it makes it too easy for C to cause E in cases like the following.
_Meat Pie_
Victim goes to the shop and orders a Steak and Vegemite Pie. (This is C.) Around this moment, Assassin3 decides that he is going to kill the next person who orders a Steak and Vegemite Pie. Hearing Victim order this, he starts trailing Victim looking for a good time to kill him. (This is T.) Psychic has had a vision that someone who orders a Steak and Vegemite Pie will be in danger. On hearing Victim order the pie, he decides that Victim needs assistance. So he starts trailing Victim as well, watching out for threats. (This is P.) Because Psychic is also trailing Victim, Assassin3 has to pass up several opportunities that would otherwise have been perfect chances to kill Victim without any observers. Eventually Assassin3 tires of this and leaves without killing Victim, so Victim survives.
Here it still seems odd to say that C caused E. C put Victim’s life at risk, and only the lucky intervention of Psychic saved the day. So C did not cause E. What seems distinctive about this case is that both BT and BP are deviant states of affairs. Normally there aren’t assassins waiting to kill the next person who orders a certain kind of pie, or psychics waiting to protect such pie-buyers. This suggests a refined version of Hypothesis 1.
Hypothesis 2: C causes E in a short-circuit case iff BT is a deviant state of affairs, but BP is a default state of affairs.
This has some plausibility I think. If C is the kind of thing that, by default, protects against threats like T, and a threat like T arises, then arguably C is a cause of E, even if C plays some role, in conjunction with a very odd state of affairs, in bringing about the threat. I have some sympathy for this conclusion, but I’m worried that it leads to mistakes in cases like the following.
_Volvo Driver_
Friend advises Victim to buy a Volvo because they have so many safety features. (This is C.) Victim does go on to buy a Volvo on Friend’s advice. The next day the Aliens invade. As everyone knows, what the Aliens really hate are Volvo drivers, so they go about throwing large boulders at all cars, including Victim’s car while Victim is driving it. (This is T.) Happily, the airbags that Friend was so stressing do their job when the boulder hits, launching and protecting Victim from the attacks. (This is P.) So Victim survives.
I’m a little torn about the case, but I think here C does not cause E. That is despite the fact that BP consists entirely of pretty default states of affairs (like that airbags work) and BT consists of pretty deviant states of affairs (like that Volvo-hating Aliens invade). So I don’t think Hypothesis 2 gets to the heart of the matter.
My preferred solution is rather more radical, but stating it requires some set-up. We all know that there is a distinction between Phi-ing and causing something to be Phi-ed. If I say something that offends my guests and they storm out, I cause my front door to be opened (and slammed), but I don’t open (or slam) my front door. In some places a new meaning for ‘direct’ is stipulated so that one opens a door iff only directly (in this proprietary sense) causes the door to open. It has to be a special sense because, as Lewis pointed out, we can open a door using arbitrarily complicated devices. (Lewis’s way of putting this was to refer to Rube Goldberg machines, but that reference might be too old for youngsters like, er, me!) But I think we can make sense of this idea that some causal chains are more apt to be the truthmakers for causatives like _open_, and we’ll call these causal chains more _direct_. This lets us state my preferred view.
Hypothesis 3: C causes E in a short-circuit case iff the chain running from C to P is (much) more direct than the chain running from C to T.
Roughly, C causes E iff C _protects_ Victim from threats like T (as opposed to merely causing Victim to be protected) and doesn’t _threaten_ Victim (as opposed to causing Victim to be threatened).
I think this does get the cases right. In particular in _Volvo Driver_ it links up with a rather delicate intuition I have about a more realistic case. If my friend advises me to buy a car with airbags, and I do so, then the friend has caused me to be protected from various threats. But my friend hasn’t protected me from those threats. On the other hand, if the friend actually installs the airbags, then she has protected me from the threats. It is the inapplicability of causatives in advice cases like this that leads to the inapplicability of ’cause’ in _Volvo Driver_.
There is though one kind of case that I’m worried about. It is widely known that the application of causatives depends on the intention of the causal agent. I’ve been in various offices where, if the door was opened, opening the window would cause the door to slam shut. (Often quite violently, which was a little unpleasant.) Now in general in such a case my opening the window would cause the door to close, but wouldn’t be a way that I’d close the door. There is one exception. Say that I wanted the door closed, but I was nearer the window than the door, and opened the window with the intent of causing the door to close, and as a result the door became closed. In such a case I would have closed the door, not merely caused it to close. That is to say, intention matters to directness. But, I worry, it doesn’t matter to short-circuits. So consider one last case.
_Door Stopper_
Victim is in Friend’s office. Friend opens the window of his office. (This is C.) The wind that comes in causes Friend’s door to slam shut. (This is P.) Assassin4 is waiting outside Friend’s office, hoping to kill Victim. When Friend moves the window (open, as it turns out) Assassin4 sees Victim’s reflection and thinks he can kill Victim. So he fires. (This is T.) Unhappily, the bullet is headed straight for Victim. Happily, the door, which is a big thick steel door, closes shut just in time to prevent the bullet getting to Victim. So Victim lives.
I’m (tentatively) inclined to say that this is a case where C causes E. More importantly, I’m worried that whether Friend intended the door to close makes a lot of difference to whether the chain from C to P is direct, but makes no difference to whether C caused E. That is to say, whether we fill out _Door Stopper_ to say that Friend intended the door to close, or was surprised by its closing, seems to make no difference to whether C causes E. But Hypothesis 3 suggests that it should make some difference.
This is a problem for Hypothesis 3, but I think it can be, to some extent, parried. Even if Friend didn’t _close_ the door, there is still a good sense in which the C-P causal chain is much more direct than the C-T causal chain. Opening a window was not a way that Friend threatened Victim’s life, or even I’d say endangered Victim, though it did cause Victim to be in some danger. So Hypothesis 3 predicts that in both variants of _Door Stopper_, C causes E.
If Hypothesis 3 is right, a rather radical conclusion follows. To know whether _cause_ applies in a particular case, we have to not just know the causal relationships between various subsidiary events, but know whether causatives are applicable to those subsidiary events. That means that any analysis of causatives that uses the concept of causation will be _circular_, since in some cases whether _cause_ is applicable depends on whether causatives are applicable.
That’s to say that our concept of causation can’t be conceptually prior to our causative concepts like _open_ and _protect_. And that, if right, is I think quite a striking conclusion.