We saw last time that there are serious problems with Hempel’s account of explanations. Next, we will look at some proposed alternatives. Some alternatives build on Hempel’s ideas. Others depart more significantly.

 

I. Causal Accounts

 

A. Preliminary Points

Some of the objections to Hempel’s view suggested that his view erred because it omitted all specific reference to causal connections. The examples were arguments that satisfied his proposed conditions, but were not good explanations. They seemed to go wrong because they did not properly reflect the causal connections.

 

It is important to understand that Hempel did not omit reference to causation simply because he didn’t think about it. There’s a deeper reason, having to do with his views about what’s meaningful and what can be part of science. A good way to get at this is understand some ideas that were emphasized by David Hume. Here’s a key idea: Hume observed that in making claims about causation, all we ever really observe are patterns of events - constant conjunctions. Once you get clear about that, two sets of problems about causation emerge. The metaphysical problems concern the nature of causation. E.g., when we say that C causes E, what more are we saying, beyond something about E’s regularly following C? It seems to include the idea of the cause “making” the effect happen. Some find this woefully obscure. If we can’t observe anything more than “this after that”, people may worry that there is no real sense to the idea. Some think there is nothing more than regularities. The epistemological questions concern our ability to know or reasonably believe anything about causal relations. All we can really observe are the regularities up till now. Hume is interpreted by some as being skeptical about this.

 

B. Statement of the View

 

Very simple:

 

  C.       X is a correct explanation of event E = X is a true statement identifying the cause of E.

 

(R. uses the phrase “efficient cause” when he describes this idea on p. 49. Aristotle said that there were four kinds of causes. Efficient causes are the kind we think about most. But he also talked about “final causes”. These are goals, or ends, or purposes. Your HW assignment has to do with that.) Perhaps we could expand on this by saying something about causal processes and patterns. So we might say that being struck by the cue stick in a particular way caused the cue to move in a certain direction and at a particular speed, and this, given the set up, caused it to strike the 8 ball, which caused the 8 ball to move in a certain direction, which led to the 8 ball going into the pocket.

 

This condition seems to be satisfied in the cases of good explanations that we’ve mentioned so far. And it is not satisfied in the examples (birth control, the flagpole, barometer) that caused trouble for Hempel.

 

For those who think that the idea of causation is clear enough to use, this may seem pretty good. But to those like Hempel who are troubled by the metaphysical question mentioned above, this will be unacceptably obscure. For them, something clearer about the concept of causation will be needed.

 

C. Questions and Puzzles

 

1. Extraneous Junk

 

Suppose you want to explain why a billiard ball moved the way it did. And suppose you set up the initial conditions just right, stating the masses and velocities of the various balls before the collision, and assume the collision does cause the resulting motion (and not, say, a gust of wind at just about the same time). But suppose also that in your list of initial conditions you include some extraneous truths, e.g., that the player “hexed” the cue ball just as the shot was made. This seems to be true:

 

Being struck by hexed cue ball in such and such a way caused the 8 ball to go in the direction of the side pocket.

 

Yet the explanation seems not so good if it has this extra stuff in it. Maybe some “no extra junk” condition is needed. But it turns out to be hard to figure out exactly how to spell that out.

 

The remaining questions come from Rosenberg.

 

2. Distinguishing causes and coincidences

 

I don’t see why this is a problem. This may show that sometimes it is hard to tell what the real cause of something is, and thus what the correct explanation is, but it does not present any reason to think that the causal account is wrong.

 

3. Causes as necessary conditions

 

Suppose that a match is struck (S) and then it lights (L). We explain L by saying S caused L. This seems to be a good explanation. Rosenberg notes a few things about causation that are important. These are points that are sometimes sources of confusion. Here is one: S is not sufficient for L. If the match were wet, you could have S w/o L. People sometimes think otherwise. Here’s an example: suppose there’s a big snow storm and the roof of a house collapses. When asked why the roof collapsed, one might cite the weight of the snow. But someone might respond: but there was just as much weight on the roof of the house next door, and it didn’t collapse. Notice that if this response shows that the weight explanation is wrong, it relies on the assumption that the weight is sufficient for the collapse. Another example: A gets a cold. Explanation: A was exposed to B, who had a cold. Response: but C was also exposed to B, and C didn’t get a cold. Again, this response relies on the assumption that the cause must be sufficient for the effect. But that’s a mistake.

 

By itself, this point is no objection at all to the causal account of explanation. But one might think that there is an objection nearby. R. notes that if causes are necessary but not sufficient conditions, then causes are simply elements of the overall set of conditions that are sufficient. Sometimes some of these elements seem to us to be explanatorily relevant, sometimes other ones seem relevant. This is especially clear when we are thinking about contrasts - why did B, but not C, get a cold? Because C was immune. But, why did B but not D, get a cold. Because B was exposed to A (and D was not). So the idea is that whether “B was exposed to A, who had a cold” is a good explanation of why B got a cold depends on what else we are thinking about. Things weren’t supposed to work out this way. But maybe the causal theorist can respond: the explanation is a good one, but not the one I was interested in . (More about this when we discuss pragmatic theories of explanation.)

 

4. Ceteris Paribus laws, Explanation and Expectation

 

If a cause is just a necessary condition, then citing a cause does not provide good grounds to expect its effect. (Bottom of p. 50) This is true. Smoking illustrates this. R. seems to think that this is a problem. It’s not entirely clear what the problem is. There’s this: we would like explanations that do enable us to predict outcomes with confidence. But this requires filling out all the other conditions to yield some that are jointly sufficient. E.g., S + presence of oxygen + not wet + not too windy ... This will get mighty complicated. In fact, we probably can’t actually do it. People often invoke ceteris paribus clauses at this point - everything else equal. But then R. says that the “S caused L” explanation is only an explanation sketch. I don’t see why. That’s not what my proposed version of the causal account implied. He must be thinking of a version of the theory in which an explanation must identify the “complete cause” of the effect. There is a puzzle about assessing ceteris paribus laws. If I say, ceteris paribus X causes Y (of if X then Y), then it’s hard to know when I’ve been refuted. Suppose X happens but Y doesn’t. I can always say that the ceteris paribus condition wasn’t satisfied. (This is his point on the top of p. 52.)

 

So - there are puzzles about ceteris paribus laws. But this is a problem for the causal account of explanations only if defenders of the account say that an explanation must identify a complete cause or a complete cause ceteris paribus.

 

5. Probability

 

This is discussed on pp. 52-5. Often explanations are statistical. See smoking/lung cancer example on bottom of p. 52. Two points get mentioned on the top of p. 53 - “some smokers never contract cancer, while some lung cancer victims never smoked.” The second is quickly dismissed. Other things can have the same effect. It’s worth noting that there is a common error in causal reasoning that turns on just this mistake. People will think that they have reason to doubt that A caused B because they know of cases of B w/o A. (“It wasn’t my staying up all night and working or partying so hard that caused me to get sick. My friend didn’t do any of that, and he got sick too.) But, as R. says, this is just a mistake - there can be different causes of the same effect.

 

The other point is much more confusing. He writes, “The first fact, that some smokers don’t contract lung cancer, is harder to reconcile with the claim that smoking causes cancer.” To understand the issue here, we need a distinction between singular and general causal statements. Give examples. Notice that our causal theory was stated in terms of singular causal statements. The puzzle that he worries about in this section has to do with understanding general causal statements. What does the general statement “Smoking causes lung cancer” mean? Not considered here is an analysis in terms of individual causal statements. Instead, something statistical (hence, more easily measured) is sought. And, just to see the puzzles involved, note that nothing about the percentage of smokers who have lung cancer adequately analyzes the statement - it can be a high or low percentage. And even comparisons - higher percentage of smokers than non-smokers get lung cancer - will not suffice. Compare - a higher percentage of people who loiter just outside buildings get lung cancer. But the corresponding general causal claim is not true. To deal with this, you get the idea stated at the end of the first full ¶ on p. 53. Notice that this makes the general causal statement dependent on what we know about. Once again, that seems not to be the desired idea. And it’s hard to see how to eliminate this.

 

So - there are puzzles about how general causal statements work. But this is no problem for the causal theory we stated, which did not make use of such statements. Perhaps R. thinks that some explanations will have to make use of general causal statements. If so, then these are serious issues. And maybe some explanations do have to make use of them - perhaps good scientific explanations explain not just individual events (Why did Jones get lung cancer?) but also general points (Why do miners get lung cancer?) The latter will not be about individuals and will not invoke singular causal statements. It will appeal to general causal statements. So, we’d better understand them. And it seems that there are puzzles concerning them.

 

Conclusion about the causal account: a) the general idea that there is an essential connection between causation and explanation seems sensible; b) we have found no clear-cut counterexamples; c) there is a hard puzzle about how to formulate the view to rule out explanations with extraneous stuff; d) there are puzzles associated with the interpretation of general causal statements, and this is a problem if such statements enter into explanations; e) there are the general worries some philosophers have about the whole idea of causation; f) and there is the issue, not discussed in class but discussed in the text on pp. 53-5, about how this applies to apparently indeterministic physical processes.

 

II. Pragmatic Accounts

 

A. The General Idea

 

This view about explanations is discussed on pp. 41-44. The central idea is relatively easy to grasp.

 

We’ve been thinking so far that there is one correct explanation of any event, or perhaps a lot of explanations of varying level of complexity and detail - e.g., you can trace the causal chain further and further back, or supply more and more of the details. A key idea is that a thing’s status as an explanation is in no way a function of what people happen to think, know, or care about. It’s an “objective” matter in the world. Defenders of Pragmatic Accounts think this attitude misses something important about the idea of an explanation.

 

Consider the example on the top of p. 42. On the D-N or Causal accounts, an explanation answering any of (a)-(c) would equally count as an explanation for all of them. On this view, what counts as a good explanation of the same event - Ms. R’s killing Mr. R - depends on what we know or what we are interested in. More details might make the idea here clearer: with respect to (a), you might be wondering why she did it herself rather hire a hitman; with respect to (b), you might wonder why she killed him rather than merely beating him; with respect to (c), you might wonder why she killed him rather than, say, the woman with whom he had an affair.

 

You might think that this has particularly to do with intentional explanations of human behavior. But that’s not right. Suppose I throw a ball and it hits and breaks a window. The background might be: a) the same ball hit another window in much the same way, and this other window did not break, or b) a different object hit the same window and the window did not break.

 

The idea to extract from this: why questions are contrastive, i.e., always (?) short for “why this rather than that”. The things referred to by “that” are the alternatives - the other things that might have happened instead. But which other things go into that class depend upon the particular interests and background knowledge of the questioner. This makes clearer the differences among (a)-(c) in the Ms. R example. They have different contrast classes.

 

There is another element to this - the relevance relation. One way to think about this is in terms of levels of explanation. In the Ms. R example, you might want an answer that explains what she was aiming at, what she hoped to accomplish. But you might want an answer that goes further back in her history and cites the childhood formative experiences that made her turn out this way. Perhaps you could also give responses in terms of underlying neural events. Another example: if you ask why a plane crashed, you might be interested in an answer that brings out the relevant decisions people made that contributed to the accident, or the factors having to do with the weather that played a role.

 

A good explanation, on this view, will be a satisfactory answer to a why question. But what will count will depend upon the particular contrast class and relevance relation. If this is right, then there is no formula along the lines of Hempel’s account or the causal account that specifies what a good explanation is. What we will accept as an explanation will always depend upon the context. What’s a good explanation of an event in one context is not a good explanation of that same event in another context. The word “pragmatic” is used here to bring out the idea that there is this kind of variation that depends upon particular details about the situation. It’s not all a matter of mere logic and the meanings of terms.

 

B. Some Details

This can be made a little more precise using the <topic, contrast class, relevance relation> pattern that R. describes on 42-3. Apply this to the example about Ms. R. Any question is associated with these elements.

 

Q: Why did Ms. R kill Mr. R?

 

Topic: Ms. R killed Mr. R.

Contrast class: (i) Ms. R. killed Mr. R; (ii) Ms. R. kicked Mr. R; (iii) Ms. R. divorced Mr. R

More on the relevance relation in a minute.

 

In this setup, the background assumption is that she did kill him and the other elements in the contrast class are false. A response to the question takes the form:

 

She killed him in contrast to kicking him or divorcing him because ____Answer___

(i.e, (i) rather than (ii) or (iii) because Ans.) We often shorten this answer to Q to: because Ans.

 

In our example, the answer might be:

 

A: She was very angry with him because he had an affair and she no longer wanted to live with him but she wanted to inherit his money.

 

So we reply to Q with “Because A.”

 

This seems pretty good. Notice that if a different contrast class were intended, a different Ans. would be right.

 

A sensible assumption here, of course, is that the Answer is true. Assume that it is in our case.

 

But there is a question that is easy to miss here. What’s the connection between the answer and the thing being explained? In the example, what’s the connection between A and (i)? Why is it true that A picks out (i) rather than (ii) or (iii)? This is where the relevance relation comes in it. There is supposed to be some relation, R, such that A has relation to (i), but not to (ii) or (iii). But what is that relation?

 

Complicated issues arise at this point. Here’s one way to think about how it works. The questioner’s background beliefs are such that, given them, the Answer makes (i) more probable than the alternatives. In the example, given our background beliefs about people, (A) favors (i) over (ii) and (iii).

 

This does a nice job of bringing out why (A) seems to be a satisfying answer to us. We might respond, “Oh, now I get it.”

 

C. An Objection

 

What follows is based on the final 2 paragraphs in R’s discussion of this topic. There is a difference between an answer to a why question that is satisfying (relieves wonder) and one that is correct (true). The account seems to capture the former rather than the latter, but it’s the latter that we desire. R. gives examples, top of p. 44: i) an explanation that is correct may not be satisfying when the questioner is a child; ii ) superceded physical explanations are good but not true. [Why?]

R’s final paragraph raises a deeper point. It’s worth getting clearer on what he’s getting at. Read key sentences in ¶. Suppose that at the meeting of the American Astrology Association the same question is posed, but proposed answer is:

 

A1: She is a Gemini

 

The group might be satisfied with this answer. We might say that A1 has “astrological relevance” to the contrast class. The defender of answer A says that A has “psychological relevance” to the contrast class. If a defender of this sort of view has a basis for ruling out A1 and ruling in A, it seems like it will come in the form of an account of what a genuine scientific relevance relation is. In effect, this amounts to what the “right” background beliefs are. But we don’t have any account of what that difference is. We are back to where we started. R. says that this doesn’t help distinguish scientific from non-scientific explanations. We’ve just seen what he means. This is a significant problem.

 

III. Unificationist Accounts

 

A. The General Idea

This idea is developed by Kitcher in the essay on reserve and discussed by R on pp. 55-6.

The idea in brief: we explain things by fitting them under general patterns. Hempel style arguments, following either the D/N or the I/S model do this. But now think about our overall set of explanations of things. The more things you can explain, using the fewest different argument patterns, the better. To reduce the total kinds of explanatory argument is to unify the explanations. Explanations that are part of a more unified system count as correct (or at least scientifically best). The idea is not that an explanation is a good one just in case people think it is. Rather, the idea is that it has to fit into the best way of systematizing the overall set of things they believe. Suppose that there are a bunch of observations and two competing theories about them. To take an overly simplistic example: suppose we wonder why someone has a heart attack. Genetic and environmental (diet/exercise) explanations are proposed. It might be that one explanation is more popular - among ordinary folks or scientists. But it might be that the other one contributes to a more unified system. The unificationist theory endorses that explanation, not the popular one. But it is best relative to the beliefs and observations. It could ultimately be wrong. [Question: whose beliefs? He doesn’t say. I think he has some idea about “the scientific community.” But I’m not sure.]

 

B. A Few Details

Sections 3-5 of Kitcher’s article contain examples of what he means by an argument pattern, using Newton and Darwin. The main idea: a bunch of seemingly diverse things are brought together using certain kinds of explanatory arguments. The individual arguments can be just like the one’s from Hempel.

 

A proposed explanation, E, is a good explanation (relative to background beliefs K) just in case E belongs to the set of arguments that best unifies K.

The set of arguments that best unifies K is the set of arguments that offers proposed explanations of the most things in K using the smallest set of argument patterns.

 

 

The examples earlier in the article, about Newton and Darwin, illustrate the idea. History of science would help here, but the short story is that Newton showed how from a single set of laws - thus using the same argument patterns - you could explain both the motion of bodies on earth (billiard balls) and the motion of the planets.

 

C. Dealing with the problem cases

See p. 292f. Three examples that were problems for Hempel. They should be familiar. Kitcher claims that his theory gets them right. The general strategy is the same in each case: the bad explanations get ruled out because any system that included them would be less unified than the one that excludes them.

 

Go over the hexed salt example. Defenders of the explanations involving hexes (presumably) agree that sometimes salt that has not been hexed dissolves. They will (presumably) appeal to ordinary chemical explanations in those cases. So they will have two argument patterns - ordinary, hexed - where the good theory has only one. And, Kitcher says, they are not able to explain anything more.

 

Consider the flagpole example. We can derive the length of the shadow from the facts, but we can also derive the height of the pole from the initial conditions and the length of the shadow. The latter is deemed not a good explanation. This is the asymmetry problem. Kitcher says that his theory rules out the second explanation. But he uses a different example - the pendulum. Note similarity of this case to the flagpole case. We’ll look at what he says about the pendulum, then apply it to the flagpole.

One explanation we have for the length of the pendulum is what he calls an “origin and development” (OD) explanation. This will explain its length in terms of how it came about. We have another explanation in terms of its period. But we’ll have to include OD explanations in our overall system in order to explain the lengths of things that aren’t pendulums. So, a comprehensive system of explanations that includes the explanation in terms of its period will have an extra argument pattern but won’t explain more stuff. A comprehensive system that included the explanation in terms of periods but dropped the (OD) patterns wouldn’t explain as much. The one that has the OD explanation but not the period explanation comes out best. So the period explanation is not part of the best system. That’s why it’s not a good explanation.

The same is true of the flagpole example. You can explain flagpole’s height using an OD explanation or the shadow length explanation. But not everything has a shadow. Compare three overall systems: 1) OD explanations + shadow explanations, 2) Shadow but no OD explanations, 3) OD but no shadow explanations. (1) explains everything that the others explain, but it has more patterns. (2) does not explain the length of shadowless things. (3) explains everything, but uses fewer patterns. So, (3) is best. The shadow explanation is not part of the best system. So it’s not a good explanation.

 

Similar things apply to the other puzzle cases.

 

D. Discussion

 

Briefly: This is an interesting idea. R expresses a worry that this relies on the assumption that the universe is “simple”. (p. 56)

 

A different question: notice that in the flag pole example the reason (3) came out best depends on the fact that not everything has a shadow. It’s not clear that all the non-explanatory regularities can be set aside for this sort of reason. If there are any systems in which you can always work back from effects to causes as well as forward from causes to effects, then you can equally well unify the system by including either kind of explanation. But that seems wrong. And that’s why adding a causal element seems like a good idea.

 

IV. Rosenberg on Teleology and Necessity

 

We’ve consider several views about the nature of explanation. R. says that there are questions about whether scientific explanations “really answer our explanatory questions” (p. 56) Two reasons for this doubt are raised. They apply equally well to all the different views we’ve consider (maybe not the pragmatic view).

 

A. Teleology

 

This is what the HW assignment was about. Causal explanations don’t make things “intelligible”, they don’t reveal their point. One application - human behavior. We discussed this earlier. We saw that these teleological explanations can be recast as causal explanations. Review an example.

 

1. He went to the bank in order to withdraw cash.

2. He went to the bank because he wanted to have cash and thought that the best way to get it was by going to the bank.

 

Notice that this does not eliminate purpose in human behavior. Rather, it says what it is, and places it within a causal framework. (1) is still true. It’s just another way of saying what (2) says.

 

A second place in which people have thought there is purpose, and for which teleological explanations apply, is in biology. People thought various systems had a purpose. People had eyes in order to see.

 

3. People have eyes in order to see

4. Plants have chlorophyll in order to produce starch.

 

One reaction: the existence of the Darwinian explanations shows that these are false. In general, Darwin eliminates purpose from nature, at least to the extent that all things like (2) and (3) come out false. Perhaps the idea is that the only way they could be true is if there were a designer whose is attempting to fulfill goals by giving us eyes and plants chlorophyll. So (3) and (4) are just like (1), but they come out false because it is false that

 

5. People were given eyes because their designer wanted them to be able to see and thought that giving them eyes was the best way to accomplish this.

 

Similarly for (4).

 

But the second reaction is to say that (3) is really true. However, this has nothing at all to do with intentions, God, or anything designer (other than the metaphorical “Mother Nature”). (Notice the final sentence on the ¶ on the top of p. 61.) The idea is that just as we could say

 

6. The function of eyes is to see

7. The function of chlorophyll is to produce starch

 

And we can explain the idea of function here in terms of evolutionary value. So the evolutionary advantage of eyes or starch spells out more clearly what (6) and (7) come to. And so on this view, (3) and (4) are true. We have “naturalized purpose.” Instead of explaining it away, we have explained what it is.

 

B. Necessity

 

We’ll skip this section.

 

V. Some Conclusions

A) We start with the question, “What counts as a scientific explanation?” We want some kind of analysis or explication of the idea. On the model of our analysis of lying. We don’t want a list. We want the factors that are nec. and suf. for something to be an explanation.

 

B) As a matter of historical fact, Hempel’s DN model is the starting point for discussion of this. The rest is reaction to that. It does seem to capture something - some good explanations are like that. Review the main idea.

 

C) Lots of explanations that we regard as acceptable fall short of the Hempel model. We sometimes appeal only to statistical factors. Hempel tried to deal with this by adding IS explanations. There were some puzzles in understanding just how this was supposed to work.

 

D) Other explanations that don’t meet Hempel’s conditions are ones in which we mention just the individual causal factors but no laws.

 

E) It’s a matter of controversy whether all of the explanations that come up short are simply incomplete - “explanation sketches” - or whether Hempel’s conditions are too demanding. A notable point in favor of the latter alternative is the fact that it would seem that we can have good explanations in cases in which it is an open question whether any Hempel-style explanation could even in principle be given, e.g., good historical explanations.

 

F) A question we did not pursue at length, and to which we will return later on, concerns his appeal to laws of nature. We have examples that illustrate the idea between laws and accidental generalizations. But there are questions about exactly what the difference amounts to and how we can have a good basis for thinking that we’ve found a law in those cases where we think we found one. We talked about supporting counterfactuals in this connection, but that seems at best to identify a feature of laws without helping us really understand what makes something a law.

 

G) Some arguments that fit Hempel’s models clearly are not explanations. (The flagpole example, the barometer example, the pregnancy example.) These examples are similar, but they are slightly differently. They show fairly conclusively that his conditions are not sufficient. This is a major problem for his view. A way to state the problem briefly: there are non-accidental regularities that can be used in arguments fitting the DN (or IS) patterns, but which are not explanatory.

 

H) One leading response to this is to defend a causal theory of explanation. I like this. If you are moved by the point in (D) to think that you don’t need laws in explanations, then you’ll like the simple causal view: a good explanation merely cites the cause of something. If you are not moved that point, then you’ll like adding a causal condition to Hempel’s conditions.

 

I) But there are puzzles about exactly what causation is. And if you want to stick to the idea that science deals only with observable things, then very serious puzzles arise since (it seems) we can’t really observe causation! Things get even murkier if we allow that explanations can involve general causal statements. So some people are not willing to accept the causal account.

 

J) A different response to the puzzles for Hempel’s account is given by the unification account. We’ve only discussed this briefly, but you should get the idea. It’s an interesting idea. But there is a worry that even though it dealt well with specific examples, this success will not carry over to all similar examples. We did not develop this issue in any detail.

 

K) And there are those who think the pragmatic theory is really correct. It holds that an explanation is whatever answers a “why question”, where what counts depends upon background information. This does help us see that one way we use the word “explanation” has a contextual or pragmatic element. An example we did not discuss brings this out very well. Q: Why is the porch light on? A1) I flipped the switch. A2) Because we are expecting company. In different settings, different answers seem suitable. But I think that one way to think about this is to note a difference between the sort of explanation that one is looking for in a particular situation and what is, in its own right, a good scientific explanation. And if you are more careful about exactly what is getting explained in that example, you’ll notice that there are really two different things: the light’s being on, my turning it on. No reason to give up the search for conditions for scientific explanations.

 

L) Given what we said about teleological explanations, (A1) and (A2) in the previous paragraph really just highlight different aspects of the same causal story.