What Good is an Explanation? (Part 1)

Peter Lipton's death in 2007 was a great loss to the philosophy of science. He was probably the foremost explicator and defender of the explanatory method which I find most congenial: inference to best explanation.

In this brief series, I am going to take a look at one of his articles. The following one to be precise:

"What Good is an Explanation?" in Cornwell, J. (ed) Understanding Explanation (Oxford University Press, 2004)

It will no doubt come as a shock to learn that in this article Lipton attempts to answer a meta-question about the good of explanations. In other words, instead of asking "what makes for a good explanation?" he asks "why should we even bother explaining something?".

The answer to this meta-question is hardly earth-shattering. We explain things for instrumental reasons, i.e. to help us achieve some other goal, and also to achieve the intrinsic good of "understanding".

The majority of the article is taken up with trying to determine what this intrinsic good of understanding really consists in. Only at the end does Lipton return to the instrumental good of explanation.

In this post we will do two things. First, we will identify three key features of an explanation, features that must be accounted for in any formal definition of "understanding". Second, we will consider five possible definitions of understanding.


1. Three Key Features of an Explanation
Lipton begins his article by highlighting three uncontroversial and yet key features that are shared by most explanations. These features serve as diagnostic tests for any sound conception of understanding.

The first of these features is the distinction between knowing that something is true and understanding why it is true. Suppose you and I have just received our exam results. We both did badly. Staring askance at our grades, we are both inclined to ask the "why" question: why did I do so badly? In this scenario, we both already know that we did badly. We now want something more: we want to understand.

The second of these features is the benign nature of the why-regress. We all know the irritating manner in which every possible answer to a why-question can be followed up by another why question. As follows:

Why did I fail my exam?
Because you didn't study properly.
Why didn't I study properly?
Because you're easily distracted.
Why am I easily distracted?
etc. etc.

The important point here is that explanations are not worthless just because there is a potentially infinite regress of why-questions. Genuine understanding can be gained at each link in the question-and-answer chain.

The third feature of explanations is that they are sometimes self-evidencing. What this means is that the data they are intended to explain can, at least partially, justify us in accepting them. So for example, my lack of study can explain why I failed the exam; and my failing the exam can provide evidence for my lack of study. There is a kind of circularity here (H explains E; E justifies H), but it is not vicious.

With these three features in place, we can proceed to consider five possible conceptions of understanding. They are: (i) the reason conception; (ii) the familiarity conception; (iii) the unification conception; (iv) the necessity conception; and (v) the causal conception.

The first two of these make understanding an epistemic matter; the last two make it an ontological matter; the middle one can go either way. This will make more sense once we have gone through them, which is what we are now going to do.

2. The Reason-Conception of Understanding
According to this conception, what we really want from an explanation of a particular fact is a reason to believe in that fact. So questions of the form "why P?" need to be restated in the form "why should we believe that P?". Bayesian evidentialists are fond of this idea.

There are two main attractions to this conception. First, it doesn't distinguish between reason-seeking why-questions and explanation-seeking why-questions. So it reduces the size of the set of epistemological questions we need to ask. Second, it does not rely on somewhat dubious metaphysical concepts, such as causation and necessity.

There is, however, one significant problem with it: it fails to account for any of the three features of explanation outlined above.

• It does not account for the distinction between knowing and understanding because, for example, your doctor's expert opinion about the state of your health may give you a reason to believe that you are sick, but it would not help you to understand why you are sick.

• It does not account for the benign nature of the why-regress because, according to one approach, in order to have a reason to believe that H explains E, we would also need a reason to believe H. This approach would have to fall foul of the why-regress.

• It does not account for self-evidencing explanations because if E is reason to believe in H, H cannot be a reason to believe in E. This would be a vicious circle.

3. The Familiarity-Conception of Understanding

According to this conception, an explanation works when it reduces the unfamiliar to the familiar. So the goal of all explanations should be to build analogies (or other logical bridges) between the known and the unknown. For example, Charles Darwin explained the "design" exhibited in the natural world by building an analogy between the processes of artificial and natural selection.

This conception fits well with the typical context in which explanations are sought, namely: the context of epistemic surprise. In other words, we tend to seek explanations when something seems unfamiliar or surprising and this tendency is captured by this conception. Also, this conception does account for the gap between knowing and understanding: something is understood when it is rendered familiar; something is known when we have reasons to believe in it.

Despite these good points, the familiarity-conception does poorly when dealing with the other two features of explanations. 

• It cannot deal with the why-regress because in order to be familiar, something must be understood, and so only what it already understood can count as an explanation. Thus, it is impossible to derive satisfaction from an explanation that is itself unexplained.

• It cannot deal with the self-evidencing nature of some explanations for the following reason. If an explanations works only to the extent that it reduces unfamiliar facts (E) to a familiar hypothesis (H), it unclear how those same unfamiliar facts could give us reason to believe in the truth of H.

4. The Unification-Conception of Understanding

According to this conception, something is understood once we can see its place within a broader, more unified, conception of reality.

This unification-conception accounts for two of the key features of explanation outlined above. First, it accounts for the gap between knowledge and understanding: we can know that a fact is true without necessarily knowing where it fits in the broader conception of reality. 

Second, it accounts for the self-evidencing nature of explanation. Lipton uses an analogy to make this point: a single piece of a broader pattern can provide evidence for that broader pattern, while at the same time the broader pattern can help us to understand the role of function of the single piece.

It may also account for the why-regress, although Lipton thinks this is less clear. It would do so by showing how E is explained by the wider pattern of H, while leaving the place of H within a still wider pattern unclear.

5. The Necessity-Conception of Understanding

According to this conception, something is understood once it has been shown that it had to occur; that there was no other way things could have turned out. This brings us face-to-face with the principle of sufficient reason, a metaphysical concept beloved by theists and atheists, which is used to argue for the existence of a necessary being.

This necessity-conception can account for the gap between knowledge and understanding. After all, something can be known to be true without being known to be necessarily true. There would also appear to be no problem with self-evidencing necessary explanations.

Where the conception appears to fall down is with the why-regress. It would seem that in order for an explanation (H) to confer necessity on a set of facts (E), the explanation would itself need to be shown to be necessary. This would rule out the possibility of being satisfied with an explanation that was not itself shown to be necessary.

6. The Causal-Conception of Understanding

According to this conception, something is understood once we have information concerning its causes. This is the conception of understanding that Lipton himself prefers.

The causal-conception can easily account for the three key features of explanation that we have been discussing: 

• Something can be known to occur, without knowing what caused it to occur (knowledge v. understanding); 
• We can know that C was the cause of E without also needing to know what caused C (why regress); 
• There is no reason why C cannot be the cause of E, while at the same time E can provide evidence for C (self-evidencing).

Despite these powerful attractions, there are, of course, challenges facing the causal conception. Lipton identifies three important ones.

• There is no completely adequate concept of causation. (See section 2 of this post for some of the different approaches)
• Some explanations, such as mathematical explanations, seem to be clearly non-causal.
• Not all causes are explanatory. Even though there is a long chain of causes linking my present actions to the origin of the universe, it would be odd to explain my behaviour by referring to the Big Bang.

Acknowledging these challenges, Lipton's goal for the remainder the article is to offer a more detailed defence of the causal conception of understanding. I will outline that defence in part two.