Inference to the Best Explanation

In recent decades, the inference to the best explanation has emerged as one of the most promising ways to formulate the logic of scientific arguments.⁴⁸⁵ The basic logic behind the inference to the best explanation (IBE) is generally thought to be “abductive” as described by C.S. Peirce.⁴⁸⁶ The idea of abductive reasoning is that if our empirical evidence would be a reasonably expectable occurrence given the truth of some hypothesis, then this gives us evidence in favour of this hypothesis.⁴⁸⁷ Peirce formulates this logic as follows:

7. The surprising fact C is observed.

8. But if A were true, C would follow as a matter of course.

9. Hence, there is reason to suspect that A is true.⁴⁸⁸

Many believe that this is the basic form of inference which underlies many forms of science, Darwinian evolutionary theory being one standard example.⁴⁸⁹ For example, if all animals have a common ancestor, then we would expect them to have biological similarities.

Biological similarities exist, so we have grounds for believing the hypothesis of common descent. This kind of abductive reasoning can fail, because the same evidence can fit several different hypotheses. In practice abductive explanations (hypotheses) are evaluated based on several different criteria such as enhanced likelihood, explanatory power and scope, causal adequacy, plausibility, evidential support, fit with already accepted theories, predictiveness, fruitfulness, precision, unifying power and the like. If one explanation emerges as superior in this comparison, we are justified in accepting it as the right explanation, until a better one emerges. Because the exact criteria and their application are controversial, deciding on the best explanation is a process dependent on background beliefs and subjective considerations.⁴⁹⁰

The history of science knows many examples where the same data could be explained by several models. In Galileo Galilei’s time, the Earth-centric cosmology of Tycho Brahe was able to explain the same data as the heliocentric cosmology of Copernicus and Galileo.

However, over time the situation has changed in favour of the heliocentric model of the solar system. This feature of inferences to the best explanation is known as explanatory

485 Lipton 2003.

486 Peirce 1955, 151. Quoted in Ratzsch 2010.

487 Ratzsch 2010.

488 Peirce 1955, 151. I am following the arguments from Ratzsch 2010 and Dawes 2009, 20-23.

489 See Banner 1990, 125-130. For more on the general logic of inferences to the best explanation, see Lipton 2003.

490 Ratzsch 2010.

underdetermination.⁴⁹¹ The ID theorists’ claim that the design argument does not reveal the identity of the designer can also be stated in terms of underdetermination: the same data can be explained based on a wide variety of different design hypotheses. For example, perhaps the designer could be a Platonistic demiurge, a space alien or Zeus instead of the God of the Bible.⁴⁹²

Within the ID movement, Meyer explicitly refers to his argument primarily as an inference to the best explanation (IBE), whereas Behe makes several different formulations of the argument. However, the probabilistic nature of Behe’s argument makes seeing it as an inference to the best explanation credible. As I will show in chapter 6.1, Behe argues that the probability of the design conclusion varies based on the properties of the natural systems under analysis. The more complex and purposeful the order appears, the more certain our conclusion of design becomes for him.⁴⁹³ He compares the likelihood of design and natural explanations and argues that our background religious and non-religious beliefs effect our assessment of the probability of the conclusion.⁴⁹⁴ He further argues that the presence of consilience, several independent lines of argument leading to the same conclusion, lends more probability to the design argument.⁴⁹⁵ These are features of inferences to the best explanation.⁴⁹⁶

Ratzsch formulates the design argument as an inference to the best explanation (IBE) as follows. Here the first premise is related to Pierce’s premise that “the surprising fact C is observed” and the second premise is the hypothesis which explains this datum. After comparison with alternative explanation, it is concluded that the design hypothesis is the best explanation and probably true.

10. Some things in nature (or nature itself, the cosmos) exhibit exquisite complexity, delicate adjustment of means to ends (and other relevant R characteristics).

11. The hypothesis that those characteristics are products of deliberate, intentional design (Design Hypothesis) would adequately explain them.

12. In fact, the hypothesis that those characteristics are products of deliberate, intentional design (Design Hypothesis) is the best available overall explanation of them.

491 Lindberg 2003.

492 For example, John Leslie (1989) and Michael Denton (1998) have advocated such non-traditional design hypotheses. See Ratzsch 2001, 81-82 for some ideas about the concept of underdetermination in relation to design arguments.

493 Behe 2006a, 256.

494 Behe 2001.

495 Behe 2007, chapter 10.

496 Paley’s design argument can also be stated as an inference to the best explanation. Schupbach (2005), responding to Oppy (2002) argues that Paley even intended his argument as an inference to the best explanation, while Oppy (2006) continues to argue that it is a deduction, but acknowledges the possibility of other interpretations. I am inclined to agree with Schupbach.

Therefore (probably):

13. Some things in nature (or nature itself, the cosmos) are products of deliberate, intentional design (i.e., the Design Hypothesis is likely true).

The conclusion (13) is not always included in formulations of the inference to the best explanation. It simply states the common belief if some explanation is the best, then we have grounds for believing that it is also probably true, not just the best currently available explanation. This may not always be true. If all considered explanations are not very convincing and the best explanation is only slightly better than the rest, then we may still not have grounds for believing that it is true. However, assuming that the criteria for selecting the best explanation are conducive to truth – that is, assuming the reliability of our rational intuitions concerning explanations and the philosophy of explanations – it seems that often we are at least rationally justified in accepting the best explanation as probably true.

However, it is not always clear what exactly makes one of the hypotheses so much better than the others that such confidence is warranted.⁴⁹⁷

Related to this, Peter Lipton makes the distinction between potential and actual explanations. A potential explanation is one that entails the data in some way; an actual explanation is simply a true potential explanation. Different criteria can then be used to try to identify the most likely actual explanation from the pool of potential explanations we have available.⁴⁹⁸ In the case of “design-like” objects in nature, potential explanations could include evolutionary explanations, design-based explanations, explaining the apparent rationality as an illusion created by our minds and so on. However, if design is at least a potential explanation – if it has explanatory force – then this comparison could in principle lead to the result that design is the best explanation.

Critiques of the Design Argument as an IBE

In the preceding formulation of the argument, premises 11 and 12 are the crucial controversial premises of the argument. As Gregory Dawes argues, there are two alternative lines of critique which can be made of any abductive theistic explanation. First (1), it can be argued (against premise 11) that theistic explanations actually don’t explain at all, meaning that they are not part of the pool of potential explanations. Divine action or (in the case of the ID movement) the actions of an unidentified intelligent designer do not show why we should expect to observe the data. This is an “in principle” objection to design arguments.

No matter what the universe looks like empirically, design cannot explain it. This means that

497 Dawes 2009, chapter 6.

498 Lipton 2004, 56-66. For Lipton, the inference to the best explanation is indeed concerned with finding reasons to believe in some hypothesis. However, it is also possible to interpret the inference to the best explanation as merely a heuristic method for comparing hypotheses that will then have to be confirmed using some other methodology. (Iranzo 2007, 340-341.)

there is no conceivable evidence that could speak in favour of design – a very strong conclusion. According to Dawes, the more promising critique of the argument is a “de facto”

critique based on the competition between different potential explanations of the same data.

This critique is targeted at premise 12 – the comparison of explanations. Though theistic explanations and design could in principle explain natural order, in practice naturalistic explanations like Darwinian evolutionary theory can be argued to work better and thus make design an unnecessary explanation.⁴⁹⁹ So, even if design is a potential explanation of the data, it can still be argued that it is not a very good explanation, and this is relevant for assessing the credibility of premise 12. Indeed, in the discussion on Intelligent Design, critiques of the explanatory power of design are common.

Common critiques concern the lack of independent support and frivolous nature of design hypotheses. Elliott Sober has illustrated the difficulties facing design arguments with an interesting parable. Suppose that we hear a strange sort of rumbling from the basement.

We could argue that the hypothesis “the noise is caused by a bunch of trolls bowling”

explains the noise quite well, as it predicts the observed empirical evidence. This is causally sufficient explanation involving designers, but it does not feel quite satisfactory.⁵⁰⁰

Two related difficulties help understand the unsatisfactory nature of the troll hypothesis. First (1), we do not have any independent reasons for believing in trolls, which means that the prior probability of the troll hypothesis is very low. We will rather hold out for a more reasonable explanation than accept something this strange. Even though the evidence increases the probability of the troll hypothesis, it is not sufficient to raise the probability into the realm of credibility, since the beginning probability is so low. Second (2), any data can be explained by a modified troll hypothesis, supposing that the trolls are postulated to have the motivation and adequate powers for producing the evidence we see.

For example, supposing that cookies are missing from the cookie jar, we could hypothesize that an invisible troll ate them. This would explain the data, but it would be totally frivolous.

The technical term for this frivolousness is that the troll hypothesis is ad hoc – a hypothesis artificially constructed just to explain this one piece of data, but which has no other grounds.⁵⁰¹

According to Sober, the design hypothesis for explaining the evolution of life also has a very low prior probability and suffers from the problem of frivolousness. What is needed for the design hypothesis to work is independent evidence of the existence of the designer and some plausible reason for supposing that the designer has the capabilities and the motivation

499 Dawes 2009, chapter 2.

500 Sober 1993, chapter 2.

501 This is also related to what Phillip Kitcher (1981, 528) has called the problem of spurious unification – if a pattern of explanation could fit any state of affairs, it cannot explain why this particular state of affairs exists rather than some other state of affairs which would fit the hypothesis equally well. (See also the commentary in Dawes 2009, 43-46.) Similarly, Robert Pennock (1999, 275) argues that explanation in terms of a designer’s purposes is non-explanatory, because a designer’s purposes can conceivably be invoked to explain anything. For example, supposing that we want to ask a question about why apples fall down from trees, we could answer that it is because God wills it.

for producing the order we see in nature. This is the only way to avoid frivolous design hypotheses which can be invoked to “explain” anything at all, but which we nevertheless have no grounds for accepting. Interestingly, Sober’s remarks are directed also against David Hume’s critique of the design argument. According to Sober, the design argument as an inference to the best explanation can avoid most of the Humean critiques of design arguments. For example, it does not depend on the exactness of the analogy between the universe and a machine. But Sober’s conclusion is very Humean: we can reliably infer the presence of human design, but our inferences about supernatural design are far less certain.⁵⁰²

So, for a design argument to work, the reasoning behind premises 11 and 12 must be made as clear as possible. It must be specified what is required before a design hypothesis can possess explanatory power, and to assess how much evidential support this really gives to the design hypothesis in relation to other hypotheses. Here the strategies of theistic natural theologians and the ID theorists differ, though both employ the same logic of the inference to the best explanation. In order to bring out the core ideas of ID and relate them to theistic natural theology, I will first introduce one way of defending natural theology and then go on to discuss ID’s approach. The similarities and differences of the two schools of thought will become clearer in the coming chapters.

Differences Between Design Arguments

The arguments of theistic natural theology are different from the troll hypothesis because they are not just artificially constructed to explain one facet of reality. Rather, there is a cumulative case of many theistic arguments which support each other. If – as many argue – the explanatory dimensions of theistic belief are not the reason for its origination, then this seems to show that theism is not an ad hoc hypothesis invented to explain some small amount of data.⁵⁰³ Kenneth Himma has similarly argued that we are justified in making design inferences only in contexts where there is already strong independent reason to think that there exist intelligent agents with the ability to bring about the occurrence of the relevant entity, feature, or property. Only in such contexts, Himma argues, is there sufficient information to justify assigning a probability to the design hypothesis that is higher than the probability that we are presumably justified in assigning to the chance hypothesis.⁵⁰⁴

502 Sober 2004. This type of conclusion is very common in the critique of ID. See Pennock 2001 and Hurd 2004 for other examples of the objection.

503 The explanatory origins of belief in God were discussed in chapter 2.5. Of course, it could be that the non-explanatory account is not correct. For example, if humans do have an intuitive tendency to explain nature teleologically (as discussed in chapter 4.1), then perhaps this is partly responsible for the origins of belief in God.

If design arguments are just more rigorous statements of the logic of this intuition, then belief in God and the design arguments come at least partly from the same source.

504 Himma 2005, 1.

As Dawes notes, it is also not the case that just any data could be explained by reference to the theistic God. This is why theistic believers would find the existence of gratuitous evil in the world very puzzling. God is thought to have certain attributes which make some ideas about what God would do more reasonable than others. For example, because God is good, he is expected to create a good world. The arguments of theistic natural theology thus often depend on clear ideas about God’s nature and motivations.⁵⁰⁵

It is nevertheless clear that such theistic explanations are quite different from any natural explanation based on lawful regularities. God is thought to be free and thus it is difficult to argue that any empirical result could be derived from the existence of God as a

“matter of course”. This difficulty doesn’t seem fatal, however: the same seems to be true of all intentional explanations, even those used of humans. Though determinists believe that it is in theory possible to predict the behaviour of humans completely, in practice it is not. We can never be sure that another human being has the exact intention which will result in exactly this sort of behaviour, though we may have very good reasons to believe so. Theistic philosophers of religion argue that the same is true of personal explanations referencing the intentions of God. Though God is free in his actions, he does not act without reason. Thus some outcome can be reasonably expected of the theistic God. For the theistic hypothesis to have explanatory power, it seems that it is enough for it to make the evidence more likely than the chance hypothesis. Its explanatory power can then be measured by how great the difference between the theistic hypothesis and the chance hypothesis is.⁵⁰⁶

The ID movement’s inference to the best explanation is different, though the following chapters will also show that there are some points where the ID movement comes closer to the theistic arguments.⁵⁰⁷ Typically ID’s design arguments are stated without the support of a cumulative case of other theistic arguments and without appealing to such theistic presuppositions. As I argued in chapter 3.5, the ID theorists recognize that theistic background beliefs can make the design argument more credible, and that atheistic background assumptions can make it less credible. However, they also emphasize that the design argument can be convincing on its own, as long as the possibility of design is not dogmatically denied a priori. ID’s design argument also leaves the question of the motives of

505 Dawes 2009, 43-46. I will look at some of these ideas in more detail in chapter five as I analyse theistic formulations of the fine-tuning design argument. The most influential defenders of theistic design arguments I will look at there are Robin Collins and Richard Swinburne. The idea of the high prior probability of theism is also crucial to Swinburne’s argument. According to Swinburne (2004a, chapter 5), because theism attempts to explain the whole world, its prior probability cannot be evaluated in the light of any background evidence.

Rather, the background probability of theism must be based purely on a priori considerations such as the simplicity of the hypothesis. Since Swinburne sees theism as a very simple hypothesis, he assigns it a relative high intrinsic probability of 0,5.

506 Dawes (2009, appendix) argues that intentional explanations are deductive, whereas Swinburne (2004a) sees them as probabilistic.

507 One way of doing this would be to construct ID as a theistic science starting from the presupposition that God exists. This would then supply the basic knowledge that a designer capable of creating life exists. The aim of ID’s project would then be to determine whether signs of the divine intelligence can be seen in the order of nature, and whether God created life through evolution or through miracles surpassing the laws of nature.

the designer open.⁵⁰⁸ As I argued in chapter 4.3, certain types of “design-like” order are thought to be best explained by design even if we know nothing about the designer’s motivations beforehand.⁵⁰⁹

So, in the ID movement's thought, the design argument does not need to predict what the designer would do. Rather, the argument simply posits that based on our experience, designers have the right sort of causal powers to explain certain types of order, and that this is sufficient to make design into a potential explanation. In chapter 3.3, I pointed out that in the philosophy of explanation, there has been a tendency to separate the concepts of

So, in the ID movement's thought, the design argument does not need to predict what the designer would do. Rather, the argument simply posits that based on our experience, designers have the right sort of causal powers to explain certain types of order, and that this is sufficient to make design into a potential explanation. In chapter 3.3, I pointed out that in the philosophy of explanation, there has been a tendency to separate the concepts of