# Lipton Key Extracts Curated passages from *Inference to the Best Explanation* (2nd ed., 2004) by [[Peter Lipton]]. --- ## Ch1: Induction — The Doing-Describing Gap and Black Box ### The doing-describing gap (p. 12) > Why is description so hard? One reason is a quite general gap between what we can do and what we can describe. You may know how to do something without knowing how you do it; indeed, this is the usual situation. It is one thing to know how to tie one's shoes or to ride a bike; it is quite another thing to be able to give a principled description of what it is that one knows. Chomsky's work on principles of language acquisition and Kuhn's work on scientific method are good cognitive examples. Their investigations would not be so important and controversial if ordinary speakers knew how they distinguished grammatical from ungrammatical sentences or normal scientists knew how they made their methodological judgments. Speakers and scientists employ diverse principles, but they are not conscious of them. The situation is similar in the case of inductive inference generally. Although we may partially articulate some of our inferences if, for example, we are called upon to defend them, we are not conscious of the diverse principles of inductive inference we constantly use. ### Black box inference (p. 13) > Since our principles of induction are neither available to introspection, nor otherwise observable, the evidence for their structure must be indirect. The project of description is one of black box inference, where we try to reconstruct the underlying mechanism on the basis of the superficial patterns of evidence and inference we observe in ourselves. This is no trivial problem. Part of the difficulty is simply the fact of underdetermination. [...] There will always be many different possible mechanisms that would produce the same patterns, so how can one decide which one is actually operating? In practice, however, as epistemologists we usually have the opposite problem: we can not even come up with a single description that would yield the patterns we observe. The situation is the same in scientific theorizing generally. There is always more than one account of the unobserved and often unobservable world that would account for what we observe, but scientists' actual difficulty is often to come up with even one theory that fits the observed facts. On reflection, then, it should not surprise us that the problem of description has turned out to be so difficult. Why should we suppose that the project of describing our inductive principles is going to be easier than it would be, say, to give a detailed account of the working of a computer on the basis of the correlations between keys pressed and images on the screen? --- ## Ch2: Explanation — No Humean Gap and Self-Evidencing Explanations ### No gap between explanation and seeming-to-explain (pp. 22–23) > We do not appear to know how to make the contrast between understanding and merely seeming to understand in a way that would make sense of the possibility that most of the things that meet all our standards for explanation might nonetheless not really explain. To put the matter another way, we do not see a gap between meeting our standards for the explanation and actually understanding in the way we easily see a gap between meeting our inductive standards and making an inference that is actually correct. > > It is not clear whether this is good or bad news for explanation. On the one hand, in the absence of a powerful skeptical argument, we feel less pressure to justify our practice. On the other, the absence seems to show that our grasp on explanation is even worse than our grasp on inference. We know that inferences are supposed to take us to truths and, as Hume's argument illustrates, we at least have some appreciation of the nature of these ends independently of the means we use to try to reach them. The situation is quite different for explanation. We may say that understanding is the goal of explanation, but we do not have a clear conception of understanding apart from whatever it is our explanations provide. If this is right, the absence of powerful skeptical arguments against explanation does not show that we are in better shape here than we are in the case of inference. Perhaps things are even worse for explanation: here we may not even know what we are *trying* to do. Once we know that something is the case, what is the point of asking why? ### Self-evidencing explanations (p. 24) > Suppose you ask me why there are certain peculiar tracks in the snow in front of my house. Looking at the tracks, I explain to you that a person on snowshoes recently passed this way. This is a perfectly good explanation, even if I did not see the person and so an essential part of my reason for believing my explanation are the very tracks whose existence I am explaining. Similarly, an astronomer may explain why the characteristic spectrum of a particular galaxy is shifted towards the red by giving its velocity of recession, even if an essential part of the evidence for saying that the galaxy is indeed moving away from us at that speed is the very red-shift that is being explained. These 'self-evidencing explanations' have a distinctive circularity: the person passing on snowshoes explains the tracks and the tracks provide the evidence for the passing. What is significant is that the circularity is benign: it spoils neither the explanation of the tracks nor the justification for the belief that someone did pass on snowshoes, neither the explanation of the red-shift nor the justification for the claim that the galaxy moves with that velocity. Self-evidencing explanations do, however, show that the reason model of explanation is untenable, since to take the explanation to provide a reason to believe the phenomenon after the phenomenon has been used as a reason to believe the explanation would be vicious. In other words, if the reason model were correct, self-evidencing explanations would be illicit, but self-evidencing explanations may be perfectly acceptable and are indeed ubiquitous. --- ## Ch7: Bayesian Abduction — The Squash Analogy and Realization Thesis ### The squash analogy (p. 108) > I want to suggest some of the ways in which explanatory considerations may be our way of running or realizing the mechanism of Bayesian conditionalization — the movement from prior to posterior probability — and our way of handling certain aspects of inference that conditionalizing does not address. [...] If these suggestions are along the right lines, then arguing that Inference to the Best Explanation is wrong because Bayesianism is right is like arguing that thinking about technique cannot help my squash game because the motion of the ball is governed by the laws of mechanics. Even if Bayesianism gave the mechanics of belief revision, Inference to the Best Explanation might yet illuminate its psychology. ### The realization thesis (pp. 106–107) > My objection to the argument that Inference to the Best Explanation is wrong because Bayesianism is right will not be that the premise is false, but that the argument is a non-sequitur, because Bayesianism and Inference to the Best Explanation are broadly compatible. It goes beyond the third response, however, in suggesting not only that Bayes's theorem and explanationism are compatible, but that they are complementary. Bayesian conditionalization can indeed be an engine of inference, but it is run in part on explanationist tracks. That is, explanatory considerations may play an important role in the actual mechanism by which inquirers 'realize' Bayesian reasoning. As we will see, explanatory considerations may help inquirers to determine prior probabilities, to move from prior to posterior probabilities, and to determine which data are relevant to the hypothesis under investigation. ### Summary of the realization relation (pp. 119–120) > This chapter has provided a brief exploration of the prospects for a compatibilist view of the relationship between Inference to the Best Explanation and Bayesianism. The relation that has motivated my discussion has been one of realization. Perhaps the simplest version of this view would make explanatory considerations a heuristic employed to make the likelihood judgments that the Bayesian process of conditionalization requires. Prior assessment of the quality of the explanation would be a way to fix on a likelihood, according to the rule of thumb that the better the explanation, the higher the likelihood. I have suggested that this may indeed be part of the story, but we have seen the relationship between Inference to the Best Explanation and Bayesianism is considerably more complicated and extensive than this, for a number of reasons. We have found that explanatory loveliness does not map simply onto likelihood, but may also play a role in assessing the priors. We have also found that explanatory considerations may play diverse roles in addition to the substantial jobs of helping to judge priors and likelihoods, such as determining relevant evidence and guiding hypothesis construction. --- ## Ch8: Explanation as a Guide to Inference — Background Constituting Standards ### Background beliefs as constitutive of explanatory standards (pp. 139–140) > A third and rather different inferential-explanatory role for the background exists because the background will incorporate particular explanatory standards. Thus a background might include a ban on explanations that appeal to teleology, to action at a distance or to irreducibly indeterministic processes, and it might privilege certain types of properties (e.g. 'primary properties'), marking them as providers of a particularly lovely explanation. Or what counts as a lovely explanation may be determined in part by previous explanations that serve an exemplary function, as Kuhn describes it (esp. 1970), or by more general 'styles of reasoning' (Hacking 1982, 1992). Variation in explanatory standard should be seen as occurring at diverse levels of generality, from features peculiar to small scientific specialties to those that may apply to almost the entire scientific community at a particular time. > > The background thus should be seen as affecting judgments of loveliness in two different ways: for a given standard, how lovely an explanation is will depend in part on what other explanations are already accepted, and the standard itself will be partially determined by the background. The importance of the background in inference, and the plausible suggestion that what counts as a lovely explanation is thus context sensitive, is entirely compatible with Inference to the Best Explanation, as I construe it. That account maintains that loveliness is a guide to likeliness, but it does not require that standards of loveliness are unchanging or independent of background belief. ### Fit with background as an inferential-explanatory virtue (pp. 122–123) > Fit with background is also an inferential factor that has an explanatory aspect. One reason this is so is because background beliefs may include beliefs about what sorts of accounts are genuinely explanatory. For example, at given stages of science no appeal to action at a distance or to an irreducibly chance mechanism could count as an adequate explanation, whatever its empirical adequacy. The role of background belief in determining the quality of an explanation shows how explanatory virtue is 'contextual', since the same hypothesis may provide a lovely explanation in one theoretical milieu but not be explanatory in another. --- ## Ch9: Loveliness and Truth — Inductive Powers, Conservatism, Preadaptation ### "What we cannot have are inductive powers without inductive achievements" (p. 158) > The moral of the story is that certain kinds of intermediate skepticism, of which the argument from underconsideration is one example, are incoherent. Because of the role of background beliefs in theory evaluation, what we cannot have are inductive powers without inductive achievements. ### The argument in full (pp. 157–158) > Scientists rank new theories with the help of background theories. According to the ranking premise of the argument from underconsideration, this ranking is highly reliable. For this to be the case, however, it is not enough that the scientists have any old background theories on the books with which to make the evaluation: these theories must be *probably true*, or at least probably approximately true. If most of the background theories were not even approximately true, they would skew the ranking, leading in some cases to placing an improbable theory ahead of a probable competitor, and perhaps leading generally to true theories, when generated, being ranked below falsehoods. [...] Hence, if scientists are highly reliable rankers, as the ranking premise asserts, the highest ranked theories have to be absolutely probable, not just more probable than the competition. This is only possible if the truth tends to lie among the candidate theories the scientists generate, which contradicts the no-privilege premise. ### Preadaptation analogy for hypothesis generation (pp. 150–151) > Let us extend the biological analogy. Darwin's mechanism faces the anomaly of the development of complex organs. The probability of a new complex organ, such as a wing, emerging all at once as a result of random mutation, is vanishingly small. If only a part of the organ is generated, however, it will not perform its function, and so will not be retained. How, then, can a complex organ evolve? The solution is an appeal to 'preadaptation'. Complex organs arose from simpler structures, and these were retained because they performed a useful though perhaps different function. A wing could not have evolved all at once, and a half-wing would not enable the animal to fly, but it might have been retained because it enabled the animal to swim or crawl. It later mutated into a more complex structure with a new function. > > Preadaptations are themselves the result of natural selection, and they form an essential part of the mechanism by which complex organs are generated. So natural selection plays a role in both the generation and the selection of complex organs. Similarly, the mechanism of explanatory selection plays a role both in the generation of the short list of plausible causal candidates and in the selection from this list. The background beliefs that help to generate the list are themselves the result of explanatory inferences whose function it was to explain different evidence. (This is like the Bayesian point that today's priors are yesterday's posteriors.) ### Inferential conservatism as a byproduct (p. 151) > Our method of generating candidate hypotheses is skewed so as to favor those that cohere with our background beliefs, and to disfavor those that, if accepted, would require us to reject much of the background. In this way, our background beliefs protect themselves, since they are more likely to be retained than they would be if we considered all the options. We simply tend not to consider hypotheses that would get them into trouble. The short list mechanism thus gives one explanation for our apparent policy of inferential conservatism. --- ## Ch10: Prediction and Prejudice — Post Hoc Ergo Ad Hoc and the Fudging Explanation ### The *post hoc ergo ad hoc* fallacy (p. 167) > Accommodating theories are obviously ad hoc in one sense, since 'ad hoc' can just mean purpose-built, and that is just what accommodating theories are. To claim, however, that a theory that is ad hoc in this sense is therefore poorly supported begs the question. Alternatively, 'ad hoc' can mean poorly supported, but this is no help, since the question is precisely why we should believe that accommodating theories are in this sense ad hoc. To assume that accommodating theories are ad hoc in the sense of poorly supported is to commit what might be called the '*post hoc ergo ad hoc*' fallacy. The simple appeal to the notion of an ad hoc theory names the problem but does not solve it. ### The fudging explanation (p. 170) > When data need to be accommodated, there is a motive to force a theory and auxiliaries to make the accommodation. The scientist knows the answer she must get, and she does whatever it takes to get it. The result may be an unnatural choice or modification of the theory and auxiliaries that results in a relatively poor explanation and so weak support, a choice she might not have made if she did not already know the answer she ought to get. In the case of prediction, by contrast, there is no motive for fudging, since the scientist does not know the right answer in advance. She will instead make her prediction on the basis of the most natural and most explanatory theory and auxiliaries she can produce. As a result, if the prediction turns out to have been correct, it provides stronger reason to believe the theory that generated it. So there is reason to suspect accommodations that do not apply to predictions, and this makes predictions better. ### The crossword analogy (pp. 170–171) > Consider a crossword puzzle. Suppose that you are trying to find a word in a position where some of the intersecting words are already in place. There are two ways you can proceed. Having read the clue, you can look at the letters already in place and use them as a guide to the correct answer. Alternatively, you can think up an answer to the clue with the requisite number of letters, and only then check whether it is consistent with the intersecting letters. The first strategy corresponds to accommodation, the second to prediction. [...] If, however, you are fortunate enough to come up with a word of the right length without using the intersections, and those letters are then found to match, one might hold that this matching provides more reason to believe that the word is correct than there would be if you had adopted the accommodating strategy for the word. --- ## Conclusion (pp. 207–210) ### Inference to the Loveliest Explanation, not merely the Likeliest (p. 207) > I also made the distinction between the explanation most warranted by the evidence — the likeliest explanation — and the explanation which would, if true, provide the most understanding — the loveliest explanation. The model tends to triviality if we understand 'best' as likeliest, since the sources of our judgments of likeliness are precisely what the model is supposed to illuminate. Inference to the Loveliest Explanation, by contrast, captures the central idea that the explanatory virtues are guides to inference, so I urged that we construe the model in this ambitious and interesting form. ### The model explains its own discovery (p. 209) > One of the main attractions of the model is that it accounts in a natural and unified way both for the inferences to unobservable entities and processes that characterize much scientific research and for many of the mundane inferences about middle sized dry goods that we make every day. It is also to its credit that the model gives a natural account of its own discovery, that the model may itself be the best available explanation of our inductive behavior since, as we have seen, that inference must itself be inductive and moreover an inference to a largely unobservable mechanism. ### Closing remark (p. 210) > I also take some comfort in the otherwise discouraging fact that an account of our inductive practices does not have to be very good to be the best we now have.