# Inference to the Best Explanation  ## Metadata - Author: [[Peter Lipton]] - Full Title: Inference to the Best Explanation - Category: #books - Summary: insert summary - My notes: - Summary: Peter Lipton explains that Inference to the Best Explanation (IBE) is a strong method for understanding scientific reasoning. He uses Semmelweis’s research to show that IBE works better than traditional deductive methods. The text also discusses challenges to IBE, but argues that it remains a valuable way to make sense of evidence and explanation. ## LLM Chats ## NotebookLM ## LLM Audio ## Highlights > It is easy to ride a bicycle, but hard to describe how it is done; it is easy to distinguish between grammatical and ungrammatical strings of words in one’s native tongue, but hard to describe the principles that underlie those judgments. In the cases of inference and explanation, the contrast between what we can do and what we can describe is stark, for we are remarkably bad at principled description. ([View Highlight](https://read.readwise.io/read/01khjtbwx156bcxsjkhes1zt9w)) > In [chapter 4](12_chap4.xhtml#c4), I flagged two general reasons one might doubt that Inference to the Best Explanation could give an account of a reliable inferential practice, reasons indeed for thinking it very unlikely that using loveliness as a guide to likeliness should be reliable. The first, ‘Hungerford’s objection’ (‘Beauty is in the eye of the beholder’), was that explanatory loveliness is too subjective and variable to give a suitably objective account of inference. The other, ‘Voltaire’s objection’, was that Inference to the Best Explanation makes the successes of our inferential practices a miracle. We are to infer that the hypothesis which would, if true, provide the loveliest explanation of our evidence, is therefore the explanation that is likeliest to be true. But why should we believe that we inhabit the loveliest of all possible worlds? If loveliness is subjective, it is no guide to inference; and even if it is objective, w­hy should it line with truth? ([View Highlight](https://read.readwise.io/read/01kh3z15r7vha905rqmhdy1q6p)) > The first point to make in response to Voltaire’s objection is that we should not be asked to prove too much. Induction by its nature is not a form of inference that is guaranteed to yield only truths from truths. Moreover, Inference to the Best Explanation is meant to be descriptively adequate, and the psychological research we discussed in the last two chapters brings out some sources of systematic unreliability in our actual inferential practices. (One might mount a parallel argument from the history of science, as in the so-called ‘pessimistic induction’ from the falsity of past scientific theories ([Laudan 1984](21_references.xhtml#r68)).) If Inference to the Best Explanation can account for some of this, that is to its credit. It is no objection to explanationism that it does not make us out to be more reliable than we actually are. ([View Highlight](https://read.readwise.io/read/01kh3z45jvgc0d1khv15v7rpaa)) > Still, I take it that our inductive practices are reasonably reliable, certainly better than random guessing. And at this level, one should perhaps say that Voltaire is right: even the moderate reliability of Inference to the Best Explanation would require a miracle. For we are talking about induction here and there is a sense in which the success of induction is miraculous or inexplicable on any account of how it is done. This is one way of putting the conclusion of David Hume’s great skeptical argument. Hume himself used an over-simple ‘More of the Same’ description of inductive inference but, as I noted in [chapter 1](09_chap1.xhtml#c1), his skeptical argument does not depend on the particular description he gave. Whatever account one gives of our non-deductive inferences, there is no way to show a priori that they will be successful, because to say that they are non-deductive is just to say that there are possible worlds where they fail. Nor is there any way to show this a posteriori since, given only our evidence to this point and all a priori truths, the claim that our inferences will be successful is a claim that could only be the conclusion of a non-deductive argument and so would beg the question. In short, the impossibility of justifying induction does not depend on a particular account of our practices, but only on the fact that they are inductive. Consequently, in the absence of an answer to Hume, any account of induction leaves inductive success miraculous. That is very bad news, but of course it is no worse news for Inference to the Best Explanation than any other account of induction. Inference to the Best Explanation would not have us infer that the loveliest possible world is actual; at most, it has us infer the loveliest of those worlds where our observations hold. And what Hume showed was that the claim that any system of choosing from among those worlds is reliable is indefensible. ([View Highlight](https://read.readwise.io/read/01kh3z6vdkbw2ch31ttamyn7pk)) > So, a response to Voltaire’s objection is to say that it reduces to Hume’s problem. This is hardly a solution, but it would show that Inference to the Best Explanation is in no worse shape than any other account of induction. Leaving Humean skepticism to one side, however, might one claim that Inference to the Best Explanation would make the reliability of induction somehow more surprising than some other account? That would be the case, for example, if there was an incoherence in explanationism that does not arise in other accounts of induction. I can think of two quite different forms this incoherence might be supposed to take, but careful readers will by now know why I hold neither of them to be realized in explanationism properly construed. The first is that Inference to the Best Explanation would not be epistemically effective, since an actual explanation must be true, so one would have to know the truth before one could infer an explanation. According to this objection, explanationism gets things backwards, because we must infer before we explain. As we saw in [chapter 4](12_chap4.xhtml#c4), however, this objection is based on a misunderstanding of Inference to the Best Explanation. Perhaps actual explanations must be true, but the account has us infer to the best potential explanation, a hypothesis that would explain if true. There is no incoherence here. The second claimed source of incoherence is probabilistic: someone who uses Inference to the Best Explanation will violate Bayesian constraints on belief revision and so will be susceptible to a dutch book ([van Fraassen 1989](21_references.xhtml#r119): 160–70). As we saw in [chapter 7](15_chap7.xhtml#c7), however, explanationism can be understood in a way that respects Bayesian constraints and indeed in a way that serves a Bayesian approach to inference. The probabilistic incoherence that the dutch book argument displays only occurs if explanatory considerations are supposed to boost the posterior probability after conditionalization has taken place; but this leaves us free to engage in the coherent use of explanatory considerations earlier in the process, for example in the determination of priors and likelihoods (cf. [Harman 1999](21_references.xhtml#r50): ch. 4; [Okasha 2000](21_references.xhtml#r88)). ([View Highlight](https://read.readwise.io/read/01kh3zb2wv0vq6d5fxxxxbk6xb)) > Indeed far from suffering from an incoherence that other accounts avoid, Inference to the Best Explanation in fact inherits whatever justification various other accounts of inference would provide. As we saw with the catch-22 of the [last chapter](19_chap11.xhtml#c11), the structural similarity between explanationism and other accounts of inference presents a peculiar challenge to the descriptive task of showing that explanatory considerations are actually driving inference; but in the context of justification and Voltaire’s objection these similarities are an advantage. Thus insofar as Bayesian inference avoids Voltaire’s objection, so can Inference to the Best Explanation, since we may see the latter as a way of realizing the former. Similarly, if the method of difference strikes you as avoiding the objection, so should Inference to the Best Explanation, in light of the structural similarity between simple contrastive inference and contrastive explanation, a similarity I exploited in [chapters 3](11_chap3.xhtml#c3) to [6](14_chap6.xhtml#c6). Moreover, as we noted in the [last chapter](19_chap11.xhtml#c11), our explanatory standards and the inferences they support are highly sensitive to our background beliefs, so it seems that Inference to the Best Explanation has the resource to capture the way reliable patterns of inference depend on taking this background into account. Since those background beliefs were themselves generated through explanatory inference, we have a kind of feedback between judgments of likeliness and judgments of loveliness. Successful inferences become part of the background, and influence what counts as a lovely explanation and thus influence future inferences. This is as it should be: as we learn more about the world, we not only know more but we also become better inferential instruments. ([View Highlight](https://read.readwise.io/read/01kh3ze3rr51410arz0kw0tm29)) > Voltaire’s objection thus appears ill founded. First, our actual inductive practices are far from perfectly reliable, so it is no criticism of explanationism that it would not make them so. Secondly, although Inference to the Best Explanation does not solve the Humean problem of induction, nothing else does either. Thirdly, there is no reason to believe explanationism is incoherent and, fourthly, indeed it can capture normatively attractive features of Mill’s methods, Bayesianism and the essential role of background belief to inference. To these points I would add an inversion of the worry I expressed earlier about the relationship between the normative and the descriptive. Hume notwithstanding, we believe that our inductive methods are pretty reliable. The worry was that a belief that the reliability of Inference to the Best Explanation would be miraculous would thus undermine confidence in its descriptive adequacy as well. My hope is rather that by this stage you are convinced of the descriptive merits of explanationism, so insofar as you believe that our actual practices are reliable, you will tend to discount Voltaire’s objection. ([View Highlight](https://read.readwise.io/read/01kh3zfqzxzyhxbesqzj997kwx)) > 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.) We consider only the few potential explanations of what we observe that seem reasonably plausible, and the plausibility judgments may not seem to be based on explanatory considerations; but they are, if the background beliefs that generate them are so based. Those beliefs now serve as heuristics that guide us to new inferences, by restricting the range of actual candidates, much as preadaptations limit the candidate organisms that are generated ([Stein and Lipton 1989](21_references.xhtml#r113)). So Inference to the Best Explanation helps to account for the generation of live candidates because it helps to account for the earlier inferences that guide this process. The analogy to the evolution of complex organs also serves in another way. By exploiting preadaptations, the Darwinian mechanism yields the result that old and new variations fit together into a coherent, complex organ. Similarly, the mechanism of generating hypotheses favors those that are extensions of explanations already accepted, and so leads towards a unified general explanatory scheme. This scheme is a complex organ that could not have been generated in one go, but it is built on earlier inferences, themselves selected on explanatory grounds and guiding the generation of additional structures. ([View Highlight](https://read.readwise.io/read/01kh3znebrn6xvyspk0wesvybb))