Inference to the best explanation | Peter Lipton (2000)
In W.H. Newton-Smith (Ed.), A Companion to the Philosophy of Science, 184-194. Oxford, UK: Blackwell.
Copied notes from paper. If I cite this author, I should re-read the original paper to make sure I do not accidentally copy phrasings, even though it looks like I made an effort to quote properly. N.B. the page numbers are relative to the beginning (for some reason my version of the paper only has relative numbering).
Only in some cases can you show the falsity of a hypothesis deductively with the truth of experimental data, namely when the observed evidence plainly contradicts the hypothesis.
It is more common to use a non-demonstrative relation between evidence and hypothesis, meaning that "it remains possible that the hypothesis is false even though the data are correct" (1)
Explanatory considerations guide (scientific) inference from the available evidence to the hypothesis that, if it is correct, best explains the evidence.
Example: the hypothesis of natural selection does not follow logically from biological evidence, but is its best explanation.
Example: Sherlock Holmes claims he uses deduction, but in fact, he uses abduction/ inference to the best explanation (IBE).
Self-Evidencing circularity
Notice the emphasis on "guide" in the following citation:
Its governing idea is that explanatory considerations are a guide to inference, that scientists infer from the available evidence to the hypothesis which would, if correct, best explain that evidence. (1)
In inference to the best explanation, a hypothesis is supported (not proven) by the observations (or "evidence" in Bayesian terms) that they are supposed to explain.
Thus there is a sort of inversion of the "natural order".
According to that natural view, inference is prior to explanation. First the scientist must decide which hypotheses to accept; then, when called upon to explain some observation, she will draw from her pool of accepted hypotheses. According to Inference to the Best Explanation, by contrast, it is by only by asking [typo in original] how well various hypotheses would explain the available evidence that she can determine which hypotheses merit acceptance. In this sense, Inference to the Best Explanation has it that explanation is prior to inference.
You would expect that you first infer a hypothesis, which then in turn explains the data (inference -> explanation). But in IBE, you only infer a hypothesis because they explain the data (explanation -> inference).
Inference is thus indeed "guided" by explanatory considerations.
Example of this "curious circularity" (2)
a star's speed of recession explains why its characteristic spectrum is red-shifted by a specified amount, but the observed red-shift may be an essential part of the reason the astronomer has for believing that the star is receding at that speed. (2)
The recession is used to explain the red-shift and the red-shift is used to confirm the recession, yet the recession hypothesis may be both explanatory and well-supported. According to Inference to the Best explanation, this is a common situation in science: hypotheses are supported by the very observations they are supposed to explain. Moreover, on this model, the observations support the hypothesis precisely because it would explain them. (2)
The enumerative model of induction
What is usually known as induction is a form of IBE that focuses on enumeration.
Lipton argues induction is inadequate for scientific inferences, because it is both too permissive and too restrictive.
- Too permissive
Consider Hempel's 'Raven Paradox' and the 'New Riddle of Induction'.
- Too restrictive.
The "more of the same"-principle of induction does not apply because scientific hypotheses often appeal to non-accessible phenomena that are not given in the evidence. E.g. some hypotheses about a particle in physics, of which it is hypothesized that it must exist: nothing like it is observed yet, and still this type of inference is possible in science.
The enumerate model conversely extends upon some countable and thus observable phenomenon that is currently observed.
The nomological-deductive model
The nomological-deductive model allows for "vertical inference" that may include assumptions about states currently not present in the evidence (cf. the previous physics example). In this model, a hypothesis explains something when, together with "auxiliary premises" (3) it allows deduction of the evidence.
However, the model is still too permissive.
Example from Lipton: If some prediction P is inferred (because some evidence can be deduced from it), then P \/ H can also be inferred for any arbitrary hypothesis H. Meaning: any hypothesis may be inferred.
There's one thing I don't really get from this example. Assume you don't infer P; then from P \/ H you couldn't necessarily infer the evidence right, because you don't know if P or H is the case?
My reasoning was this: Assuming that some evidence E may be deduced from a hypothesis, then E \/ P may also be deduced. Meaning: the hypothesis may explain anything, quite literally.
Notice the difference between the meanings.
This is a problem because we expect IBE to discriminate between competing explanations.
In the nomological-deductive model IBE would be reduced to/isomorphic to the hypothetico-deductive model, rather than a form of inductive reasoning. It would amount to reducing inductive reasoning to deductive reasoning!
The main requirement for an account of inference to the best explanation
- What is an explanation?
- Which explanation is best (i.e. better than other explanations)?
Inference to the best explanation suggests that inference always implies choosing one hypotheses over the others, but this is not entirely accurate:
Yes scientists are sometimes agnostic, unwilling to infer any of the available hypotheses, and they are also sometimes happy to infer more than one explanation, when the explanations are compatible. (5)
Horrible rephrasing: "inference to the best of the available competing explanations, when the best one is sufficiently good" (5)
An account of IBE then must answer:
- When is an explanation sufficiently good?
- And what factors make one explanation better than another?
With respect to the latter point, Lipton introduces the distinction between most likely and most lovely explanations.
Lipton uses a clear example to illustrate the difference:
if one says that smoking opium tends to put people to sleep because opium has a 'dormative' power, one is giving an explanation that is very likely to be correct but not at all lovely: it provides very little understanding. (5)
Reducing 'best' to 'the most likely' would "severely reduce the interest of the model by pushing it towards triviality" (5). This is also a point that Glass keeps repeating.
The crux is that in principle the most likely explanation does not necessarily add any understanding; and would thus not have any value for a scientist. IBE is an account that aims to describe how, among others, scientists reach a particular inference. That does not in any way imply that the loveliest explanation would not be (the most) likely.
Scientists do infer what they judge to be the likeliest hypothesis, but the main point of a model of inference is precisely to say how these judgements are reached, to give what scientists take to be the symptoms of likeliness. (5)
Beautiful phrasing: symptoms of likeness.
And then the home-run with a great analogy:
To say that scientists infer the likeliest explanations is perilously similar to saying that great chefs prepare the tastiest meals: true perhaps, but not very informative if one wants to know the secrets of their success. (5-6)
- Is "best" the most likely or the most lovely?
- The former is a strong push towards triviality (p.5)
- The latter is an explanation that increases understanding.
TODO: review this whole section.
Inference to the loveliest explanation
Loveliness is a scientists guide to what is considered most likely.
Its central claim is that scientists take loveliness as a guide to likeliness, that the explanation that would, if correct, provide the most understanding, is the explanation that is judged likeliest to be correct. (6)
Note that there is an apparent contradiction in this sentence, but it's not in fact contradictory. Lipton's program is to show how loveliness guides estimates of likeliness, and ideally, (that's the hardest part perhaps), to show that these two interpretations of "best" are often in agreement.
However, despite this not necessarily being a contradiction, I'm not so convinced by Lipton's arguments that loveliness guides likeliness in such a way that they in practice agree. That would imply that the more lovely explanations are then also the more likely ones and vice versa.
But it's not hard to give counter arguments where likely and lovely diverge. When a doctor says "the patient is ill" this is likely to be correct but not because it offers most understanding, i.e. likeliness and loveliness can strongly diverge.
But perhaps it makes a difference here if Lipton would only commit to saying that lovely explanations are also likely, whereas likely explanations are not necessarily lovely. But this distinction is not sufficiently made. Now Lipton simply says that they tend to "agree" which suggests mutual co-occurrence.
Three challenges:
- What are explanatory virtues? The features of a good explanation by which it adds much understanding.
- These explanatory features match judgments of likeliness
- Glass discusses this in terms of "truth-tracking".
- Given the previous two, show that loveliness is in fact guiding likeliness.
Candidates for explanatory virtues:
- Scope (explain more types of phenomena)
- Precision
- Mechanism (provide information about underlying mechanisms)
- Unification (unify apparently disparate phenomena)
- Simplicity (cf. Ockham's Razor)
Example: there is no "uncontroversial analysis of unification or simplicity"; and are these artefacts of the specific formulation of the explanation or the explained hypothesis?
Contrastive explanation
p. 7-8: Good explanation of why explanatory virtues probably require a contrastive why question: Why P rather than Q.
This means that the quality of the explanation for P is dependent on the foil Q.
Explanatory and inferential virtues
If Inference to the Best Explanation is along the right lines, then the lovelier explanations ought also in general to be judged likelier. Here the situation looks promising, since the features we have tentatively identified as explanatory virtues seem also to be inferential virtues, that is, features that lend support to a hypothesis. (7)
When do they overlap, when do they not, and when is this not an issue?
The overlap between explanatory and inferential virtues is certainly not perfect, but at least some cases of hypotheses taht are likely but not lovely, or conversely, do not pose a particular threat to Inference to the Best Explanation. As we have already seen, the doermative power explanation of opium's soporific effect is very likely but not at all lovely; but this is not a threat to the model, properly construed. (7)
Why? The banal account and the deeper molecular account are not competing according to Lipton and both can be inferred without problem. So I guess here the idea of "contrastive explanation" is used in another way: to indicate where IBE has to choose between hypotheses and where not.
p.8: his favorite Semmelweis example: "contrasts in why-questions often correspond to contrasts in the available evidence" (8)
Semmelweis inferred a hypothesis that explains the constrast between wards. I think the idea is that this is an explanatory virtue and, in turn, aids the correctness/likelihood of the hypothesis.
Is loveliness guiding?
Lipton's arguments here are not that strong. He says:
- If IBE gives a better account than other related theories, this is support for IBE. Fair enough. But the actual argument is still lacking at this point.
- "if there is a good match between loveliness and likeliness, as the guiding challenge grants, this is presumably not a coincidence and so itself calls for an explanation. Why should it be that the hypotheses that scientists judge likeliest to be correct are also those that would provide the most understanding if they were correct?" (9)
The argument w.r.t the second point is curious, to say the least.
If scientists select hypotheses on the basis of their explanatory virtues, the match between loveliness and judgments of likeliness follows as a matter of course. Unless the opponents of the model can give a better account of the match, the challenge has been met. (9)
From description of inferences to their (inductive) justification
One page 9 there is a switch from description to justification.
On the issue of justification: Hume's problem of justification is that induction is by definition not deductively valid, and that we also can't justify induction based on inducing the next instance will be confirmatory; because this presupposes induction. E.g. you can't argue that induction will most likely be accurate in the future. This is itself a form of induction.
IBE does not solve this problem, even though Hume clearly discusses enumerative induction only.
IBE and defense of scientific realism
Constructive empiricism of Van Fraassen on p.11. a theory may be adequate, but only observable consequences (deduced?) are true.
Putnam's Miracle Argument:
- IBE good solution for descriptive problem
- IBE can be used to defend scientific realism
Suppose that all the many and varied predictions derived from a particular scientific theory are found to be correct: what is the best explanation of this predictive success? According to Putnam, the best explanation is that the theory itself is true. If the theory were true, then the truth of its deductive consequences would follow as a matter of course; but if the hypothesis were false, it would be a 'miracle' that all its observed consequences were found to be correct. (11)
So Putnam used a 2nd order inference.
P12-13 discuss counterarguments.
The conclusion is that the uses of IBE for justification do not appear promising and have many problems. It's main ambitions are descriptive.