1.5: Inference to the Best Explanation
In making this inference one infers, from the fact that a certain hypothesis would explain the evidence, to the truth of that hypothesis. In general, there will be several hypotheses which might explain the evidence, so one must be able to reject all such alternative hypotheses before one is warranted in making the inference. Thus, one infers, from the premise that a given hypothesis would provide a “better” explanation for the evidence than would any other hypothesis, to the conclusion that the given hypothesis is true.
—GILBERT HARMAN1
Inference to the Best Explanation
We have been treating the expression inference to the best explanation as technical jargon. It is a way of looking at evidence or at least purported evidence in an inductive argument. If we look at the component words in this expression, we will discover quite a lot. First of all, we are dealing with an inference . For most purposes, we can consider this as just another way of saying that we have an argument to be considered. This inference is to an explanation . But we are not dealing with just an inference to an explanation but to the best explanation. This implies two very important things. First, in order for there to be a comparison, there must be other possible explanations of the data in the argument, rival explanations . And the argument is also committed to this original explanation being better than all these rivals. Therefore there seems to be some rank ordering of the explanatory candidates, even if this is not explicitly stated.
I will use all this as a way of articulating a test of the quality of evidence within an argument. This test will be most straightforward when you are what I have called a consumer of an argument. Connie thought she had evidence that her boyfriend was smooching Mary Jane. Holmes had evidence about Watson’s decision about the investment and about what happened at Ridling Thorpe Manor. We must decide whether these arguments are any good. Was the evidence for these hypotheses strong? What I am going to call the inference-to-the-best-explanation (IBE) “recipe” is a procedure for answering these kinds of evaluative questions.
Inference-to-the-Best-Explanation Recipe
1. Schematize the argument.
2. List some serious (hopefully challenging) rival explanations.
3. Rank order all the explanations—the original along with the rivals.
4. Based on the rank order, see if the original is the best explanation. If it is, the evidence has passed the test and looks pretty good. If it isn’t, it’s failed the test, and the evidence is weak and maybe nonexistent.
Let’s first apply the test or recipe to the simple argument presented in the pop song “Lipstick on Your Collar” that we introduced in chapter 1.
Schematizing Connie’s Argument
That fateful evening at the record hop, Connie was confronted with data, mainly her own simple observations, which cry out for explanation. Where did the lipstick stain come from? Why was he gone for so long? Why did he say it belonged to her when the stain was red and her lipstick was baby pink? Why when Mary Jane appeared was her lipstick all a mess? Although neither a trained natural scientist nor an experienced detective, Connie easily forms an explanatory hypothesis. When she then writes her sad song, she implicitly asks us to account for what happened. Here’s how I would schematize Connie’s evidence for her theory that her boyfriend had been smooching Mary Jane when he left her alone at the record hop.
e 1 . He left Connie all alone at the record hop.
e 2 . He was gone for half an hour or more.
e 3 . When he returned, there was a lipstick stain on his collar.
e 4 . When confronted, he claimed that the stain came from Connie’s lipstick.
e 5 . The stain was red.
e 6 . Connie’s lipstick was baby pink.
e 7 . Mary Jane’s lipstick was all a mess.
t 0 . He had been smooching Mary Jane during the half-hour absence.
Rival Explanations (of Connie’s Data)
For our purposes, rival explanations will be accounts of the data that flat-out deny the original explanation and substitute a completely different story of the data offered as evidence. It will be useful to imagine each truly rival account of evidence as starting out with a lengthy preliminary phrase—“ no, no, no , he was not smooching Mary Jane during his absence from the record hop; rather . . .” This is important because the original explanation might be phrased in very different language.
t′ 0 . He and Mary Jane ditched Connie so they could make out.
Or an account might offer a more (or less) detailed account of what happened.
t″ 0 . He headed for a soda pop but met Mary Jane and couldn’t control himself.
Neither t′ 0 nor t″ 0 will count as rival explanations. If you were to challenge Connie with them, I don’t think she would say, “Oh, yeah, maybe I was wrong,” but rather she’d exclaim, “Exactly!”
So what else might have happened? Connie never suggests any rival explanations, but they are easy enough to formulate. He went out for a soda pop, just as he said. When asked about the lipstick stain, he responded that it came from Connie, since she was the only one he had been smooching. The laundry detergent his mother uses left a residue on his collar that chemically changed the baby pink lipstick to a bright red color. Mary Jane had been smooching a new guy she met at the record hop, and this messed up her lipstick. We can label this rival explanation t 1 .
t 1 . The lipstick changed from pink to red because of a chemical reaction with his mother’s laundry detergent.
Or the circumstances might be more sinister. He left Connie all alone because he was feeling ill but thought it more decorous to say he wanted a soda pop. Mary Jane has been harboring a grudge against Connie since the last student council meeting. She found him in the lobby, distracted him, and wiped lipstick on his collar. After he left to return to Connie, Mary Jane smudged her lipstick with the back of her hand. When he returned and was asked about the stain, he told Connie it was hers because she was the only one he had been smooching. Let’s label this one t 2 .
t 2 . Mary Jane staged the whole thing out of revenge.
t 1 and t 2 were the rival explanations that I came up with when I first used this example in a conference paper several years ago now. I subsequently used the example in quizzes in several of my critical thinking courses. Many of my students suggested a rival explanation that I now believe is much more challenging to Connie’s original theory than either of my earlier attempts. Perhaps the stain really did come from Connie but not that evening at the record hop. She might have been wearing bright red lipstick when they smooched last weekend. He’s not too hot at doing his laundry regularly and wore the stained shirt to the record hop.
t 3 . The stain came from a previous episode of smooching when Connie was wearing red lipstick.
Rank Ordering Explanations (for Connie’s Argument)
We now have on the table four competing accounts of what happened at the record hop.
t 0 . He had been smooching Mary Jane during the half-hour absence.
t 1 . The lipstick changed from pink to red because of a chemical reaction with his mother’s laundry detergent.
t 2 . Mary Jane staged the whole thing out of revenge.
t 3 . The stain came from a previous episode of smooching when Connie was wearing red lipstick.
Inference to the best explanation asks us to judge one of these explanations as better than all the rest. How in the world do we start the process of judging one explanation as superior to another? What counts and what doesn’t count in such a comparison? We will address this in some detail in a later chapter, but for now, let’s simply phrase the question as “ Which account makes the best sense of what we know? ”
I assume that both t 1 and t 2 would rank way down on your list, compared to t 0 and t 3 . Isn’t part of the reason for this, the fact that both of them introduce something “out of the blue” to explain the absence? Where did this mysterious laundry detergent come from? Or this whole grudge on the part of someone she had considered her best friend?
What about t 0 and t 3 , however? They both seem reasonable enough. Let me simply assert some factors that do not count in rank ordering explanations. The best explanation is not necessarily the one we like the best, nor the one that best accords with our politics, religion, or moral perspectives. It is the one that is most plausible .
Here comes a scary fact! You have to make the judgment about which explanation is best. There is no “objective,” “reliable” test or formula you can utilize that automatically identifies the best explanation. The whole recipe, therefore, rests on a step that is candidly, unavoidably subjective . When it comes to flavors of ice cream or styles of beer, being subjective means that people’s preferences are relative to who they are and are, consequently, all over the place. If evidence evaluation is the same, we’re done for, and I can stop writing my book and teaching my courses as I do. Fortunately, I believe, explanatory plausibility is very different from beer preferences. Even though each of us, individual subjects, must rank order alternative accounts for ourselves, it turns out that in a great number of contexts—courts of law, the natural sciences, and even stories about suspicious lipstick stains— subjective judgments about plausibility can turn out to be intersubjective . When all is said and done, when we think about it as free from prejudice and bias as we can be, we discover widespread agreement about what the best explanation is. We are the most intelligent species that has ever existed, and part of being intelligent is being darn good at spotting the best explanation of what’s happening around us.
I rank order our four explanations in the following order:
t 0 . He had been smooching Mary Jane during the half-hour absence.
t 3 . The stain came from a previous episode of smooching when Connie was wearing red lipstick.
t 1 . The lipstick changed from pink to red because of a chemical reaction with his mother’s laundry detergent.
t 2 . Mary Jane staged the whole thing out of revenge.
I grant you that t 0 and t 3 are pretty close to one another, but I think Connie would not have been so surprised at all of this if she regularly wore bright red lipstick, and besides, the whole idea of Connie having red lipstick is sort of out of the blue as well.
Assessment of (Connie’s) Evidence
The whole purpose of the inference-to-the-best-explanation recipe is to assess the quality of evidence in an argument. We need to find the best explanation. The whole test depends on what is in first place. In my considered judgment, Connie’s theory was the best explanation, and therefore, her evidence is pretty good . For all the talk about intersubjectivity, I fully realize that some of you will have ranked t 3 ahead of t 0 . Those of you who have come to that judgment would say that since there is a better explanation of the facts at the record hop, Connie’s evidence is weak .
I have been asking my students to use the inference-to-the-best-explanation recipe to assess the quality of evidence presented in an argument for more than three decades. The single most common mistake that my students make, including some of the best and most intelligent, is to forget about the purpose of the recipe and neglect to offer an assessment of the evidence in the argument . They often beautifully schematize it, come up with some challenging rival explanations, and offer subtle and insightful comments about how and why they have rank ordered as they have but then remain silent on the quality of the evidence. I am almost tempted to include a fifth step in the recipe saying something such as the following:
- 5. Conclude your analysis with one of the following two sentences: “ Since the original theory proved to be the best explanation of the data in the evidence, the argument’s evidence is pretty good (strong, etc.) ” or “ Since there is a better explanation of the data in the evidence, the argument’s evidence is weak (poor, nonexistent, etc.) .”
Step 4 requires an explicit evaluation of the evidence, as it was presented and schematized, in the original argument!
What about Ties?
Suppose you came to the conclusion that smooching Mary Jane and smooching Connie last weekend were equally plausible explanations of all the data you had? What happens in the recipe when the original and one of the rivals are tied for first place?
This is a classic half-full, half-empty kind of dilemma. You might say that since the original is tied as the best explanation , there’s some evidence for that conclusion. You might also say, however, that since there’s a rival explanation that’s tied as the best explanation, the evidence is not so hot. I think that whichever way we go, the message is really the same. The original’s being tied as the best explanation allows us to see why someone would offer the argument in its defense in the first place and why there is some evidence that seems to support it. A rival being tied as the best explanation tells us that the evidence is far from conclusive. Ideally, in such a case, we go out and do a little more investigating and see if we could discover some new data that would help break the tie. And indeed, the whole subject of new data is the topic for our next chapter. But before heading there, let’s apply the recipe to scientific argument.
The Origins of Natural Language
The following comes from an article by two prominent cognitive scientists, Stephen Pinker and Paul Bloom:
All human societies have language. As far as we know they always did; language was not invented by some groups and spread to others like agriculture or the alphabet. . . . The grammars of industrial societies are no more complex than the grammars of hunter-gatherers. . . . Within societies, individual humans are proficient language users regardless of intelligence, social status, or level of education. Children are fluent speakers of complex grammatical sentences by the age of three, without benefit of formal instruction. They are capable of inventing languages that are more systematic than those they hear, showing resemblances to languages that they have never heard, and obey grammatical principles for which there is no evidence in their environments. . . . The ability to use a natural language belongs more to the study of human biology than human culture; it is a topic like echolocation in bats or stereopsis in monkeys, not like writing or the wheel. . . . We argue that language is no different from other complex abilities to such as echolocation or stereopsis, and the only way to explain the origin of such abilities is through the theory of natural selection. 2
Pinker and Bloom’s thesis is that our knowledge of syntax or grammar is not something we learn but is innate, something we are born with. Spiders don’t learn to spin webs; they simply spin them. Bats don’t learn to use echolocation; they simply use it to navigate. Babies don’t learn grammar; they already possess it as they learn their native language.
Please take a moment to try your hand at schematizing Pinker and Bloom’s argument before reading further.
The Argument Schematized
Pinker and Bloom are defending a scientific hypothesis about the origins of natural language and their conviction that its history lies in natural selection.
t 0 . The only way to explain the origin of language is through the theory of natural selection.
They present a good deal of data in support of their theory. Here is how I would schematize their evidence:
e 1 . All human societies have language.
e 2 . They always have had language.
e 3 . Language was not invented and did not spread.
e 4 . Contemporary grammars are no more complex than those of hunter-gatherers.
e 5 . Humans are proficient language users regardless of intelligence, social status, or level of education.
e 6 . Children are fluent speakers of complex grammatical sentences by the age of three, without benefit of formal instruction.
e 7 . Children are capable of inventing languages that are more systematic than those they hear, showing resemblances to languages that they have never heard and obeying grammatical principles for which there is no evidence in their environments.
t 0 . The origin of language is explained through the theory of natural selection.
Rival Explanations (of Pinker and Bloom’s Data)
A superficially similar theory was first introduced by Noam Chomsky in the late 1950s. He argued that natural selection produced larger brains and that the ability to master a natural language so easily was a happy by-product of this larger brain size. He was quite emphatic that language was not “selected for” in our evolutionary history. Chomsky’s view was expanded upon by the important evolutionary biologist Steven Gould:
Yes, the brain got bigger by natural selection. But as a result of this size, and the neural density and connectivity thus imparted, human brains could perform an immense range of functions quite unrelated to the original reasons for the increase in bulk. The brain did not get big so we could read or write or do arithmetic or chart the seasons—yet human culture, as we know it, depends upon skills of this kind. . . . The universals of language are so different from anything else in nature, and so quirky in their structure, that origin as a side consequence of the brain’s enhanced capacity, rather than a simple advance in continuity from ancestral grunts and gestures, seems indicated. (This argument about language is by no means original with me, though I ally myself fully with it; this line of reasoning follows directly as the evolutionary reading for Noam Chomsky’s theory of universal grammar.) 3
t 1 . Natural selection produced larger, and more neurally dense, human brains. It was a “side consequence” that these brains gave us such remarkable language abilities.
My next rival explanation comes from my own teaching. Several years ago, I was teaching a philosophical psychology course, and as a part of it, I had my students read Pinker and Bloom’s article and the one by Stephen Jay Gould from which I have taken the above quote. In a take-home essay exam, I asked my students to discuss the controversy and take sides on which argument was stronger. One of my students, a philosophy minor who had taken several courses from me and knew all about inference to the best explanation offered a rival explanation of Pinker and Bloom’s evidence, which she argued was better than either t 0 or t 1 . I was so taken with the originality of her argument that I offered to coauthor with her and see if we could get her idea published. We were successful!4
Joyclynn Potter is a committed theist. But she is also a good philosophy student. Her belief is a cornerstone of who she is and how she thinks. She is, however, intellectually curious and far from close-minded. She sympathetically read and understood Pinker and Bloom’s argument and Gould’s rival argument. She rejected both, not because they were both secular naturalist in spirit nor because they both endorsed evolution by natural selection, but because she felt that both had explanatory problems and that traditional theism offered a better account of what we know about language. Here’s how I tried to express Joci’s position.
What is the best explanation of these facts about human language? There is wide consensus that there is something innate and almost certainly biological, but a totally secular evolutionary account is maddeningly difficult to produce. Theists, however, can easily hypothesize that both a uniquely human ability to acquire and use a natural language as well as mental syntax that structures human thought in a quasi-linguistic manner (a language of thought) are the products of an infinitely wise and beneficent creator. 5
t 2 . The uniquely human ability to acquire and use a natural language is a gift from God.
Ideal Agnostics
I want to share with you an idea that I am very taken with these days. It comes from a contemporary philosopher, as it turns out a very candid Christian philosopher, named Peter van Inwagen. He proposes an audience for arguments (at least those that occur in philosophical debates) that is psychologically impossible but is useful to imagine nonetheless.
The audience is composed of what we might call ideal agnostics. That is, they are agnostic as regards the subject-matter of the debate. . . . Each member of the of the audience will have no initial opinion about [the subject of the debate]. . . . My imaginary agnostics . . . would very much like to come to some reasoned opinion [on the debate] . . . indeed to achieve knowledge on that matter if it were possible. . . . They don’t care which position . . . they end up accepting, but they very much want to end up accepting one or the other. 6
Ideal agnostics are absolutely indifferent—intellectually, personally, and in every way that might bias them—about what the best explanation is. But that doesn’t mean they don’t care. They are also passionately committed to figuring out which explanation is the strongest.
I’m no ideal agnostic and neither are you. But I think we are both well served in our discussions and investigations to pretend that we are. Indeed, I am suggesting that any time we evaluate another’s potential evidence, we try as hard as possible to adopt the position of the ideal agnostic, knowing all along that we will fail in certain respects. When we are presenting our own argument, I would also suggest that we pretend our audience is not composed of partisans but rather ideal agnostics.
This whole little subsection might strike you as a tedious distraction. I am belaboring all this because we all carry with us biases that will inevitably affect some of our rank ordering of explanations. That is the position I find myself in with the current argument. I care very deeply about arguments in the philosophy of religion and cognitive science. I have great respect for all the scientists involved in the debate about language, and I also have great respect for Joyclynn Potter and the tradition she represents within natural theology. I have thought and written about these issues for my entire career. Certainly my lifelong skepticism about religion affected my evaluation of the evidence just as Joci’s committed faith affected hers. In the end, we had to agree to disagree, but hopefully, we understood one another’s arguments better, and we were ultimately in a position to share our joint thinking with a larger professional audience.
Rank Ordering the Explanations (for Pinker and Bloom’s Argument)
While reading and grading Joci’s exam and later while collaborating with her, I came to agree even more strongly with Pinker and Bloom. Here’s how I rank order the three competing accounts of what we know about language.
t 0 . The origin of language is explained through the theory of natural selection.
t 1 . Natural selection produced larger, and more neurally dense, human brains. It was a “side consequence” that these brains gave us such remarkable language abilities.
t 2 . The uniquely human ability to acquire and use a natural language is a gift from God.
Chomsky and Gould would undoubtedly invert t 0 and t 1 . Joyclynn was forced in the exam to commit herself on whether t 0 was better than t 1 , and as I remember, she preferred t 0 . But she dramatically disagreed with her teacher and ranked the three hypotheses as follows:
t 2 . The uniquely human ability to acquire and use a natural language is a gift from God.
t 0 . The origin of language is explained through the theory of natural selection.
t 1 . Natural selection produced larger, and more neurally dense, human brains. It was a “side consequence” that these brains gave us such remarkable language abilities.
Disagreements
What in the world do we do about passionate, but reasoned, disagreement? Stephen Jay Gould and Noam Chomsky were two of the most important scientists of the latter twentieth century. Steven Pinker and Paul Bloom are stars of the twenty-first century. Based just on their credentials, it’s impossible to take sides. Joyclynn Potter is no natural scientist, but she’s a very smart and thoughtful woman. What are we to make of the obvious fact that very intelligent and very honorable people disagree about where the evidence points?
Some might argue that all this shows a fatal flaw in the whole inference-to-the-best-explanation approach to evidence. How can I continue to argue—as I already have and intend to even more vigorously in a later chapter—that we are skilled explainers when equally smart and committed people so disagree as to what the best explanation really is? The short answer is that this is simply the nature of evidence. Lots of times, it points in a clear direction, and we can expect something like intersubjective agreement. In these easy cases, which I believe constitute the vast majority of times when we consciously evaluate evidence, inference to the best explanation brings us close to the standard of knowledge we developed in chapter 3. The evidence for the hypothesis that smoking is a causal factor in lung cancer is so strong that we don’t simply say that the evidence points in that direction; we rather say that we now know that smoking causes lung cancer.
We’re probably not at that degree of certainty about what happened at the record hop nor do we yet possess the full story about the origin of natural language. Still, we possess lots of relevant evidence. Inference to the best explanation helps us reach our personal evaluation of the evidence and hopefully helps us understand the reasoning of those who see things differently. None of us—not our greatest scientists, Supreme Court justices, nor just the smart people we interact with regularly—possess the so-called God’s-eye view, which would allow the simple “perception” of the truth. Since we don’t, the best we can do is rely on evidence to help point us in the direction of the truth. And as the history of science or contemporary debates in jurisprudence and cognitive science tell us, we simply have to expect a certain amount of reasoned disagreement.
Don’t Forget about the Final Assessment of the Evidence!
When I assess the evidence for Pinker and Bloom’s hypothesis utilizing the inference-to-the-best-explanation recipe, my rank ordering in step 3 commits me to my final evaluation. For me, t 0 provides the best explanation of the evidence adduced in support of Pinker and Bloom’s hypothesis. Therefore, the evidence they marshal is very strong .
Chomsky and Gould would come to a very different evaluation of the evidence. For them, t 0 fails to provide the best explanation of Pinker and Bloom’s evidence; t 1 provides a better explanation. Therefore, the evidence presented in the article is poor .
Joyclynn Potter would agree with Chomsky and Gould but for a very different reason. She also believes that t 0 fails to provide the best explanation of Pinker and Bloom’s evidence, but she is convinced that t 2 is the superior explanation. Therefore, she would also say that the evidence in their article is poor .
A Magical Encore?
Quite by accident, I discovered a glitch in the iPod software. On a Saturday night last year, my wife and I went to a banquet for the League of Oregon Cities. The entertainment was Pink Martini, a Portland band I like a lot. I had already planned that I was going to ask for two songs when they came back for an encore—“Lilly” and “Que Sera Sera.” As it turned out, they did “Que Sera Sera” as part of their concert, and there was no chance to ask when they did their encore. On Sunday, as we drove back from Portland, I plugged in my iPod to listen to them again. I set the settings to “All” and to “Shuffle Songs.” This meant that my iPod searched through both of their albums, found all thirty-six songs and played them in “random” order. That’s the glitch! The last two songs were “Lilly” and “Que Sera Sera.” The exact encore I had imagined the night before! What are the odds of this? My theory is that these two songs came up last, not randomly, but because of all the Pink Martini songs, I listen to these two the most often. I am thinking of writing to Apple to tell them about the problem.
This crazy philosopher has a theory that there is a glitch in the iPod software. For practice, and to make sure you’ve got the IBE recipe down pat, take a few minutes, and using all four steps in the inference-to-the-best-explanation recipe assess the quality of evidence he has for this theory.
EXERCISES
- 1. What is a rival explanation?
- 2. What are the four steps in the inference-to-the-best-explanation recipe? Why was I tempted to add a fifth step?
- 3. What is the advantage of imagining all argument analysis, or evidence evaluation, as a discussion between ideal agnostics?
- 4. At the end of chapter 4, I ask you to schematize Leslie’s argument for her theory that Johnny had left her and taken up with Judy (p. 35). That was step 1 in the IBE recipe. Now use the other three steps to determine whether Leslie’s evidence is strong, weak, or just so-so.
QUIZ FIVE
On July 20, 2013, an article appeared in the New York Times arguing that female family members can make males more generous. Here’s a link to the article: https://www.nytimes.com/2013/07/21/opinion/sunday/why-men-need-women.html .
Use all the steps in the IBE recipe to assess the quality of evidence for the claim that “the mere presence of female family members—even infants—can be enough to nudge men in the generous direction.”
Notes
1. G. Harman, “The Inference to the Best Explanation,” Philosophical Review 74, no. 1 (1965): 89.
2. Steven Pinker and Paul Bloom, “Natural Language and Natural Selection,” in The Adapted Mind , ed. Jerome H. Barkow, Leda Cosmides, and John Toby (New York: Oxford University Press, 1992), 451.
3. Stephen J. Gould, “Tires to Sandals,” Natural History , April 1989, 8–15, quoted in Daniel C. Dennett, Darwin’s Dangerous Idea (New York: Simon & Schuster, 1995), 390.
4. Jeffery L. Johnson and Joyclynn Potter, “The Argument from Language and the Existence of God,” Journal of Religion 85, no. 1 (January 2005).
5. Johnson and Potter, 84.
6. Peter van Inwagen, The Problem of Evil (New York: Oxford University Press, 2006), 44.