by Dan Sperber
Unless our reputation is at stake, we are unlikely to seriously examine our own first-order intuitions in the light of our metarepresentational intuitions about reasons. Even if we do, the reasons that come easily to our mind are likely to confirm or even strengthen our initial intuitions. And even if there is a mismatch between our first-order intuitions and our second-order intuitions about reasons, we may not automatically trust the latter more than the former. Victims of scams like Jeb may have some higher-order doubts but fall into the trap all the same: first-order intuitions are too strong. So the fact that our first- and second-order intuitions might not match doesn’t, by itself, make us wise.
We began this book with a double enigma, the second part of which was: How come humans are not better at reasoning, not able to come, through reasoning, to nearly universal agreement among themselves? It looks like now we might have overexplained why different people’s reasons should fail to converge on the same conclusion and ended up with the opposite problem: If the reason module is geared to the retrospective use of reasons for justification, how can it be used prospectively to reason? How come humans are capable of reasoning at all, and, at times, quite well?
9
Reasoning: Intuition and Reflection
Reasons can justify an opinion already formed or a decision already made—this is the retrospective use of reasons. But what if you have a question in mind that you don’t know how to answer, a decision to make but you are not sure which? Isn’t this where a prospective use of reasons—reasoning proper—should help? In principle, yes. If your intuitions on the issue are not clear or strong enough to sway you one way or another, nothing could be better, it seems, than to think in an impartial way of reasons showing which answer is right or which decision is best. Is this, however, what the reason module helps you do?
Intuitive Arguments, Reflective Conclusions
The intuitions the reason module provides are not, we have stressed, about facts that could be a reason for some unspecified conclusion; their form is not “P is a reason” (for example, “That Amy has a fever is a reason”). These intuitions are about facts taken together with the conclusion that they support; their form is “P is a reason for Q” (for example, “That Amy has a fever is a reason to call the doctor”).
In both justification and reasoning, the output of the reason module is a higher-order conclusion that there are reasons for a lower-order conclusion. In the case of justification, the lower-order conclusion has already been produced and possibly been acted upon; say, Amy’s parents have already called the doctor. What is now added is the higher-order conclusion that Amy having a fever justified this decision. This retrospective justification may work even if Amy’s parents called the doctor just because she was feeling dizzy, before they actually took her temperature.
In reasoning, the reason module produces not one but two new conclusions, the second conclusion embedded in the first. The first conclusion is the higher-order argument itself, that is, the metarepresentational intuition that such-and-such reasons support a particular conclusion: Amy’s fever is a good reason to call the doctor now. The second new conclusion is the conclusion—Let’s call the doctor!—embedded in the overall argument and supported by it.
If you catch a fish that has just swallowed another fish whole, you catch two fish at once. Similarly, if you infer that some new conclusion is supported by good reasons, you infer two new conclusions at once: the whole argument and the conclusion the argument supports. When the embedded conclusion is relevant on its own, as is generally the case in reasoning, you may disembed it and assert it, store it in memory, or use it as a premise in further reasoning.
Two illustrations:
You are quite confident that you have left a book you need in your study, but however hard you look, you fail to find it. You try to think where else it might be, and you think, yes, it might also be in the living room. At this point, you reach a conclusion in the form of an argument:
Argument: Since I’m pretty sure the book is either in the study or in the living room and I cannot find it in the study, I should look for it in the living room.
This is the kind of argument one typically accepts without considering higher reasons for it. Given that an intuition, simply defined, is a conclusion accepted without attention to, or even awareness of, reasons that support it, your argument as a whole is definitely an intuitive conclusion, an intuition. This intuitive conclusion, however, is about reasons and about the support these reasons give to a second conclusion, which is embedded in the argument:
Embedded conclusion: I should look for the book in the living room.
Is the conclusion that you should look for the book in the living room also an intuition? No, because it doesn’t fit the definition of an intuition. It is, after all, supported by reasons, reasons that you very much have in mind. In fact, it occurs embedded in a representation of these very reasons.
If the conclusion embedded in an argument is not an intuitive conclusion, what kind of conclusion is it? It is a reasoned conclusion, or, to use a term common in the literature on reasoning and that we have often used ourselves,1 a reflective conclusion, a conclusion accepted because of higher-order thinking (or “reflection”) about it. As this example illustrates, the amount of reflection that goes into reaching a reflective conclusion may be minimal. Still, there is a difference between intuitively believing, without any reasoning involved, that you should look for the book in the living room and having intuitive reasons for this conclusion.2
Here is a second example of a reflective conclusion, this time involving a little more reflection. You are sitting with a friend in a café. She challenges you: “I offer you two bets. Accept one of the two bets, and if you win it, I’ll pay for the drinks.” The two bets she offers are the following:
Bet A: You win if at least three of the next five people to enter the café are males.
Bet B: You win if at least three of the next six people to enter the café are males.
Without having to consider higher-order reasons, you conceive of an intuitive argument that helps you choose one bet.
Argument: The chances of winning bet B are greater than the chances of winning bet A, making bet B the better choice.
Embedded conclusion: Bet B is the better choice.
Your conclusion that bet B is preferable is supported by your intuition about your relative chances of winning the two bets. This intuition provides a reason for what is, therefore, a reflective conclusion.
What these two examples show is that a reflective conclusion need not be the output of a mechanism of reflective inference that could be contrasted to mechanisms of intuitive inference. A reflective conclusion may be an indirect output of a process of intuitive inference. The direct output of this process is an intuitive conclusion about reasons R1, R2, …, Rn for some conclusion C—in other terms, an argument; the indirect output is the reflective conclusion C. This is represented in Table 1.
Table 1 Outputs of the reason module
Types of conclusion Form of conclusion
Direct output Intuitive argument Reasons R1, R2, …, Rn support conclusion C
Indirect output Reflective conclusion C
We want to make an even stronger claim: not just some but all reflective conclusions in human thinking are indirect outputs of a mechanism of intuitive inference about reasons.
In the basic and most common cases (for instance, in the misplaced book or the choice of bet illustrations we just gave), a reflective conclusion is embedded in an intuitive argument. In slightly more complex cases, there may be not one but two levels of embedding, with an intuitive argument supporting a reflective argument that itself supports a final reflective conclusion.
Take the bet example. We described you as having the intuition that the chances of winning bet B are greater than the chances of winning bet A. This intuition would give you a strong argument for choosing bet B. You might, however, be more reflective and accept this argument in f
avor of bet B only on the basis of a higher-level intuitive argument:
Intuitive argument: If only two of the next five people to enter the café are males, you would lose bet A, but you might still win bet B. All it would take is that the sixth person to enter the café be male. Hence, the chances of winning bet B are greater than the chances of winning bet A, making bet B the better choice.
Reflective argument: The chances of winning bet B are greater than the chances of winning bet A, making bet B the better choice.
Reflective conclusion: Bet B is the better choice.
In this example, the higher-level argument (in which the reflective argument and the final conclusion are embedded) is, for most people, likely to be intuitive. For someone who doesn’t find this more general argument intuitive or is unwilling to accept it without reflection, it could itself be embedded in an even more general argument about probabilities that would support it. There may be more than two levels of embedding. Still, ultimately, any reflective conclusion is the indirect output of a process of intuitive inference.
It would be a mistake to assume that higher-order explicit arguments make conclusions easier to understand and accept. To take an extreme example, for most people, it is a plain, incontrovertible fact that 1 + 1 = 2. Alfred North Whitehead and Bertrand Russell famously devoted hundreds of pages of their landmark book, Principia Mathematica,3 to deriving this conclusion through a long chain of complex arguments. Very few people are able to follow these arguments. They were not aimed, anyhow, at strengthening the rock-solid common intuition that 1 + 1 = 2 but at demonstrating that it is provable and, in so doing, at providing logical foundations for mathematics.
What proportion of the many conclusions humans come to in their life are reflective rather than intuitive? What proportion of these reflective conclusions are themselves embedded in reflective rather than intuitive arguments? How common are higher and higher-order levels of reflection? These are empirical questions that have not been properly studied.
Note that the people who would be most likely to express with the confidence of experts their opinion on the importance of reflection in human inference are logicians, philosophers, and psychologists of reasoning. These experts are hardly typical of the population at large; they commonly resort to higher-level arguments as part of their trade. They might be prone, therefore, to mistake their own professional twist of mind for a basic human trait. We surmise that most human reasoning, even excellent reasoning of the kind that is supposed to make us humans so superior, rarely involves more than one or two levels of arguments.
We had suggested in Chapter 8 that the reason module produces, as direct output, intuitions about reasons and, as indirect output, reflective conclusions supported by these reasons. If this is right, then there is no need—and, in fact, no room—for a psychological mechanism (“system 2” or otherwise), the job of which would be to directly produce reflective conclusions. Reasoning can be more or less reflective depending on the degree to which the arguments involved are themselves embedded in higher-order arguments. More reflective, however, does not mean less intuitive. From elementary reflective conclusions directly embedded in an intuitive argument to the many-level reflections characteristic of some of the most impressive scientific achievements, reasoning is always an output of a mechanism of intuitive inference.
In describing reasoning as a use of intuitions about reasons, we adopt a purely psychological approach quite different from more traditional logicist approaches to reasoning. Doesn’t logic play a central role in reasoning? Aren’t we missing something essential?
To better understand the issue, compare the psychology of reasoning to that of mathematics. The science of mathematics itself is about objective mathematical facts. The psychology of mathematics is about the way in which people learn about these facts and use them to calculate, for instance, accrued interests or the surface of a garden. The psychology of mathematics cannot be approached in a purely psychological way; mathematical facts have to be part of the picture. Similarly, it could be argued, reasoning is the use of objective logical facts. Intuitions about reasons that are not about their logical properties are no more part of reasoning than beliefs about, say, lucky numbers are part of mathematical thinking. If this argument is correct, then the psychology of reasoning should have logic at its core.
The acquisition and use of mathematical competence rely on the existence of mathematical symbols for numbers, operations, and so on. Even before the invention of writing, many words in spoken languages served as mathematical symbols and allowed a modicum of mathematical competence to develop. Logic, likewise, relies on symbols to represent propositions, logical relationships, and so on. The use of a whole range of specialized symbols for logic is relatively recent, but just as in the case of mathematics, many words and expressions of ordinary language can serve as logical symbols. Language, so the argument goes, made reasoning possible well before the development of more formal logic.
So, according to the classical approach to reasoning, language is essential to reasoning. We agree, but for completely different reasons.
Reasoning Relies on Language
Reasons, we have argued, are for social consumption. To be socially shared, reasons have to be verbally expressed and, indeed, reasons appear on the mental or public scene in verbal form. Reasons may serve to justify oneself, to evaluate others, or to convince people who think differently. All this involves verbal communication.4 Even when you think of reasons on your own, often it is as if you were mentally answering what others have said or might say, as if you were readying yourself to question others’ opinions and actions. Even when you think of reasons to answer your own questions, it is as if you were engaging in a dialogue with yourself. For this, you resort to inner speech.
Language is uniquely well adapted to represent reasons. To understand a reason is to mentally represent the relationship between at least two representations: the reason itself and the conclusion it supports; in other words, it is a metarepresentational task. Language and metarepresentation are closely associated (even if metarepresentation, and in particular mindreading, may well be possible without language).5 Language is a uniquely efficient tool for articulating complex metarepresentations and for communicating them. Language, in particular, may be uniquely well suited to metarepresent relationships between reasons and conclusions.
Linguistic expressions can be embedded within linguistic expressions, and in particular, sentences can be embedded within sentences: “It is nice” is a sentence, and so is “Yasmina said that it is nice.” Several sentences can be embedded in a more complex sentence to articulate a reason. For instance,
Molly isn’t smiling.
Molly is upset.
are sentences that may represent states of affairs. They can be combined, as in
The fact that Molly isn’t smiling is a reason to believe that she is upset.
This complex sentence metarepresents the relationship between a reason and the conclusion it supports.
To express all kinds of metarepresentational relationships and in particular reason-conclusions relationships, language offers a variety of linguistic devices. Nouns such as “argument,” “reason,” “objection,” and “conclusion” can describe statements and their relationships, and so can verbs such as “argue,” “object,” and “conclude.” A number of so-called discourse markers, such as the connectives “so,” “therefore,” “although,” “but,” “even,” and “however,” have an argumentative function: they focus attention on a reason-conclusion relationship and give some indication of its character.6
Compare, for instance, these two possible answers to the question, “How was the party?”
a. It was a nice party. Pablo had brought his ukulele.
b. It was a nice party but Pablo had brought his ukulele.
Both answers state the same facts, but they put them in a different perspective. In the (a) answer, the information that Pablo had brought his ukul
ele would be understood as an elaboration and a confirmation of the statement that it was a nice party; the music contributed to the success of the party. In the (b) answer, on the contrary, what “but” does is suggest that some of the consequences you might have inferred from the assertion that it was a nice party, such as that everything went well, do not actually follow in this case: yes, it was a nice party, but this was in spite of rather than because of Pablo’s music.
Argumentative devices such as “but” play a heuristic role in argumentation. What they do is facilitate inference and suggest which implications should and shouldn’t be derived in the context. According to the classical view, however, the main role in verbal reasoning belongs to verbally expressed logical symbols—in particular, logical terms such as “or,” “if … then,” “only if,” “and,” and “not” (and other logical devices such as quantifiers and modals). Such logical devices are what make it possible to construct a valid deductive argument. They are the linguistic tools that make reasoning (or at least so-called deductive reasoning) possible. Really? Are the logical devices the protagonists of reasoning, and other linguistic devices mere supporting characters? We want to tell a different story.
