The Enigma of Reason: A New Theory of Human Understanding

by Dan Sperber

  What, anyhow, is the explanatory import of the whole scheme? There are many more than two mechanisms involved in human inference. The key to explaining human inference is, we will argue, to properly identify the common properties of all these mechanisms, the specific features of each one of them, and their articulation, rather than classifying them into two loose categories on thin theoretical ground.

  Still, for a while, dual process theory seemed to help resolve what we have called the enigma of reason by explaining why reasoning so often fails to perform its function. True reasoning (type 2 processes), the theory claimed, is indeed “logical,” but it is quite costly in terms of cognitive resources. If people’s judgments are not systematically rational, it is because they are commonly based on cheaper type 1 processes. Type 1 processes are heuristic shortcuts that, in most ordinary circumstances, do lead to the right judgment. In nonstandard situations, however, they produce biased and mistaken answers. All the same, using type 1 processes makes sense: the lack of high reliability is a price rationally paid for day-to-day speed and ease of inference. Moreover, type 2 reasoning remains available to double-check the output of type 1 intuitions. Intellectual alertness—intelligence, if you prefer—goes together with a greater readiness to let type 2 reasoning take over when needed. Enigma resolved? Not really.

  The more dual process approaches were being developed, the more they inspired experimental research, the less this simple and happy picture could be maintained. Evans and Stanovich now call it a fallacy to interpret dual process theory as committed to seeing type 2 processes as necessarily “better” than type 1 processes. In fact, they acknowledge, type 2 reasoning can itself be a source of biases and even introduce errors where type 1 intuition had produced a correct judgment. We are not quite back to the early approach of Evans and Wason in the 1970s, if only because the picture is now so much richer, but the problems that dual process approaches seemed to solve are just posed in new and somewhat better terms. The enigma of reason still stands.

  We won’t discuss dual process theory in any detail: it is too much of a scattered, moving, and in part blurry target.17 Our ambition, anyhow, is to offer something clearly better. More relevant to us than the varieties of dual process theories is the way the whole approach has shaken and in some sense shattered psychology of reasoning. For decades, the central question of the field had been: What is the mechanism by means of which humans reason? “Mental logic!” argued some psychologists; “mental models!” argued others. Some still see this as the central question and have offered novel answers, drawing on new ideas in logic or in probabilities. But with the dual process approach, doubt has been sown.

  First there was the idea that there are not one but two types of processes at work. Then several dual system theorists came to see type 1 processes as carried out by a variety of different specialized mechanisms. More recently, even the homogeneity of type 2 processes has been questioned. The more it is recognized that human inference involves a variety of mechanisms at several levels, the less adequate become the labels “dual process” and “dual system theory.” Reason and logic have split, and reason itself now seems to be broken into pieces. This is both a good end point for one kind of research and a good start for another.

  II

  * * *

  UNDERSTANDING INFERENCE

  The elephant trunk is a type of nose. However impressive it may be, it would not make sense to think of it as the epitome of noses. Similarly, reason is one type of inference mechanism; it is neither the best nor the model of all others. To understand reason, we must first understand inference in general, its diversity, and its powers. In Chapters 3 through 6, we show how widespread inference is in humans and other animals, and we look at its basic mechanisms and procedures. Just as the elephant trunk does things that other noses don’t do, we explain how a category of higher-order human inferential mechanisms (that include reason) happens to have a uniquely wide reach.

  3

  From Unconscious Inferences to Intuitions

  Animals don’t think, Descartes had maintained. The eighteenth-century Scottish philosopher David Hume disagreed. He took it, on the contrary, as evident that animals think and are capable of drawing inferences in the same way humans do. He wrote, “Animals, therefore, are not guided in these inferences by reasoning: Neither are children: Neither are the generality of mankind, in their ordinary actions and conclusions: Neither are philosophers themselves, who, in all the active parts of life, are, in the main, the same with the vulgar.” And he added: “Nature must have provided some other principle, of more ready, and more general use and application; nor can an operation of such immense consequence in life, as that of inferring effects from causes, be trusted to the uncertain process of reasoning and argumentation.”1

  Reasoning and argumentation had been viewed, not just by Descartes but by most philosophers, as the path to greater certainty and, moreover, as the only method for drawing inferences. Hume, unfazed, described reasoning as so unreliable that nature must have provided some other means for performing inferences. But if there are means better than reasoning that, moreover, are “of more ready, and more general use and application,” why, then, bother to reason at all?

  The two words “reasoning” and “inference” are often treated as synonyms. What Hume implied was that reasoning is only one way of performing inferences, and not such a reliable way at that. We agree.

  The Why and the How

  What is reasoning? Everybody has at least some working knowledge of it: it is, after all, something we all do and do consciously. But it is one thing to have a working knowledge of some mechanism, and another to really understand what exactly it does and how it does it. You know how to swallow, but do you know how swallowing works? To understand reasoning (rather than just make use of it), one needs, to begin with, an effective way of telling it apart from other psychological processes. And this is where the difficulty begins.

  In the philosophical and psychological literature, reasoning is commonly defined in two ways, in terms of either its goal or its process. These two definitions, alas, fail to pick out the same phenomenon: the standard characterization of the goal picks out inference in general; the standard characterization of the process picks out reasoning proper.

  Why do we reason? The goal of reasoning, so the story goes, is that of coming to new conclusions not through mere observation or through the testimony of others but by drawing these new conclusions from information already available to us.

  How do we reason? The process of reasoning consists in attending to reasons for adopting new conclusions.

  You hesitate, say, between spending the evening at home reading a novel and going to the cinema. You might, at some point, just find yourself reading the novel without having deliberated about what to do. Or you might think, “There is no very good film playing tonight and the weather is bad; I might have to walk back in the rain. On the other hand, there is this novel that Tomoko gave me and that looks really good …” If such was your train of thought, then your decision to stay at home involved reasoning (leaving open the question of whether you used reasoning to actually make the decision or just to rationalize it).

  When we reason, conclusions do not just pop up in our mind as self-evident; we arrive at them by considering reasons to accept them. Or, if we already accept a given conclusion, we might still engage in reasoning in order to find reasons that justify our conclusion, and that should convince others to accept it too. This is quite a rudimentary sketch of the process of reasoning. Unlike most approaches, it does not even mention the role of logic in identifying reasons (which many would claim is essential). Because it eschews the logical framework, it speaks not of premises but, more broadly, of reasons. Still, it already raises a simple question: Is this process—attending to reasons—the only way to pursue the goal of extracting new information from information that we already possess? And Hume answers with force: Of course not! After all, even animals form expectations ab
out the future. Their life depends on these expectations being on the whole correct. Since the future cannot be perceived, it is through inference that animals must form expectations. It is quite implausible, however, that, in so doing, animals attend to reasons.

  Are we humans so different from other animals? Like them, we cannot perceive the future but only infer it. Like them, we base much of our everyday behavior on expectations that we arrive at unreflectively. You play tennis, for example, and have to adjust your position to return the ball; you infer the best position without reflection, in a fraction of a second. Or you call your father on the phone and, just from his tone of voice, you infer that he is in a bad mood. You refrain from telling him that you won’t be able to come to the next family reunion. You don’t have to reflect in order to understand your father’s mood, to expect that he might be agitated if told that you won’t come, and to avoid mentioning the issue: all this seems immediately obvious in the situation and you act accordingly. Just like other animals, humans are capable of forming expectations and drawing various kinds of inferences in a spontaneous and unreflective way.

  Following Hume’s example, we will use the term “inference” for the extraction of new information from information already available, whatever the process.2 We will reserve the term “reasoning” for the particular process of pursuing this goal by attending to reasons. Humans, we will argue, cannot spend a minute of their waking life without making inferences. On the other hand, they can spend hours or even days without ever engaging in reasoning.

  As long as attending to reasons was taken—or rather mistaken—for the only way of extracting new information from information already available, there was no need or incentive to distinguish “inference” from “reasoning.” It is no surprise, then, that the two terms should have been, for so long, used as synonyms.

  Even today, many philosophers, disagreeing with Hume, want to keep a narrow notion of inference as meaning more or less the same thing as reasoning in the traditional sense. This leaves them with two words, “inference” and “reasoning,” for the same thing and no word for the many forms of inference in which reasons play no causal role. Apart from conservatism, there is no clear motivation for such an unhelpful terminological policy.

  In some contexts, mind you, treating “inference” and “reasoning” as synonyms is innocuous and may even serve a purpose. Psychologists, for instance, often use the word “reasoning” to describe the remarkable inferences that animals and infants turn out to be capable of making. Unlike philosophers who insist on narrowing down the meaning of “inference” to that of “reasoning,” what these comparative and developmental psychologists are doing is broadening the use of “reasoning” to cover all kinds of inference. This use of the superior-sounding word “reasoning” expresses well-deserved respect for the intellectual capacities of creatures—animals and infants—who had so often been described as quite dumb.

  Here, however, we want to explore the very special place of reasoning among other forms of inference. For this, we had better sharply distinguish the two. When one looks for inference not just in reasoning but wherever it might occur, one sees it occurring everywhere, in lowly animals, in infants, and in human adults even when they are not reasoning at all.

  Ants in the Desert

  Charles Darwin read Hume’s remarks on animal inference and commented in his diary that one should consider “the origin of reason as gradually developed,” the first hint ever that reason should be approached in an evolutionary perspective.3 It took more than a century for the evolution of human reason to become a serious topic of study, in particular in the new field of evolutionary psychology. Well before this, however, Darwin’s ideas inspired the study of animal psychology. In the very process of investigating what distinguishes animal species from one another, “comparative psychology” (as this new discipline came to be called) has highlighted what all cognitive systems have in common—and, to begin with, the fact that they all perform inferences.

  Take desert ants.4 The University of Zurich biologist Rüdiger Wehner has devoted more than thirty years to their study. He explains:

  The salt pans of the Sahara Desert—vast expanses of flat, hot, and dry terrain—are inhabited by very few animals. Cataglyphis fortis, a skillful and vivacious species of ant, is certainly the most remarkable of these species. It dashes, leaps, and scrambles across the desert surface, and sweeps it for widely scattered food particles, mostly carcasses of other insects that have succumbed to the stress of the harsh desert environment. In searching, the ant leaves its underground colony and scavenges across the desert floor for distances of more than 200m, winding its way in a tortuous search for fodder. Once it has found its bit of food, the ant grasps it, turns around, and runs more or less directly back to the starting point of its foraging excursion, a tiny hole that leads to its subterranean nest [see Figure 7].5

  How do these insects manage, on their way back, to orient themselves in precisely the right direction and to stop speeding forward when they are in the vicinity of their nest? Wehner and his collaborators have, through many careful experiments and observations, shown that ants have in their “navigational toolkit” a “celestial compass” that allows them to assess their changes of direction and an “odometer” that keeps track of the distance covered between two such changes of direction. Needless to say, both tasks involve much more than mere recording of sensory information.

  The celestial compass uses the ants’ sensitivity to the polarization of the sun’s light to determine an axis and to record the angle of each segment of the ant’s outbound run relative to this axis. The odometer infers the distance covered between two changes of direction on the basis of the number of steps used to cover it. Ants’ brains then put together the information inferred by their compass and odometer to further infer the direction and the distance to their nest. This process of “path integration” is comparable to the technique of dead reckoning used by sailors and aviators in order to compute the position of a ship or an airplane in the absence of landmarks (and, nowadays, of GPS input). While human dead reckoning is a complex intellectual task done with the help of measuring instruments, ants’ path integration is achieved within their minute brain by means of these automatic and unconscious computations.

  Figure 7. The trajectory of a desert ant from its nest (N) to food (F) and back.

  For all its specificity, the desert ants’ case well illustrates three of the basic properties of all cognitive systems:

  Cognition is first and foremost a means for organisms that can move around to react appropriately to risks and opportunities presented by their environment. Cognition didn’t evolve in plants, which stay put, but in animals that are capable of locomotion. Cognition without locomotion would be wasteful. Locomotion without cognition would be fatal. Desert ants in particular moving out of their nest would, without their cognitive skills, quickly fry in the sun.

  Cognition involves going well beyond the information available to the senses. All that sensory organs get by way of information, be it in ants or in humans, are changes of energy at thousands or millions of nerve endings. To integrate this information, to identify the events in the environment that have caused these sensory stimulations, to respond in an appropriate manner to these events, cognition must, to a large extent, consist in drawing inferences about the way things are, about what to expect, and about what to do. Foraging ants draw inferences every second.

  Inferences may be performed by specialized mechanisms that are each remarkably good at dealing with just one quite specific task: inferring distance on the basis of number of steps, inferring angular changes of direction on the basis of the position of the sun in the sky, inferring a best path back to the nest on the basis of distance and direction, and so on.

  Ptolemy on a Boat

  Do humans have specialized mechanisms that each deal with one kind of cognitive task? Well, of course they do—to begin with, in perception. That perception is performed by specia
lized mechanisms—vision, audition, olfaction—is a truism. What is less immediately obvious, however, is that these mechanisms perform inferences. We experience ordinary perception as a mere registration of the way things are, as a “taking in” of facts rather than as a construction of mental representations. If there are inferences involved in perception, they are typically unconscious.

  In cognitive psychology textbooks, the scholar usually credited with having discovered unconscious inference in perception is the nineteenth-century German scientist Herman von Helmholtz. The discovery, however, is much older. It is the Greco-Roman scientist Ptolemy, who in the second century CE first talked about the role of unconscious inference in vision. He was also the first to use perceptual illusions as providing crucial evidence on the workings of the mind.6

  You may have had the experience of sitting on a train that actually had imperceptibly started moving, giving you for a brief instant the impression that it was not your train but rather the train on the other side of the platform that was in motion. There were no trains at the time of Ptolemy, but he described a similar illusion on a boat. Suppose, he wrote, “we sail in a boat along the shore … and we do not sense the motion of the [boat] carrying us, then we judge the trees and topographical features of the shoreline to be moving. The illusion stems from the fact that … we infer that the visible objects are moving” (emphasis added).