Anyone who has ever taken an aptitude test with an “analytical” section will undoubtedly have encountered little word puzzles such as the following:
The Marriage Problem:
Bill is looking at Nancy, while Nancy is looking at Greg.
Bill is married. Greg is unmarried.
Is a married person looking at an unmarried person?
Answer: A) yes, B) no, C) cannot be determined.1
If you are like most people, you will get this question wrong the first time you try to answer it. The obvious answer is C, “cannot be determined.” This is what our intuitive, rapid problem-solving system tells us. How do we arrive at that conclusion? Through a style of pattern matching. We are looking for a married person looking at an unmarried person. So we take the first couple: Bill is looking at Nancy. Bill is married, but we have no idea whether Nancy is married or not, so there is no match. Now we take Nancy looking at Greg. Greg is unmarried, but again, we have no idea about Nancy, so there is no match. Bill looking at Greg would be a match, but Bill isn’t looking at Greg, so we get no matches. Response: We can’t say. In order to answer the question, we would need to know Nancy’s marital status.
This is how we tackle the problem when we are being less than fully rational.2 There’s a problem, though. Indeed, just walking through the steps, the way I did in the paragraph above, explicitly articulating how the pattern-matching approach works, will make the problem stand out for many people. (Hint: Do we really need to know whether Nancy is married?)
Now consider what a solution to this problem looks like when you use the rational part of your brain. We know that Bill is married and that Greg is not. We do not know whether Nancy is married. Yet there are only two possible states she can be in. The first is that she is married, the second that she is unmarried. Now, suppose that Nancy is married. Is a married person looking at an unmarried person? Yes, Nancy is looking at Greg. Now suppose that Nancy is unmarried. Is a married person looking at an unmarried person? Yes, Bill is looking at Nancy. So it doesn’t matter whether Nancy is married or not. Under either state, a married person is looking at an unmarried person. The correct answer to the question is A (“yes”). This is actually quite obvious, from a rational point of view. But it is also unintuitive, which is why people tend to get it wrong on first pass.
The term reason traditionally refers to a particular mental faculty, one that is associated with a distinctive style of thinking. David Hume famously described reason as a “calm passion,”3 and a degree of detachment and distance from immediate circumstances is a hallmark of the rational style. But perhaps the more significant feature of rational thought is that it can be made fully explicit. To the extent that we are reasoning, we are fully aware of what we are doing and we are able to explain fully what we have done—hence the connection between the faculty of reason and the practice of giving reasons, or argumentation and justification. For any particular claim, we must be able to explain what entitles us to make it and we must be willing to acknowledge what it commits us to.4
This provides the basis for the traditional contrast between reason and intuition. An intuitive judgment is one that you make without being able to explain why you made it. Rational judgments, on the other hand, can always be explained. This doesn’t make intuitive judgments wrong or defective; it just means that they are produced by a different sort of cognitive process. Malcolm Gladwell helped to popularize this distinction in his book Blink, using a number of very striking examples. One involved a forged statue and a group of art historians, many of whom were convinced that the piece was inauthentic but who were hard-pressed to explain why. Something about the statue just felt wrong. According to one of these experts, the first word that came to mind when he saw the (supposedly ancient) statue was “fresh.” Another said that the statue “felt cold,” as though he were seeing it through a pane of glass.5
These judgments were clearly the product of cognition—in fact, they were the product of very sophisticated expert judgment, a system of discernment built up over the course of decades of experience. But they were not rational judgments. Why? Because the experts themselves had no access to the basis of these judgments. They could not explain what exactly it was about the statue that triggered the reaction.
We make this sort of judgment all the time. Look at a photo of a young child, maybe five years old. Is it a boy or a girl? In most cases you can easily tell. Yet how do you form that judgment? What exactly is it about a boy’s face that makes him look like a boy, not a girl? Most of us would be hard-pressed to say. Judgments of age are similar. How do you tell the difference between an eighteen-year-old and a twenty-five-year-old? The judgment is intuitive, not rational. We can go back afterward and try to figure out how we made the decision, but the basis of that decision is not available to consciousness as we are making it. What intuitive judgments provide us with are simply the outputs of a set of cognitive procedures.
Rational judgments, on the other hand, are based on reasons—considerations that the American sociologist Harold Garfinkel described as “detectable, countable, reportable, tell-a-story-about-able, analyzable—in short accountable.”6 With a rational decision, we have conscious access to the inputs and the decision procedure, as well as to the output. If the experts assessing the statue had been able to point to an aspect of the technique, the material, or the style and show that it was anachronistic, this would have provided a basis for rational judgment. Like Sherlock Holmes solving a crime, they would have been able to explain precisely how their process of deduction had unfolded. But they weren’t. They just knew, without being able to say how they knew.
Rational thinking is slow and onerous, which is typically why we try to avoid doing it. The “marriage problem,” however, shows us why it is nevertheless indispensable. One might be tempted to think that that first time we tried to solve the problem we simply made a mistake but that the second time, when we looked at it more carefully, we got it right. This idea—that the mistake is attributable to what psychologists call performance error7—is highly misleading. Our brain, according to this view, is like a cigarette lighter that sparks but sometimes fails to produce a flame, so we shake it a bit and try again. In fact, our brain is more like a bureaucracy or a customer service center, which strives to solve every problem at the lowest possible level. It is only after we have tried and failed to solve a problem using frontline resources that we decide to kick it up to a higher level, and maybe get management involved in the decision making.
In other words, the first time we look at the problem, we typically use a limited set of cognitive resources to produce an answer. Specifically, we try to solve it using a fast, intuitive pattern-matching approach. It is only after finding out that the answer is wrong that we go back and bring additional resources to bear upon it (the mental equivalent of calling a manager). This is when reason becomes engaged. When we decide to concentrate more carefully on the problem, our motive for doing so is not to eliminate a source of error, but rather to facilitate the operation of this new set of cognitive resources. We are, in effect, kicking it up from an intuitive to a rational level. The need to concentrate is just a sign that we are bringing these new resources online. The reason we are forced to do so is that some problems cannot be solved simply by using an intuitive thinking style.
The thought process that leads to the correct solution to the marriage problem has five characteristics that are widely recognized as the hallmarks of rational thinking:
1. Working out the solution requires explicit linguistic articulation. Many people will look at the problem and very quickly come to the intuitive, incorrect conclusion, without any explicit awareness of how they got there. It just seems to them that the answer must be indeterminate. When told that the correct answer is actually “yes,” they may stare at the problem blankly for a while. It is only by explicitly talking their way through it—either to themselves or out loud—that they are able to see the rationale for that answer. It is only the very odd person who is able to come to the correct conclusion without any explicit awareness of how he or she got there. (Consider another problem: “Everyone is prejudiced against prejudiced people. Nancy is prejudiced against Bill. Is Greg prejudiced against Bill?”8 The correct answer is “yes,” but very few people can see why without talking their way through the problem, step by step.)
2. It requires decontextualization. The mistaken response to the marriage problem starts with the tendency to think about it too concretely. We imagine that in order to solve the problem, we need to know some fact about the world: namely, whether Nancy is married or not. We fail to realize that there is a more abstract fact that, alone, is sufficient to resolve the problem: namely, that Nancy is either married or unmarried. Seeing this more abstract fact and its relevance requires an insight into the logical structure of the problem, which in turn requires abstraction from the specific information that we have been given about Bill, Nancy, and Greg. This is a major difference between intuitive and rational thinking. Intuitive thinking relies upon contextual information to solve problems—often supplementing existing knowledge of the problem with additional information that might seem relevant. Rational thinking moves in exactly the opposite direction, stripping the problem of contextual details in order to get at elements of structure.9
3. It makes use of working memory. The chain of reasoning that leads to the correct solution requires us to hold an intermediate conclusion in what psychologists call working memory.10 To see what this is, consider how you go about solving a long multiplication problem in your head. For simplicity, consider 8 times 23. Most people don’t have the answer to this memorized, and so need to work it out. The standard procedure for doing this is to break it down into two smaller problems. First, multiply 8 by 3 to get 24. Then multiply 8 by 20 to get 160. All you have to do now is add up 160 and … what? At this point you have to pull the 24 back out of memory to get the answer of 184. That place where you put the 24, while you were working out the second part of the problem, is your working memory. It’s called that because, first, you are using it as part of an ongoing computation and, second, you will dump it some time shortly after you’re finished. That may seem simple enough, but there is a very widespread consensus among cognitive scientists that rational thought processes depend crucially on this working memory system.11 The pattern-matching system fails to solve the marriage problem because it moves through, making pair-wise comparisons, looking for a “hit” on the pattern it is scanning for. Failing to find one, it concludes that the problem is unsolvable. The rational solution requires first considering what follows from the assumption that Nancy is married, storing that in working memory, then considering what follows from the assumption that she is unmarried, retrieving the first result from working memory, and integrating the two. (The “prejudice” problem is similar. In order to get the correct answer, you need to first figure out the crucial intermediate conclusion, that Bill is prejudiced against Nancy, then use that as a basis for further reasoning.)
4. It is capable of hypothetical reasoning. The intuitive problem-solving system likes to work with facts; it is no good at handling suppositions. The marriage problem forces us to construct a hypothetical scenario (“Suppose Nancy is married”) then figure out what follows from it. Furthermore, we know that the supposition underlying at least one of the two scenarios is false (Nancy is either married or she is not, but she cannot be both). Thus one of the two scenarios is not just hypothetical, but counterfactual. Only reason can deal with these sorts of constructs.12 Intuition is useful when it comes to thinking about and responding to the real world, but whenever we need to think about some possible world (including the negation of the real world), we need to use the rational part of our brain. This means that we need to use reason in order to engage in contingency planning (“What if Plan A doesn’t work?”), strategic thinking (“If I do this, then she may do that”), modal reasoning (“That’s not necessarily true”), and, most importantly, so-called deontic moral reasoning (“He ought to pay her back”).
5. It is hard and slow. Thinking rationally is difficult, which is why most of us try to avoid doing it until absolutely forced. As we shall see later on, there is a good reason for this—you are basically making your brain do something that it was never designed to do. Much of this difficulty is a consequence of the fact that rational processing is slow, relatively speaking, and cognitively demanding, primarily in terms of attention. Maintaining this sort of attention involves inhibiting or controlling all sorts of other thought processes, which in turn involves significant self-control. This is why it is so hard to think clearly when you’re in a rush, or when there is a lot of distracting noise or competing stimuli. (One sure way to irritate other people when they are trying to count is to stand beside them and shout out random numbers. They are bound to lose their place. Yet our own brains are constantly doing the same thing to us: “I’m hungry,” “Look at that bird!” “I’m itchy,” “It’s been a while since you checked your email,” etc.)
The picture that emerges from this is of a mind that is capable of two very different styles of thinking. At one extreme, we have fully reflective, rational thought. This style of cognition is linear, conscious, and explicit, requires attention, and has access to working memory. On the other extreme, we have completely automatic, “modular” systems, which are fast, unconscious, and implicit and can run in parallel (since they require no attention and do not make use of working memory, several tasks can be performed at the same time).
The classic example of mental modularity is our capacity for facial recognition. You see a person, you instantly recognize him as familiar (even if you can’t quite put your finger on who he is or where you met him). The cognition is triggered automatically by the visual stimulus (at no point do you get to decide whether to run your facial-recognition program on people—you can’t stop it even if you want to). You might be hard-pressed to say exactly what it is about him that is familiar, or how you recognized him. Furthermore—as we know from trying to program computers to do facial recognition—an astonishing amount of extremely complicated visual processing must be going on. Not only are faces very complex, but they are seen from different perspectives and angles, not to mention that they change over time (for instance, people’s noses and ears get longer as they age). Yet people are able to recognize each other in all sorts of different circumstances and after years of separation. Thus the processing must be happening very quickly, much faster than anything we can reproduce in consciousness.
The facial-recognition program is also domain specific, which means that it is by and large good for only one thing. For example, while we are very good at recognizing individual faces, we are very bad at recognizing individual trees. A person who spends a lot of time in the woods and worries about getting lost might want to become better at recognizing trees (say, by remembering slight differences in the patterns on their bark). Unfortunately, whatever portion of our brain we use to recognize faces can’t be redeployed to this task (the best we can do is to try to find “faces” in the bark). The domain of this competence is innately specified. This is what leads many cognitive scientists to describe these sorts of intuitive processing systems as part of the “hardware” of the brain.
One other feature of our facial-recognition program is that, because it is triggered and runs automatically, it doesn’t require any attention on our part. It also doesn’t seem to make use of any shared or central resources: there is nothing to stop this program from running at the same time that you are doing other things. When someone you know walks into the room, you recognize that person immediately, regardless of what else you happen to be doing at the moment. It’s not as though you have to wait until you’re finished before turning your attention to the task of deciding whether you know this person. Thus many psychologists have argued that these cognitive competencies are modular, in that they are self-contained and self-sufficient and in that they can run parallel, which is to say, at the same time without interfering with one another. Supporting this contention is the fact that some people suffer from a condition called prosopagnosia, which is a specific impairment in their ability to recognize faces. What is striking about this condition is that pretty much all of these people’s other visual and cognitive competencies are unaffected. This lends support to the description of facial recognition as a module that can be added to or subtracted from the brain without really affecting other systems.
The style of cognition associated with rational thought is quite different. Not only is it slow, domain general, explicit, and demanding in terms of attention, but it has a completely different structure from what is going on at the level of intuition. Reason is what computer types call a serial processing system.13 Unlike a parallel system, which can do many things at the same time, a serial system can do only one thing at a time. This produces an obvious weakness: it gets bottlenecked at the central processor, and so can run very slowly. The advantage, however, is that it can chain together a coherent sequence of operations in a sustained fashion, in order to produce what we refer to (revealingly) as a “train of thought.” Indeed, the kind of argument analysis involved in solving the marriage or the prejudice problem, which involves generating intermediate conclusions then using these as the basis for further inferences, is a perfect demonstration of how serial processing works. The central weakness of parallel processing is that it lacks this sort of focus and coherence. It’s like dividing up a writing assignment between ten people, having each person do one section independently, then pasting it all together. The job may get done faster, but it’s unlikely to hang together as a coherent piece of writing. If the task happens to divide naturally into ten different topics, then the approach may work. But if the goal is to produce an argument, where what gets claimed in one section depends upon what is decided in some previous section, then the job simply can’t be broken down in this way, with the different sections being written simultaneously. They need to be written in order, preferably by one person, even though this arrangement will make things go a lot slower.
The fact that serial processing gets bottlenecked is actually one of the features that psychologists can use to distinguish rational from nonrational processing in the brain. One technique they use is called cognitive loading. Investigators start by giving a person a mental task to complete that involves the use of working memory (say, memorizing a list of five words). They then interrupt that person and give her a second mental task (say, doing some arithmetic or tracking the movement of an object). When that task is finished, the person is instructed to return to the first. Investigators look to see if the interruption has caused any degradation of performance (how many of the five words can the person still remember?). Since working memory space is extremely limited, asking the person to do something that involves reasoning (such as mental arithmetic) will typically result in weak performance. Asking the person to do something that can be handled intuitively, on the other hand (such as visual tracking), has no impact on performance. The processing occurs “parallel” to the rational task that is being performed.14
The view that we have two rather different cognitive styles is referred to as dual process theory, although that suggests a somewhat oversimplified picture of how the mind works.15 Many of the things that we learn explicitly can, through practice and repetition, become “second nature” to us, and so no longer require any attention (think of positioning one’s fingers on the frets of a guitar, or shoulder-checking before changing lanes while driving). And many of the heuristics that we use when engaged in intuitive cognition can be introduced into the rational sphere, in order to provide us with shortcuts that we can take in our reasoning.16 (This is arguably what is going on when we produce the incorrect solution to the marriage problem.) Thus a lot of performances are hybrids, in which we are making use of the two different styles of cognition simultaneously and to varying degrees.
For our purposes, it doesn’t really matter whether we want to call these different processing systems. What is important is the general characterization of reason that dual-process theory suggests—that reason is serial, language dependent, explicit, and that it has access to working memory. Because of this, the style of cognition that it supports is quite different from what goes on elsewhere in the brain.
Sigmund Freud left an indelible mark on psychology by arguing that unconscious, intuitive cognition played a much more important role in everyday life than Enlightenment rationalists had believed. But he made the mistake of characterizing the unconscious as being too much like a human agent. He used “us”—our conscious, rational selves, the part that is available to introspection—as a model for thinking about “it.” Thus he treated the unconscious mind as having complex desires and relatively unified drives. It is no accident that as Freud’s theories slowly drifted into popular psychology, the unconscious id came to be described as an “inner child.” Modern cognitive scientists, on the other hand, are more inclined to describe the unconscious as an “inner Martian.”17 Why? Because the style of thinking that goes on there is extremely foreign to us. As the philosopher Daniel Dennett has observed, we actually find it easier to understand how computers work than we do our own brains.18 This is because computers were designed, from the ground up, to be serial processing machines. The operations of a parallel processing system, on the other hand, are not something that we have introspective access to. We therefore find them confusing to think about.
The extent to which we misunderstand our own thinking was powerfully impressed upon cognitive scientists by early attempts to produce artificial intelligence systems. Computers are extremely good at doing things that people find very hard, like adding up numbers, solving equations, performing tasks in sequence, and remembering things. By the late 1960s, even the most primitive calculators were able to put seasoned engineers to shame. How long could it be, people figured, before computers start doing all of the things that we do, and doing it all much better? The famous HAL 9000 computer, from the movie 2001: A Space Odyssey, was supposed to have been built in 1992. And yet here we are, over two decades later, still using keyboards to communicate with our computers, because computers can’t even transcribe spoken language properly, much less understand it. What went wrong? On the technology side, nothing. It just turned out that much of the stuff we find really easy to do—like recognizing faces, or even objects; learning a language and talking to others; figuring out when someone is joking, or bluffing, or making a mistake; walking without falling over—are actually incredibly hard to do. The only reason we think they’re easy is that our brains take care of these tasks for us, and don’t really keep us posted on what they’re doing or how they’re doing it. (Protein synthesis is also very complicated; luckily, our cells take care of it for us, without requiring any sort of active intervention on our part.) It’s only when we try to reproduce explicitly what our brains are doing implicitly that we realize how little we understand about our own cognitive systems.
The history of chess-playing computer programs provides a marvelous illustration of this. Every so often there will be a new “man versus machine” match, typically presented as though the honor of humanity were at stake and the “rise of the machines” near. What is astonishing, however, is not that the computer sometimes wins, but that any human being is able to beat even a mobile phone at chess.19 Why? Because even though we find chess to be extremely hard—which is why an aura of genius surrounds “grandmaster” players—mathematically it is a very simple game. It actually belongs to the same family of games as tic-tac-toe, because it is finitely solvable.20 This means that there is a single correct way to play it—a “solution,” if you will. The only reason the game is interesting to play is that the solution is too large or complex for anyone to work out (much the way tic-tac-toe is for four-year-olds). It is only when the match moves into the endgame that a portion of the solution becomes apparent. So playing chess requires a lot of hypothetical thinking, of the form “if I do this, then he will do that, then I can do this, which will stop him from doing that,” and so on. This is exactly the sort of reasoning that humans find very hard and computers find very easy. As a result, it was not long before chess-playing computers were developed that could analyze millions of branches of the decision tree in order to pick out an optimal move. And yet humans were still beating them!
This led scientists to study more carefully the thinking habits of human chess players, in order to figure out how they were doing this. What they found, to their surprise, was that grandmaster chess players actually investigate a surprisingly small number of branches of the decision tree.21 The reason they tend to win is that this process of in-depth investigation is preceded by a heuristic pruning of the decision tree, guided by an intuitive sense of what seem to be the most promising moves or of what sort of position they want on the board. Significantly, no one is able to articulate how this initial pruning is done. It is all based on “feel.” It is driven, in other words, by intuition (probably pattern recognition, built up through decades of practice and exposure to games). Thus human chess-playing is a hybrid performance, involving both intuitive and rational thought. To this day, no one has ever succeeded in reproducing the intuitive style of thinking in a computer, simply because we don’t know how it is done (despite the fact that we ourselves do it). Instead, chess-playing computers continue to rely upon brute-force rational methods—trying to search more and more of the solution tree. Deep Blue, for instance, which attracted considerable attention when it first beat Garry Kasparov, was able to examine 200 million positions per second and had access to an enormous database of positions and past games. The current champion, Hydra (described by some as “the chess monster”), is a hybrid software-hardware parallel system running on a huge cluster of workstations in Abu Dhabi. The fact that this much computing power can be deployed without yet achieving the “final, generally accepted, victory over the human”22 is a monument to the power and sophistication of nonrational thought processes in the human mind.
Whenever one sees a practiced expert in action, whether it be a grandmaster playing chess, a doctor sizing up a patient’s symptoms, or a basketball player setting up a play, one is witnessing the seamless integration of intuitive and rational styles of thought.23 Unfortunately, this is very far from the norm. What we encounter more often is a conflict between the two styles of cognition. Indeed, it is precisely because they conflict so often that we are inclined to talk about different systems of thought, rather than just two components of a single system. Call it head versus heart, brain versus gut, or reason versus emotion, we are all familiar with the experience of being “of two minds.”
Here is how one contemporary cognitive scientist, Jonathan Evans, describes the phenomenon:
I define “mind” as a high-level cognitive system capable of representing the external world and acting upon it in order to serve the goals of the organism. The two minds hypothesis is that the human brain contains not one but two parallel systems for doing this. Animals, according to this view, have but one system corresponding to the “old mind” in human beings. Humans have a second “new” mind, which coexists in uneasy coalition with the first, sometimes coming into direct conflict with it.24
This “two minds” phenomenon is actually quite puzzling and, over the years, was responsible for a lot of speculation. It probably explains why most religions posit two distinct realms, the physical and the spiritual. Human beings are typically regarded as the point of intersection of these two realms, so that despite having physical bodies and desires (like animals), we are also thought to have immaterial souls (like angels or spirits) that can survive the destruction of the body. This is then appealed to as an explanation for the qualitative difference between us and other animals. The immortal soul corresponds to the “new mind,” the higher cognitive faculty that no other animals have.
There are plenty of things wrong with this view, but there is one point on which is it right. There is something genuinely puzzling about the “rational part of our soul,” or the “new mind,” precisely because of the discontinuity one sees between human and animal cognitive competencies. To see this, consider what happens when one takes the standard approach and substitutes “evolution through natural selection” for “God” in the explanation of how we came to be rational. The “old mind” has the fingerprints of evolution all over it. Indeed, it has exactly the sort of structure that one would expect to see in a cognitive system that was the product of piecemeal natural selection. The “new mind,” however, is in many ways the opposite. Explaining it in evolutionary terms, therefore, creates something of a puzzle.
The fundamental thing one needs to understand about evolution is that it is a tinkerer, not a designer. It works through a process of small, incremental steps. It’s governed by the same principle as the children’s game hot-warm-cold, in which you hide an object then try to guide a blindfolded child to it by calling out “You’re getting warmer” when she moves closer to it and “You’re getting colder” when she moves away. If you just blindfolded her and tossed her in the room, she would never guess where the object is. But if you provide her with feedback after every step, she can quickly find her way—even though each time she moves, she is still just guessing where to go.25
Evolution is the same. Coming up with a four-chambered heart is highly improbable, just like finding an object while blindfolded. But through a series of small steps, each of which constitutes an improvement over the last, one can construct amazing things. Starting with a two-chambered heart (fish), one can add a third chamber (amphibians), a third with partial separation (reptiles), and finally arrive at a full four chambers (birds and mammals). Each step represents an improvement over the previous step (“You’re getting warmer!”), so evolution is able to produce a very complex structure that would never have been hit upon out of the blue. Of course, even then it takes an almost unimaginably long time for these processes to take place (with the heart, over 500 million years).
Because of this, evolution has something of a signature style when it comes to the design of organisms. First, it is extremely conservative. It doesn’t do radical makeovers or system overhauls. It never goes back to the drawing board. Once it hits on something that works, even if it doesn’t work that well, it sticks with it. This is why, although there seems to be extraordinary variety in the forms of life on earth, when you look more carefully, you can see an astonishing lack of variety. Go to a good natural history museum and you can see skeletons of 50-million-year-old mammals that have almost the same body plan as you have. You don’t even need to know any anatomy—you can just look at each bone in the skeleton then feel for the corresponding bone in your own body. The same is true at a cellular level. The fundamental mechanism that your body uses to store energy, for instance, is shared with not only the rest of the animal kingdom, but with plants as well. Once nature hits on a good trick, it is extremely reluctant to throw it away.
Because of this conservatism, nature is full of designs that are incredibly inefficient. To see why, imagine trying to guide a child through a maze using the standard hot-warm-cold rules—so that you call out “You’re getting warmer” whenever the child moves physically closer to the exit. The chances of succeeding are practically zero. Why? Because the child is almost certain to wander into a dead end. Once she gets there, she will get stuck, since the only moves available to her will be ones that make her “colder.” Indeed, one can imagine a situation where the only way to get out of a dead end would be to take a complex set of backward steps, each of which takes the child farther from the exit. Nature has exactly the same problem. One can easily wind up with arrangements that are only satisfactory, and yet possible improvements cannot be reached, because getting there would first require a series of backward steps.
The mammalian body plan provides a wealth of examples. Obviously, it is “good enough” for most purposes—the fact that it has remained almost unchanged since the Eocene suggests that it is actually quite successful. And yet for us it has some huge problems. Most of these are related to the fact that it is not a good design for bipedal locomotion. In particular, the backbone is not well designed for weight bearing. If you want to support something heavy, it makes sense to have it rest on big, thick, solid long bones—such as one can find in our legs. The way our limbs and back are organized makes sense for a four-legged creature, but no sense at all for a two-legged creature. (Having a single post supporting one’s upper torso is crazy, the biological equivalent of transporting something on a unicycle instead of in a wagon.) Similarly, for an upright stance, it would be much better for our knees to bend backward (the way our ankles do). Unfortunately, we’re very unlikely to develop an extra backbone or reverse the direction of our knees. Developmentally, making these changes would require big alterations, and getting there would require several backward steps. Thus we are stuck with the generic mammalian body plan, which specifies one backbone and knees with the hinges on the front. The best we can do is try to avoid knee injuries and learn to live with back pain.
The conservatism one sees on display in our bodies is present in our brains as well. (Actually, if you think your body is badly designed, wait until you see what the brain is like!) Take, for example, the hiccup. It is an involuntary action consisting of sharp inhalation “followed by a closure of the glottis.”26 It is caused by a distinctive pattern of electrical stimulation originating in the lower brain stem. It is also completely useless. At least in humans it is useless. What is striking, however, is that almost the exact same breathing pattern can be observed in amphibians, and in them it does serve a useful purpose. For tadpoles, this motion is what allows them to switch between lung and gill breathing. Interestingly, several of the tricks that we use to stop hiccups (such as breathing into a paper bag, or taking a deep breath and holding it) have also been shown to inhibit gill breathing in tadpoles (exposure to carbon dioxide, or stretching of the chest wall). Crazy as it may sound, this suggests that when you hiccup, what’s really going on is that an (evolutionarily) old part of your brain is trying to switch you over from lung to gill breathing. The physical machinery needed to carry out this instruction disappeared hundreds of millions of years ago, but there was no need to get rid of the electrical control system. And so there it remains, going off every so often, a glorious artifact of our evolutionary prehistory.
Hearing about the hiccup, one naturally wonders how much other leftover junk there is cluttering our minds. The answer is “a lot.” The basic approach that evolution seems to have taken, when it comes to the structure of the brain, is that rather than getting rid of old stuff, it has simply layered the new stuff over top of it, then given the newer control systems some capacity to override the older ones when needed. As William Hirstein puts it, “It is a truism of neurobiology that throughout our evolution the brain grew outward from the brainstem by adding and enlarging structures to what already existed, rather than replacing or eliminating existing structures.”27 This is why we have a huge number of partially redundant systems in the brain, even for relatively straightforward operations like processing visual input. It also explains why people really do “regress” when they become disinhibited (through alcohol consumption or brain damage) and the “lower” functions acquire more control over behavior.28
Thus the brain has all the signs of having been put together piecemeal, over a very long period of time. One can see evidence of this in a number of different areas of cognitive function, such as memory. Unlike a computer, which has a single, well-organized memory system, all fully indexed and classified, we have at least three different memory systems, localized (sort of) in different parts of the brain. We know this because some people suffer from selective memory impairment—they lose all long-term memory but retain their short-term memory, or they lose all knowledge of facts but can still remember all of the skills they have acquired.
When it comes to problem-solving competencies, our intuitive judgment system seems to have the same piecemeal quality. At one level, our mind seems to be nothing but a grab bag of tricks, put together to solve very specific problems—such as how to tell people apart, in order to separate those you know from those who are strangers. Furthermore, many of our fast, intuitive problem-solving capacities have the same rough-and-ready quality that one sees throughout the natural world. They offer solutions that are good enough for all practical purposes. The reason for this is not hard to find. When it comes to any problem-solving competence, there is always going to be a trade-off between speed and accuracy. Furthermore, each will be subject to diminishing returns. The first 10 percent gain in accuracy can be achieved with minimal loss of speed; the next 10 percent gain, however, requires a slightly greater loss, the next an even greater loss. In an environment where being eaten by a predator is an everyday risk, the value of rapid decision making is likely to begin to outweigh that of increased accuracy long before perfect accuracy is achieved. One should therefore expect evolution to generate problem-solving systems that never produce exactly the right answer except when it is trivially easy to do so.
Reason stands as an exception to this rule, which is why it presents something of a puzzle. But with respect to our intuitive problem-solving capabilities, the trade-off between speed and accuracy has almost always left its mark. This is why, throughout much of the cognitive science literature, the intuitive problem-solving system is often called the heuristic system. Heuristics are basically rules of thumb, or shortcuts, used to solve problems when you can’t be bothered to work out the correct solution. (A classic example in everyday life is the “rule of 72” used to calculate compound interest: “72 divided by the rate of return equals the number of years it takes for an investment to double in value.” This doesn’t give you exactly the right answer, but it’s an answer that is close enough that you almost never have to do the real calculation.)
Our brains are chock-full of these shortcuts and tricks. For instance, we all come equipped at birth with a system of “intuitive physics” that allows us to anticipate the trajectory of moving objects. (Some rather clever experiments have shown that even newborn babies react differently to objects that are coming toward them, depending upon whether the object is on a collision course or not.29) This system of intuitive physics is very powerful, but it is also based on heuristics, which means that it sometimes produces the wrong answer. In particular, it has a huge bug when it comes to anticipating the trajectory of dropped objects. Intuitively, we distinguish between thrown objects and dropped objects. With thrown objects, we take into account both the force that is acting on the object and the effects of gravity. With dropped objects, however, we look only at gravity, and so assume that they will go straight down. As a result, when an object that is already in motion is dropped, we ignore the existing motion and assume that it will go straight down.
It’s easy to see this in action. Put some children in front of a video game that requires anticipation of ballistic trajectories, such as Angry Birds, and they will instantly pick it up. They don’t have to study the game carefully to figure out where things are going to land, they just know. Put them in front of a combat flight simulator, however, where they are required to bomb a target on the ground, and they all get it wrong. They wait until they are directly over the target before releasing the bomb, failing to realize that the bomb is going to continue moving forward at the same initial speed as the aircraft while it descends. (One can see the same mistake in children’s drawings of airplanes dropping bombs, which inevitably show the bomb descending in a straight line.) This is a remarkably persistent error, one that even large numbers of university students studying physics and engineering have been known to make.30
But what else should one expect? Our brains are a product of evolution. There is no reason to expect to find the laws of Newtonian—much less quantum—mechanics inscribed in our subconscious. The environment of evolutionary adaptation didn’t require early human hunters to go on bombing sorties, or even to drop rocks from moving objects. As a result, the expectation that dropped objects will go straight down was good enough for all practical purposes. Exceptions to the rule would have been sufficiently infrequent or inconsequential that there would have been no survival advantage to be had from the development of a more nuanced system of intuitive physics.
The discovery of these cognitive shortcuts in the early 1970s spawned a truly gigantic literature on “heuristics and biases” and their role in human judgment. The most important contributors were Amos Tversky and Daniel Kahneman, who had a peculiar genius for designing scenarios that would cause our cognitive heuristics to misfire. They were able to show that in a vast number of different areas, our brains weren’t really solving the problems posed—or at least weren’t working out a proper solution—but were actually just guessing, using very clever techniques that could be relied upon to produce the right answer most of the time under standard conditions.
This research has received an enormous amount of attention, particularly among people like David Brooks, who are keen to show that human rationality is hopelessly compromised—that we are nothing but the sum of our biases. Yet seldom does anyone stop to ask how we are able to discover our own biases. If our brains are the product of evolution, and if evolution produces all sorts of “good enough” problem-solving capacities, how did we ever acquire the capacity to reflect upon our capacities and discover that they are less than perfect? The only way to detect a broken instrument is by comparing it to one that isn’t broken. But what do we have to compare ourselves to? It’s all well and good to show that our intuitive judgments of probability, for instance, are out of whack when compared to the axioms of probability theory. But where does our capacity to grasp and apply the axioms of probability theory come from?
An unreasonable answer to this question would be simply to say, “Oh well, that evolved too.” There is no doubt that it evolved. But it is very unlikely to have evolved in the same way as our other cognitive abilities. Obviously, when we correct our intuitive judgments of probability or of motion, we are using our capacity for explicit, rational thinking. The problem with reason, from an evolutionary perspective, is that it lacks all the usual characteristics of the “adapted mind.”31 Most importantly, it is domain general, which means that it can be applied to solve problems in any area. And yet in every other area of cognition, and in every other species, evolution has clearly favored cognitive adaptations aimed at solving very specific problems.32 Furthermore, given how much more difficult it would be to evolve a domain-general problem-solving system, compared to a collection of domain-specific ones, there simply is not enough time for such a system to have developed within the human lineage. Evolutionary theorists often talk about how “exquisitely adapted” particular organisms are relative to their environment. And it’s true, there are some exquisite adaptations out there. But these usually take millions of years to develop. Ants, for example, exhibit an astonishing array of complex social behaviors. But ants have also been on this planet, in pretty much their present form, for 140 million years. This means they’ve had time to work out all sorts of fancy adaptations. Modern humans, by contrast, branched off from the rest of the Homo lineage only about 250,000 years ago. The idea that between, say, the dawn of Homo heidelbergensis and rise of Homo sapiens—a period of maybe a half million years—we managed to evolve a completely new type of cognitive system, from scratch, which allows us to produce absolutely precise answers to any sort of question we might care to ask, is completely implausible.33
To see the problem, consider the case of mathematical ability. Human infants appear to come equipped with two innate modules for dealing with numbers.34 The first (called the subitization system) allows us to deal with small numbers and to tell at a glance the difference between sets of one, two, and three objects. So if you show a small child two balls, cover them with a handkerchief, then add a third ball, the child will notice the difference when you pull the handkerchief off. But if you put five balls under the handkerchief then add a sixth, the child won’t notice. The second module allows us to guesstimate with respect to the size of large groups. So a child will be able to see—again, at a glance—that a pile of twenty is more than a pile of thirty balls. This module cannot do fine discriminations, though, so a pile of twelve and a pile of thirteen will be treated the same.
One of the most striking things about these two modules is that they are shared with other primates. Chimpanzees and baboons, in particular, have been shown to possess exactly the same discriminatory capabilities. There is every reason to believe that our competencies in this area were inherited from a common ancestor (and thus constitute part of the “old mind”). At around age four, however, the time when children acquire advanced linguistic competence, they begin to develop new abilities. In particular, they develop mastery of the counting procedure and begin to integrate it with their innate competencies. Chimpanzees, on the other hand, even those who have been taught to sign, seem to be incapable of taking this step. They can be taught the signs for one, two, and three, but then they treat “three” as being equal to “anything greater than two.” Teaching them the number four involves unlearning this understanding of three, but then they simply come to understand “four” as “anything greater than three.” In order to teach them the number five, one must start all over again, unlearning this understanding of “four.” Each new number must be taught through the same arduous process. At no point does the chimpanzee ever “get it,” the way human infants do. They learn numbers the way we learn the names of objects. They can list the names, but at no point do they ever learn the procedure that we call counting.
Some evolutionary psychologists have been inclined to explain this by saying that humans have a third “mathematics module,” which allows us to do all these advanced calculations. This could be true, but consider how implausible it is. In terms of sheer computational power, the two innate modules that we share with other primates are fairly pathetic and, furthermore, have undergone no obvious improvement in the past five million years. (How hard would it be to expand our range a bit, so that instead of being able to count up to three intuitively, we acquired the ability to count to four, or maybe even five?) And yet, at the same time as nothing was going on in either of these two modules, we somehow acquired—from scratch—a brand-new module, which no other animal has, that allows us not only to count to infinity, but also to understand fractions, decimals, vectors, limits, negative numbers, irrational numbers, imaginary numbers, transcendental numbers, and much more? It makes no sense.
The idea that we have one module for counting to three and another module for “all the rest of mathematics,” and that these both came about through the same kind of natural selection, is completely unbelievable. First of all, there can be little doubt that both the subitization and the guesstimation modules evolved in order to do what they do. In other words, they were selected for precisely because of the fitness benefits that came from being able to carry out these sorts of discriminations. With the rest of mathematics, on the other hand, one can be quite certain that the underlying cognitive system was not selected for because it would facilitate the development of calculus, analytic geometry, or linear algebra—since these came about only within the scope of historical time. Even something as simple as counting to one hundred has no obvious fitness benefits (and if it did, why wouldn’t one see modification to the small numbers module?). Thus we can be fairly certain that the ability to do mathematics in the way that we do is a byproduct effect of whatever cognitive adaptations happened to make it possible. In other words, the ability to do mathematics cannot have evolved in order to allow us to do mathematics; the fact that we acquired this ability must be just a happy coincidence.
The same argument can be generalized across other cognitive domains. There is simply no advantage to be had in building a single cognitive system that can be used to figure out the gravitational constant of the universe, build a machine to catch neutrinos, unlock the secrets of life on earth, assess counterfactual probabilities, or think about what to serve for dinner next Thursday.35 It is difficult to imagine what sort of environmental challenge could require the construction of a system that flexible. Furthermore, such a system would simply be too overpowered relative to both the nature and the scale of the challenges facing our ancestors in the environment of evolutionary adaptation. It would be like putting a full-scale AI into a Roomba.
This leads us inevitably to the key conclusion: our capacity to reason must be a byproduct of adaptations that were intended to achieve other goals. In other words, our primate brains aren’t designed for rational thought.36 The “rational part of the soul,” far from being the pinnacle of creation, turns out to be just an accident, a strange little offshoot in an otherwise unpromising lineage.
This helps to explain one thing: why thinking rationally is so hard. Dennett describes the conscious, rational mind as a “serial virtual machine implemented—inefficiently—on the parallel hardware that evolution has provided for us.”37 The key word here is inefficient. The mind simply didn’t evolve to support the sort of linear, explicit processing that is the hallmark of rational thought. Furthermore, there hasn’t been enough time for it to develop the adaptations that would greatly facilitate or enhance this style of thinking. Thus the way that your brain feels after writing an exam is like the way your back feels after a long day spent lifting boxes—neither was designed for the task that it is being asked to perform.
This is of enormous importance when it comes to updating the ideals of the Enlightenment. Reason is not natural; it is profoundly unnatural. At the same time, it is the only thing that allows us to escape from the straitjacket of our animalistic minds. So while it has the potential to free us from the state of nature, there is no reason to expect this process to be easy. The great thinkers of the first Enlightenment tended to believe that once prejudice and superstition were overthrown, reason would naturally take their place, without any sort of slippage or backsliding. We now know that this isn’t true. But more importantly, we know why it isn’t true.
So what is the adaptation that created all of these astonishing byproduct effects? Many cognitive scientists (and, for what it’s worth, philosophers) believe that the answer is language.38 Indeed, an increased appreciation for language and its importance in human cognition has been at the center of the most important conceptual revolution in our understanding of the mind since the beginning of the modern era. Throughout most of history, it was simply assumed that our minds came fully equipped with the ability to formulate complex thoughts all on their own. Beliefs were thought to be like pictures in the mind, images of things we have seen. Language was something that came along later, when we began to feel the need to communicate these ideas to one another. The universally shared assumption was that a human being could be fully rational all on her own, without ever having spoken a word to another person.
A number of developments began to chip away at this consensus over the course of the nineteenth century. Philosophers, in particular, realized that beliefs had a lot more in common with sentences than with pictures.39 (Take, for instance, a belief like “The Battle of Waterloo was fought in 1815.” What would a picture of that look like? How about “The ball is not red”?) Psychologists began to notice that all sorts of advanced cognitive abilities—including counting and arithmetic40—start to show up in children only after they have acquired mastery of a language. And linguists, analyzing the structure of syntax and grammar, began to establish connections between the formal operations we use to construct complex sentences and the operations we use to construct complex thoughts and arguments.
The net consequence of all this was an almost precise inversion of the traditional order of explanation. Rationality had always been taken to precede language. Now it began to look as though language preceded reason, or, more specifically, as though our ability to string together a set of thoughts into a coherent sequence—the centerpiece of serial processing—was a byproduct of our ability to produce a set of sentences in sequence. Language, according to this view, evolved primarily because of its role in facilitating communication. The adaptations needed to get it off the ground were well within the reach of the “old mind.” Yet once a more complex system of communication developed, particularly one with the type of grammatical structure that ours has, it provided the basis for a new style of computation.41 Thus the development of language produced significant cognitive benefits, leading to the type of thinking that we associate with the “new mind.”
For example, one of the most basic uses of language is that it can be used to tell other people what to do. Not only can you tell people what to do, but you can give them a series of instructions in sequence, with the expectation that they will carry them out in the order specified. What is more, this tool can be used not just to control others, but to control the self as well. A set of self-directed imperatives is basically a plan. Mental rehearsal of actions—talking your way through them in advance—has been shown to improve performance. Thus we start by ordering people around and being ordered around, but in the process we acquire a new tool for controlling and planning our own actions.42 As the psychologist Lev Vygotsky wrote, “the most significant moment in the course of intellectual development, which gives birth to the purely human forms of practical and abstract intelligence, occurs when speech and practical activity, two previously completely independent lines of development, converge.”43
Rational thought, according to this view, is essentially a form of inner speech. When you listen to small children talking, they are not merely expressing their thoughts. You are literally hearing them think.44 It is only over time that they develop the capacity to inhibit the physical motions, to compress the speech, and thus to think silently, to themselves. This process of internalization can be observed in other domains as well. In China, one can sometimes see older merchants doing extraordinary complex arithmetic “in their heads,” without the benefit of any external devices. If you look carefully, however, you may see them making slight motions with their fingers. This is because they were trained using an abacus. After using one for long enough, they develop a clear enough picture of the device in their mind and a feel for where the beads would be, so that they no longer need the actual physical device. The slight movement of the fingers is the only remaining trace of the external origins of the competence.
In early twentieth century, Vygotsky and his colleague Alexander Luria were able to detect a similar process of internalization in children learning to write, but this time with speech, rather than with an abacus. Noticing that there was a “constant buzz” of quiet speech in a class of young students learning to write, they suggested that half of the class be allowed to whisper to themselves but the other half be instructed to hold the tip of their tongue between their teeth.45 The second group suffered a significant degradation of performance. On the other hand, simply being asked to hold their teeth together or clench a fist had no such effect. It was suppression of the tongue movements specifically that generated a degradation of performance, because it prevented them from vocalizing, and therefore led to a greater number of mistakes.
All of this suggests that talking is actually the primitive form of thinking—or, more specifically, it is a developmental precursor to the form of explicit, conscious thought that is the hallmark of rationality. By the time we become adults, the process of internalization is so complete that we imagine that we always had the ability to think silently, “in our heads.” There are times, however, when the earlier form breaks through. It is not unusual, for instance, to see someone who is struggling to follow a set of complex instructions stop and read them out loud to himself. Indeed, people often resort to self-directed talk when they are having difficulty exercising self-control, maintaining attention, or even just carrying out a difficult processing task. (Various psychological studies have shown that asking people to explain their thinking out loud as they are completing a task typically improves performance, rather than impairing it.46 This is the opposite of what should occur if talking were a fundamentally different process from thinking.)
While language can be used to enhance self-control, it also allows us to think in a different way.47 One can see this quite clearly when small children, having mastered the counting procedure, are suddenly able to perform feats of mathematical reasoning that are qualitatively different from anything else found in nature. The dominant view is that our grammar enables this, because it permits recursion—the familiar process of embedding one clause within another.48 As anyone who has programmed a computer knows, recursion is hugely important and incredibly powerful. When you say something like “I wish that he wouldn’t do that,” it may seem like just a boring old sentence, but in fact you’re doing something that your “old mind” can’t do. By embedding one sentence (“he wouldn’t do that”) within another (“I wish that …”), you are constructing a recursive function. Embed yet another sentence in it, resulting in, say, “I wish that he wouldn’t do that thing that he does with his finger,” and you suddenly become the most powerful computational system in the natural world. This is why chimpanzees can’t count and we can. They treat numbers as names, associated with groups of objects. We understand them in terms of a function, which can be applied to itself again and again, in order to produce an infinite sequence of numbers. This is what gives us the ability to do “all of mathematics.”49
It is this foundation in language that also explains the two features of rational thought that were identified at the beginning of the chapter as fundamental. First of all, rational thinking is explicit. What this means, in effect, is that the sequence of steps that we go through in order to arrive at some conclusion can be put into words. With intuitions, we can usually state what the final judgment is, but we cannot articulate how we got there. The fact that rational thought is already linguistically formulated explains why we are able to articulate it so precisely and easily. It explains why we are, as the philosopher Michael Dummett put it, able to communicate these thoughts in language “without residue.”50 There is nothing that remains unexpressed when we have finished articulating a rational train of thought. “Thought,” Dummett writes, “differs from other things also said to be objects of the mind, for instance pains or mental images, in not being essentially private. I can tell you what my pain is like, or what I am visualizing, but I cannot transfer to you my pain or my mental image. It is of the essence of thought, however, that it is transferable, that I can convey to you exactly what I am thinking.”51
The second major feature of rational thought is its universality. Reasons that are good for one person are presumptively valid for all persons. It is not difficult to see why, if one keeps in mind that language is essentially public. In order to function as a means of communication, language must work according to a set of rules that are, if not identical, at least broadly similar for all persons. This means, however, that the same rules that govern public speech are also going to govern private rational thought. As a result, arguments that we find compelling are likely be found compelling by others, in a way that private intuitions or feelings may not.
This is why, even though people’s brains are very different, and even though some people have remarkably different abilities in different areas, people all reason in very much the same way. Although the metaphor is slightly misleading, there is something to be said for thinking of human reason as akin to a software application that runs the same way on different hardware platforms (or a website that looks the same in any browser). With men and women, for instance, there are very significant differences in brain chemistry, physiology, and development. And yet clear-cut sex differences in cognition show up only in peripheral, modular systems (such as spatial rotation of objects or the ability to detect slight changes in the environment). For all the hype surrounding differences between the male and female brains, it is actually extraordinary how little difference there is in the way that men and women reason.52
For a long time, the discovery that rational thought depended on language, and not the other way around, was thought to have relativistic consequences. Indeed, a lot of postmodernism, and the associated assault on reason, was nothing but an early overreaction to these findings. When the dust settled it became apparent that, far from the evidence supporting relativism, the opposite was true. There are many different languages in the world, which means that there will be many different flavors of thought. And yet one of the striking features of these languages is that they are all basically intertranslatable. We have yet to encounter a language that is not learnable by someone willing to make the necessary investment of time and energy. Thus there is no tension between the idea that reason depends upon language and the idea that it possesses a universal structure.
The idea that our minds have different “parts,” partially redundant and often conflicting, is one of the oldest and most troubling claims in the history of philosophical reflection on the nature of thought. Modern cognitive science has gone some way toward piecing together the puzzle. According to the “dual process” view, we have two different styles of cognition, distinguished by the following characteristics:
System 1: Intuitive, heuristic
Unconscious, automatic
Rapid, computationally powerful, massively parallel
Associative
Pragmatic (contextualizes problems in the light of prior knowledge and belief)
Does not require the resources of central working memory
Functioning not related to individual differences in general intelligence
Low effort
System 2: Rational, analytic
Linked with language and reflective consciousness
Slow and sequential
Linked to working memory and general intelligence
Capable of abstract and hypothetical thinking
Volitional or controlled—responsive to instructions and stated intentions
High effort53
In order to assess the prospects of a new Enlightenment, we need to understand more clearly the strengths and weaknesses of the unglamorously named “System 2.” The great thinkers of the first Enlightenment tended to dismiss System 1 entirely, treating it either as nonexistent or else as inferior to reason. We now know better. Indeed, the pendulum has swung so far in the other direction that many people think System 2 is an illusion—that what we call “rational thought” is a sham, either hopelessly mired in biases or consisting of post hoc rationalizations. The reality is somewhat more prosaic. Reason is far less powerful and autonomous than it was once taken to be, yet it is still indispensable for certain forms of cognition. Evans sums up the state of play with admirable simplicity: “In terms of quantity, the analytic system can meet very little of the cognitive demands on the brain but in terms of quality it can do things that the heuristic system cannot.”54 The central question then becomes: What can reason do that intuition can’t?