by Dan Sperber
Most importantly, Heisenberg shared a dense correspondence with Wolfgang Pauli. It is by confronting Pauli’s arguments and rising to his challenges that Heisenberg was able to push his ideas to their full potential. Indeed, the first draft of Heisenberg’s uncertainty paper appears in a letter to Pauli. Even geniuses need people to argue with to develop their best ideas.
The exchange of arguments between Heisenberg and Pauli was essential to the formulation of the uncertainty principle, but it was far from an everyday conversation. Instead of short statements briskly flying back and forth, we find long, well-structured arguments separated by days of solitary thinking. From what we can piece together of it, the thought process that leads to scientific breakthroughs doesn’t look like the type of solitary reasoning depicted by psychology experiments, no unexamined arguments piling up endlessly in support of the scientist’s views. There is a myside bias, yes, but it is tempered by a more demanding quality control that weeds out the weakest arguments.
In Chapter 12, we argued that in an everyday conversation, it makes sense for reasoners not to spend much energy anticipating potential counterarguments. Anticipating counterarguments is both difficult and not that useful, since failing to convince one’s interlocutor right away carries little cost. It is improbable, however, that throughout our evolution reasoning was only ever used in these conditions. The stakes were bound to be higher at times, and in some situations counterarguments could be more easily anticipated. So we should expect some flexibility in how much and how well people scrutinize their own arguments.
Some institutions put a premium on well-crafted arguments. Lawyers only get one final plea. Politicians must reduce their arguments to efficient sound bites. Scientists compete for the attention of their peers: only those who make the best arguments have a chance of being heard. To some extent, reasoning rises to these challenges. With training, professionals manage to impose relatively high criteria on their own arguments. In most cases this improvement transparently aims at conviction. Lawyers must persuade a judge or a jury. If there are other people, including the lawyers themselves, whom arguments fail to convince, so be it.
By contrast, scientists—at their best—seem to be striving for the truth, not for the approval of a particular audience. We suggest that in fact their reasoning still looks for arguments intended to convince but that the target audience is large and is particularly demanding.24 If anyone finds a counterargument that convinces the scientific community, the scientist’s argument is shot down. She cannot afford to appeal to a judge’s inclinations or to play on a jury’s ignorance of the law. Her arguments must be aimed at universal validity—in a universe composed of her peers.
At least as important as the incentive to strive for universal assent are the means to achieve it. In Chapter 12, when Hélène was trying to convince Marjorie to dine at Isami, she could hardly have anticipated Marjorie’s counterarguments, since they were based on idiosyncratic pieces of information such as where Marjorie had dined the week before or how much she could afford to spend. By contrast, experts in a given scientific field are likely to share much of the information relevant to settle a disagreement. This makes actual argumentation much more efficient. This also improves the power of individual reasoning. If a scientist, while evaluating her own arguments, finds a counterargument, then others could have discovered it as well; and if she doesn’t find any, it’s a decent indication that others won’t find any either, or at least none that will be recognized as compelling by the members of the relevant community. The more overlap there is between one’s beliefs and those of the relevant audience, the more useful individual ratiocination can be. Sitting at the end of this spectrum, mathematicians from Newton to Perelman have been known to produce remarkable new results in complete isolation.
Of all the scientific communities, mathematicians are those who are most likely to recognize the same facts and to be convinced by the same arguments—they share the same axioms, the same body of already established theorems, and the arguments they are aiming at are proofs. From a formal point of view, a proof is a formal derivation where the conclusion necessarily follows from the premises. From a sociology of science point of view, a proof is an argument that is considered, in a scientific community, as conclusive once and for all. In logic and mathematics, the formal and the sociological notions of proof tend to be coextensive.25 In 1930, Kurt Gödel presented his first incompleteness theorem, destroying the dreams of building a consistent and complete set of axioms for mathematics, shattering the most ambitious project mathematicians had ever devised. It was a truly revolutionary and threatening result. Yet it was promptly accepted by every mathematician who read Gödel’s proof.26
If mathematicians are able to reach a consensus so quickly, if they all agree on the rules of the games, then they should also be able to anticipate each other’s counterarguments very efficiently. More than any other group, they have both the incentive and the means to evaluate their own arguments thoroughly. The community’s many minds are bound to uncover flaws overlooked by a solitary mind, but a lone mathematician still has a chance of achieving great results. Grigory Perelman solved the Poincaré conjecture in a mostly empty, decrepit Russian institute and in his mother’s apartment.27 Andrew Wiles worked for six years in near-total secrecy to prove Fermat’s conjecture.
Dialogue can still bring great benefits to mathematicians—Paul Erdős became one of the twentieth century’s great mathematicians through hundreds of collaborations. But it is not as necessary as in other disciplines where, even in the so-called hard sciences, a researcher can hardly hope to achieve the same degree of exigency toward her arguments as a mathematician, leaving more room for improvements through discussion.
The Social Context of Science Drives Improvement in Solitary Reasoning
Because science offers its practitioners the incentives and the capacity to engage in productive solitary reasoning, the most brilliant scientists seem endowed with a preternatural ability to generate great insights—and none more so than Isaac Newton. The historian Richard Westfall, Newton’s famous biographer, tells us:
I have never, however, met one [man] against whom I was unwilling to measure myself, so that it seemed reasonable to say that I was half as able as the person in question, or a third, or a fourth, but in every case a finite fraction. The end result of my study of Newton has served to convince me that with him there is no measure. He has become for me wholly other, one of the tiny handful of supreme geniuses who have shaped the categories of human intellect, a man not finally reducible to the criteria by which we comprehend our fellow beings.28
Newton’s deification began long before Westfall,29 but there were also some early skeptics, such as Joseph Priestley, who was careful to bring Sir Isaac’s achievements back to human scale: “Could we have entered into the mind of Sir Isaac Newton, and have traced all the steps by which he produced his great works, we might see nothing very extraordinary in the process.”30 Priestley was, among many things, a chemist, and Sir Isaac’s work in that domain seems to justify this irreverent assessment.
Prying in Newton’s notes, one does encounter some surprising passages:
The Dragon kild by Cadmus is the subject of our work, & his teeth are the matter purified.
Democritus (a Grecian Adeptist) said there were certain birds (volatile substances) from whose blood mixt together a certain kind of Serpent ([symbol for mercury]) was generated which being eaten (by digestion) would make a man understand the voice of birds (the nature of volatiles how they may be fixed).
St John the Apostle & Homer were Adeptists.
Sacra Bacchi (vel Dionysiaca) [the rites of Bacchus (or Dionysus)] instituted by Orpheus were of a Chymicall meaning.31
These are hardly unique. Newton wrote hundreds of pages on chemistry and alchemy, some describing experiments, others trying to understand the deep meaning of such allegories. While to many the passages from Newton’s greatest physics work—the Principia—would sound e
qually obscure, they come from different operations of reasoning. The most relevant difference is not that Newton happened to be right in one case and wrong in the other but that the quality of his arguments varies widely. It would take an extreme relativist to argue that the tight mathematical arguments of the Principia are not any sounder than, say, this: “Neptune with his trident leads philosophers into the academic garden. Therefore Neptune is a mineral, watery solvent and the trident is a water ferment like the Caduceus of Mercury, with which Mercury is fermented, namely, two dry Doves with dry ferrous copper.”32 Yet both are the product of the same undeniably brilliant mind.
An obvious difference between Newton’s reasoning about astronomy and about alchemy is the quality of the data he had access to. On the one hand Tycho Brahe’s precise recording of stellar and planetary positions. On the other hand a mix of hermeneutical texts, vague rumors about people able to transmute metal, and bogus recipes. But there is another difference.
When reasoning about astronomy, Newton knew he would have to convince the most brilliant minds of his time, and he could try to anticipate many of their counterarguments. Even when his academic colleagues weren’t there to talk with him, they were influencing the way he thought. This preoccupation is reflected in Newton’s publication choices. When he published his revolutionary ideas, Newton made sure his colleagues would be convinced, even if that meant not reaching a broader audience. While the first version of the Principia was written “in a popular method, that it might be read by many,” Newton then realized that “such as had not sufficiently entered into the principles could not easily discern the strength of the consequences, nor lay aside the prejudices to which they had been many years accustomed” and so, “to prevent the disputes which might be raised upon such accounts,” he “chose to reduce the substance of this book into the form of Propositions (in the mathematical way).”33
By contrast, in his alchemical pursuits, Newton lacked serious interlocutors: at the time there were only a “few ‘chemical philosophers’ besides J. B. van Helmont, long dead, Robert Boyle, and Isaac Newton.”34 Moreover, the whole topic was shrouded in secrecy. Newton “intend[ed] not to publish anything”35 on this subject, and he complained about Boyle keeping his recipes to himself.36 At Boyle’s death, Newton asked John Locke, who was one of Boyle’s executors, to share some of Boyle’s previously hidden notes, while forcefully denying that they contained anything valuable.37 Such a social context put no pressure on Newton to produce strong arguments, and provided him with little possibility to anticipate counterarguments anyway.
When reasoning about gravity, Newton had to convince a community of well-informed and skeptical peers. He was forced to develop better arguments. When reasoning about alchemy, there were no such checks. The same brilliant mind reasoning on its own went nowhere.
Conclusion: In Praise of Reason after All
Philosophers have depicted reason as a superior power of the human mind. Experimental psychologists have suggested that this superpower is, alas, badly flawed. From an evolutionary point of view, the idea of a flawed superpower makes little sense. So, we set out to rethink what reason is and what it is for.
What Reason Is (and Isn’t)
Reason is, we argued, one module of inference among many. Inferential modules are specialized: they each have a narrow domain of competence and they use procedures adapted to their narrow domain. This contrasts with the old and still dominant view that all inference is done by means of the same logic (or the same probability calculus, or logic plus probabilities).
But isn’t reason characterized by the fact that it is general? How, then, could it be a specialized inference module? The first part of our answer—let’s not rush—has been to insist that reason is indeed specialized; it draws intuitive inferences just about reasons.
Reason draws inferences about reasons? This may look like a vague truism or a cheap play on words (at least in English and in Romance languages, where a single word of Latin origin refers both to the faculty of reason and to reasons as motives). Yet in the history of philosophy and of psychology, reason and reasons have been studied as two quite distinct topics. So the hypothesis that what reason does is draw inference about reasons, far from being a truism, is a serious challenge to dominant views. But how does it help?
True, humans can reason about any topic whatsoever. How can a mechanism that draws intuitive inferences just about reasons be the mechanism of reason itself? The second part of our answer has been that the reason module produces not only intuitive conclusions about reasons—and indeed only about reasons. In doing so, it also indirectly produces reflective conclusions about the things reasons are themselves about. Since reasons may be about anything—rabbits, rain, boats, people, law, or numbers—reason may, indirectly, produce reflective conclusions about all kinds of topics.
So, for instance, if there are dark clouds in the sky and you want to go out, you might intuit, “It might rain. This is a strong reason to take my umbrella.” When, on the basis of that intuition, you decide to take your umbrella, your decision is an indirect, reflective conclusion of your intuitive inference. Your intuition was about the strength of a reason; your reflective decision is about taking your umbrella.
Instead of assuming that intuition and reasoning must be produced by two quite different types of mechanisms—an old idea currently refurbished in dual process theories—we show how reasoning itself can be achieved by an intuitive inference mechanism. The mechanism is highly specialized but it indirectly contributes to our ideas in all domains: it has what we called “virtual domain-generality.”
Compare: the mechanism of visual perception is highly specialized. It processes patterns of stimulation on the retina caused by photons and draws unconscious inferences on the things in the environment that may have emitted or reflected these photons. Not in spite of its high specialization but thanks to it, vision indirectly contributes to most of our thoughts and decisions. Vision and reason are specialized in very different ways, but they both exemplify virtual domain-generality.
The procedures of vision exploit regularities in the way objects reflect light in a normal environment, such as the fact that light generally come from above, and more specific regularities concerning types of objects, such as the fact that faces are generally seen upright. Similarly, the procedures of reason exploit properties of reasons in general, for example, relevance, clarity, or strength; they also exploit properties of specific types of reasons, for example, the force of precedent in reasons concerning coordination, from parent-children relationships to legal matters.
We show, in other terms, how reason fits among other modules of intuitive inference rather than being a towering superpower. Notwithstanding its virtual domain generality, reason is not a broad-use adaptation that would be advantageous to all kinds of animal species. Reasons, we argued, are for social consumption. Reason is an adaptation to the hypersocial niche humans have built for themselves. First part of the enigma of reason solved.
What Reason Is (and Isn’t) For
We have been working together on reason for more than ten years. While our account of the mechanisms of reason is developed for the first time in this book, we have been presenting our earlier work on the function of reason—the “argumentative theory of reasoning”—in a number of publications and conferences.
Most of the philosophers and psychologists we talked to endorse some version of the dominant intellectualist view: they see reason as a means to improve individual cognition and arrive on one’s own at better beliefs and decisions. Reason, they take for granted, should be objective and demanding. Still, when we present evidence that, on the contrary, reason is hopelessly biased and lazy, they accept it without a hitch. Indeed, many of them are familiar with this evidence—some have even contributed to collecting it.
Reading this book, you might have felt somewhat the same way. When you discovered Bertillon’s system or read about Pauling’s obsession with vitamin C, you might have been
more entertained than surprised. Psychology books make their hay showcasing human biases—as we did in Chapters 11 to 14. This has become part of pop scientific culture. In a way, we all know how biased and limited reason is—well, other people’s reason at least.
We are as good at recognizing biases in others as we are bad at acknowledging our own.1 Perhaps this explains why many people can both hold onto an intellectualist position (for themselves and some kindred spirits) and firmly believe that reason is biased and lazy (particularly in individuals who disagree with them).
Actually, the usual defenses of the intellectualist approach to reason are themselves good examples of biased and lazy reasoning. It is an undisputed fact that individual reasoning is rarely if ever objective and impartial as it should be if the intellectualist approach were right. In discussing what to do with this mismatch between theory and evidence, the possibility that the approach itself might be mistaken is rarely considered. Failures of reasoning are lazily explained by various interfering factors and by weaknesses of reason itself. Again, this doesn’t make much evolutionary sense. A genuine adaptation is adaptive; a genuine function functions.
In our interactionist account, reason’s bias and laziness aren’t flaws; they are features that help reason fulfill its function. People are biased to find reasons that support their point of view because this is how they can justify their actions and convince others to share their beliefs. You cannot justify yourself by presenting reasons that undermine your justification. You cannot convince others to change their minds by giving them arguments for the view you want them to abandon or against the view you want them to adopt. And if people reason lazily, it is because, in typical interactions, this is the most efficient way to proceed. Instead of laboring hard to anticipate counterarguments, it is generally more efficient to wait for your interlocutors to provide them (if they ever do).