We consider it a good principle to explain the phenomena by the simplest hypothesis possible.
—Claudius Ptolemy
It is futile to do with more things that which can be done with fewer.
—William of Occam
To the same natural effects we must, so far as possible, assign the same causes.
—Isaac Newton
Whenever possible, substitute constructions out of known entities for inferences to unknown entities.
—Bertrand Russell
What counts as knowledge, and what qualifies as an explanation, are two of the main questions discussed in this book. They are also central concerns for philosophers of science. In the answers they give to these questions, philosophers of science perform both the function of describing what it is that scientists are doing and critiquing what they do. Conversely, some philosophers of science use the findings of scientists—and experimental philosophers—to address traditional philosophical questions (although this is a more controversial practice among philosophers than you might guess).
Some of the important issues addressed by philosophers of science include: What constitutes a good theory? How economical or simple should a theory be? Can a scientific theory ever be confirmed, or is “not yet falsified” the best it can do? Can a theory be a good one if there is no way we could falsify it? What’s wrong with special-purpose, “ad hoc” fixes to theory? All of these questions are as relevant to the theories and beliefs we hold about everyday life events as they are to the activities of scientists.
KISS
In graduate school, I had a professor who was prone to generating highly complicated theories—much more complicated than I thought testable or likely to be supported by evidence in a convincing way. He defended himself by saying, “If the universe is pretzel-shaped, you better have pretzel-shaped hypotheses.” My response, prudently uttered only to myself, was, “If you start out with pretzel-shaped hypotheses, the universe better be pretzel-shaped or you’ll never find out what shape it is. Better to start with a straight line and go from there.”
The injunction against complexity has come to be labeled Occam’s razor: theories must be succinct—unnecessary concepts have to be shaved away. In the scientific arena, the simplest theory capable of explaining the evidence wins. We abandon a simple theory only when there’s a more complicated theory that explains more evidence than the simple theory. Simpler theories are also to be preferred because they tend to be easier to test, and, in the more precise sciences, more readily modeled mathematically.
Ptolemy didn’t follow his own advice very well. Figure 7 shows the path of Mars around the earth as specified by Ptolemy, providing epicycle after epicycle in order to match the perceived motion of Mars. An epicycle is a circle on a circle. There was a strong prior assumption in Ptolemy’s day that the universe was constructed on elegant geometric principles, employing especially the circle. If lots of circles were required to model planetary movement, so be it.
Ptolemy’s theory fit the data perfectly. But since no one could come up with laws of motion that were remotely plausible explanations for such a path, it seems puzzling that it took a very long time for people to realize that there was something drastically wrong with the theory.
KISS—Keep It Simple, Stupid—is a good motto for lots of things. Complicated theories and proposals and plans are likely to cause foul-ups. In my experience, people who sacrifice comprehensiveness and complexity for simplicity are more likely to come up with answers—to something at least, if not to the original question.
Figure 7. Ptolemy’s epicycles to explain the movement of Mars around the earth.
Simple theories are to be preferred even when you know they’re inadequate to explain all the available evidence. Testing more complicated theories is more labor-intensive, and more likely to lead the investigator down a garden path.
Early in my career, I studied the eating behavior of the obese. I found that their behavior resembled the behavior of rats with lesions to the ventromedial hypothalamus (VMH). Damage to that area of the brain made the rats act as if they were hungry all the time, and they ate enough to become obese. The analogy proved productive, and I was able to show that the feeding behavior of the obese is highly similar to that of rats with VMH lesions. This strongly suggested that obese people are hungry most of the time. I argued that they are attempting to defend a “set point” for weight that is higher for them than for most people.1 The best evidence for that comes from the fact that the eating behavior of obese people who are not trying to lose weight is the same as that of normal-weight people, whereas the eating behavior of normal-weight people who are trying to lose weight resembles that of obese people who are trying to lose weight.2
Experts in the field of eating behavior and obesity told me that the facts couldn’t be fully explained by the simple hypothesis of defense of a set point for weight. True enough. But most of the people telling me that didn’t really learn much about obesity, whereas people exploring simple hypotheses about obesity have learned a lot.
What’s sensible in science is also likely to be sensible in business and other professions. The KISS principle is explicit policy for some highly successful companies and is recommended by many business consultants.
McKinsey & Company instructs its business consultants to keep hypotheses as simple as possible at first, and to allow complications only as they are forced on them.
People who offer advice to start-up companies insist on keeping it simple at first: Release products quickly to get feedback rather than obsessively create the best possible product; target markets where it’s possible to maximize early profits rather than aim for a broad range of markets; don’t demand complete knowledge of markets or any other aspect of business before acting; keep business models presented to potential investors as simple as possible.
As they say at Google: “Done is better than perfect.”
Overly complicated ways of trying to solve a problem are sometimes called Rube Goldberg machines, Goldberg being the fellow who used to draw hilariously convoluted ways of solving a simple problem. For the all-time most spectacular Rube Goldberg machine, click on www.youtube.com/watch?v=qybUFnY7Y8w.
The part of Occam’s razor enjoining multiple hypotheses doesn’t fully apply to practitioners such as doctors. The more hypotheses the better when we’re trying to decide which explanations are in contention and how they can best be examined. I don’t want my doctor to entertain only the most plausible hypothesis. I want my doctor to pursue every hypothesis with some reasonable chance of being correct, as well as the possibility that two or more hypotheses are necessary to explain my symptoms. Even for medical diagnosis, though, some rules of parsimony apply. Medical schools teach their students to employ simple and inexpensive diagnostic procedures prior to more complicated and expensive ones and to pursue the most likely possibilities first. (“Think horses, not zebras.”)
Reductionism
An issue central to many philosophical and scientific debates concerns reductionism, a principle that at first blush resembles Occam’s razor. A reductionist explanation is one holding that some seemingly complex phenomenon or system is nothing more than the sum of its parts. Reductionist explanations sometimes go further and maintain that the parts themselves are best understood at some level of complexity simpler than, or lower than, the phenomenon or system itself. That position denies the possibility of emergence—in which phenomena come into being that are not explicable solely by invoking processes at a simpler, more basic level. The example par excellence of emergence is consciousness. It has properties that don’t exist at the level of the physical, chemical, and electrical events that underlie it (and, so far at least, are not explainable at that level).
If you really can get away with reductionism in either of the senses above, you justifiably win. But the people who study phenomena at a given level are naturally going to be opponents of people who try to dismiss events as mere epiphenomena—events secondary to the underlying events and lacking true causal significance.
Some scientists believe that macroeconomics (the aggregate behavior and decision making of the economy as a whole) is fully explained by microeconomics (choices that individuals make). Other scientists believe that microeconomics is fully explained by psychology. Still other scientists believe psychological phenomena can be fully explained by physiological processes, or undoubtedly will be at some future time. And so on. Physiological processes are fully explained by cellular biology, which is fully explained by molecular biology, which is fully explained by chemistry, which is fully explained by the quantum theory of the electromagnetic force, which is fully explained by particle physics. Of course no one proposes that degree of reductionism. But at least some scientists have endorsed one or more of the individual reductions in that chain.
Many reductionist efforts are useful. The principle of parsimony requires us to explain phenomena at the simplest level possible and to add complications only as they become necessary. And the effort to explain things at a level one down in the hierarchy can be useful even if the ultimate conclusion is that there are indeed emergent properties that prevent a full explanation in terms of simpler underlying processes.
But one person’s simplification is another person’s simple-mindedness. Scientists from other fields are continually trying to explain phenomena in my field of psychology by asserting that they are “nothing but” the operation of factors at some lower level of complexity.
I’ll describe two examples of reductionism for psychological events that seem to me to be misguided and off base. Full disclosure: recall that I’m a psychologist!
A decade or so ago a new editor of the prestigious Science magazine announced that under his regime the magazine would not accept papers in psychology that failed to show pictures of the brain. This reflected his opinion that psychological phenomena could always be explained at the neural level, or at least that advances in our knowledge of psychological phenomena require at least partial understanding of the brain mechanisms underlying them. Few psychologists, or neuroscientists for that matter, would accept the idea that we’re at a stage where purely psychological explanations of psychological phenomena should be considered useless or inadequate. The editor’s insistence on physiological reductionism was at best premature.
A much more consequential example of what the philosopher Daniel Dennett calls “greedy reductionism” is the policy formulated a decade or so ago by the head of the National Institute of Mental Health (NIMH) of refusing to support basic research in the behavioral sciences.
The NIMH continues to support basic research in neuroscience and genetics, reflecting the director’s highly controversial view that mental illness originates in physiological processes and can be understood primarily or even solely in terms of such processes, rather than as part of a loop among environmental events, mental representations, and biological processes.
Despite the $25 billion that has been spent annually on basic neuroscience research by the National Institutes of Health and the $10 billion spent on basic genetic research, neither type of research has produced new treatments for mental illness. There have been no major advances in the treatment of schizophrenia in fifty years or in the treatment of depression in twenty years.3
In contrast, there are many examples of effective treatments for mental illness resulting from basic research in behavioral science, and many more interventions that improve mental health and life satisfaction for normal individuals who would not be considered mentally ill.
We can start with the fact that the theory behind Alcoholics Anonymous, according to its cofounder, came from adopting William James’s theories of the role of religion in banishing despair and helplessness.
The best diagnostic procedure available for assessing the likelihood that a person hospitalized for a suicide attempt will make another attempt is called the Implicit Association Test.4 This measure was originally devised by social psychologists to assess a person’s tacit, unrecognized attitudes toward various objects, events, and categories of people. A person whose tacit associations regarding the self are closer to concepts related to death than to life is likely to make a second attempt. Neither the person’s self-report, physician’s judgment, or any psychiatric test does as good a job at predicting a second attempt.
The most effective treatment for phobias derives from basic research on animal and human learning.
The best available intervention for psychological trauma, discussed in Chapter 10, derives from basic research in social psychology.
And many other examples could be cited.
Finally, behavioral science has been critical in establishing the ineffectiveness, or actual damaging effects, of mental health interventions invented by non–behavioral scientists.
Know Your Own Strength
We don’t recognize how easy it is to generate hypotheses about the world. If we did, we’d generate fewer of them, or at least hold them more tentatively. We sprout causal theories in abundance when we learn of a correlation, and we readily find causal explanations for the failure of the world to confirm our hypotheses.
We don’t realize how easy it is for us to explain away evidence that would seem on the surface to contradict our hypotheses. And we fail to generate tests of a hypothesis that could falsify the hypothesis if in fact the hypothesis is wrong. This is one type of confirmation bias.
Scientists make all of these mistakes: they sometimes generate hypotheses too readily, they may fail to recognize how easy it is to explain away contrary evidence, and they may not search for procedures that could falsify their hypotheses. Some of the more interesting and important controversies in science involve accusations of unconstrained theorizing, overly facile explanations for apparently contradictory evidence, and failure to recognize opportunities to falsify hypotheses.
An American psychologist once wrote to Freud describing experiments he believed supported Freud’s theory of repression. Freud wrote back to him saying that he would have ignored the experiments if they had found “evidence” contradicting his theory; therefore he was obliged to ignore any experimental evidence allegedly supporting it. To his psychoanalytic colleagues, he sniffed “ganz Amerikanisch” (completely American).
Freud’s put-down seems odd inasmuch as Freud was a committed and highly successful experimenter himself when researching questions of neurology and hypnosis. But Freud’s philosophy of science regarding psychoanalysis was that his interpretation of what his patients told him was the royal road to truth. Anyone who disagreed with these interpretations was simply making a grievous error—which he often made clear to any student or colleague who had the temerity to disagree with him.
The scientific community can’t accept a claim that only a single individual’s judgment counts as evidence. If a theory includes the proviso that only its progenitor (or his acolytes) can assess its truth, then the theory exists to that extent outside of science.
Freud’s certainty and dogmatism are the sure signs of someone who is on shaky epistemic ground. And shaky ground is what many, if not most, psychologists and philosophers of science now believe to have been under Freud’s feet most of the time.
However, Freud’s work gave rise to many hypotheses that are testable by normal scientific means, and some of these have received strong support (and not just from Americans!). The notion discussed in Chapter 3 that the unconscious is a preperceiver is one such hypothesis. The evidence is by now overwhelming that people register countless stimuli simultaneously, only some of which get referred to the conscious mind for consideration, and that such nonconscious stimuli can markedly affect behavior. Research strongly supports other psychoanalytic theories. These include the concept of transference—by which feelings about parents or other important individuals that are formed in childhood are transferred more or less intact to other individuals later in life5—and sublimation, by which feelings of anger or sexual desire that are unacceptable to a person are channeled into unthreatening activities such as artistic creation.6
In the hands of many of its adherents, psychoanalytic theory lacks sufficient constraints. For Freud, and for many of his followers, anything goes. If I say that the patient has an “Oedipus complex” (a desire to have sex with his mother), who is to say that’s baloney? And on what grounds? “Oedipus shmedipus,” as the Jewish mother said, “as long as he loves his mother.”
Freud’s theory of the psychosexual stages—oral, anal, phallic, latent, and genital—included the assertion that arrested development at one of the early stages was possible and would have a major impact on behavior. Toddlers who withheld their feces instead of making for Mama would be stingy and obsessive-compulsive in adulthood. Freud would never have thought it worthwhile to try to find support for such hypotheses outside of the consulting room. And I very much doubt he would have been successful had he tried.
We would say today that one of the chief ways that psychoanalysts derived their hypotheses was by applying the representativeness heuristic, matching cause and consequence on the basis of their perceived similarity.
The psychoanalytic theorist Bruno Bettelheim deduced that the reason the princess in the fairy tale dislikes the frog is that its “tacky, clammy” feel is connected to children’s feelings about sex organs. Who says children don’t like their sex organs? (And tacky, clammy? Well … never mind.) And what’s to keep me from saying that the princess dislikes the frog because the bumps on it remind her of pimples, which she dreads having? Or that she’s a nervous Nellie who’s startled by the frog’s rapid movements?
The concept of the pleasure principle guided Freud’s understanding of human nature until the 1920s. Life centered on satisfying the demands of the id for satisfaction of bodily needs, sex, and the discharge of anger. Dreams were usually about wish fulfillment.
But the driving motives of wish fulfillment and the id’s desires for life-satisfying gratifications seemed to be contradicted by the need of some Great War trauma victims to keep returning to thoughts about the disastrous events they had encountered. Freud also began to notice that children in their play sometimes fantasized the death of loved ones. Patients who were dealing with painful memories that had previously been repressed kept returning to them obsessively and without seeking resolution. And therapists regularly encountered masochists—people deliberately seeking out pain.
Clearly all these people were not being motivated by the pleasure principle. So there must be some drive opposed to it. Freud labeled this drive the “death instinct”—the desire to return to an inorganic state.
The role of the representativeness heuristic in this hypothesis seems pretty clear. People’s main goal in life is the pursuit of pleasure, but sometimes they seem to be seeking the opposite. Therefore there is a drive toward personal extinction. Facile and utterly untestable.
My favorite example of the role of the representativeness heuristic in generating psychoanalytic hypotheses comes from reactions to a paper published in the American Journal of Psychiatry by Jules Masserman, at the time the president of the American Psychiatric Association. The burden of the paper, which was intended as a joke, was that ingrown toenails are symbols of masculine aspirations and intrauterine fantasies. To Masserman’s chagrin, the journal was flooded by admiring commentary on his perspicacity.7
Theories more venerable and better supported by evidence than psychoanalytic theory also have problems with constraints, confirmation, and falsification.
Evolutionary theory has generated thousands of testable and confirmed (or just as frequently disconfirmed) hypotheses about the adaptive nature of the characteristics of organisms. Why are female animals of some species faithful to a single male and the females of other species promiscuous? Perhaps many mates increase the likelihood of reproduction in some species and not in others. Indeed, that turns out to be the case.
Why do some butterflies wear flashy clothes? Explanation: to attract mates. Evidence: male butterflies that have their colors toned down by researchers don’t do well in the mating department. Why should the viceroy butterfly mimic nearly perfectly the appearance of the monarch butterfly? Because the monarch is poisonous to most vertebrates and it’s to the viceroy’s advantage. An animal only needs to fall ill once after eating a monarch in order to avoid ever after anything that resembles a monarch.
But the adaptationist perspective is subject to considerable abuse, and not just by armchair evolutionists.
A construct popular with both cognitive scientists and evolution theorists is the notion of “mental modules”—cognitive structures that evolution has developed for us that guide our ability to deal with some aspect of the world. Mental modules are relatively independent of other mental states and processes and owe little to learning. The clearest example of a mental module is language. No one today would attempt to explain human language as a purely learned phenomenon. The evidence for some degree of prewiring for language is overwhelming: human languages are all similar at some deep level, they are learned at about the same age by people in all cultures, and they are served by specific areas in the brain.
But module explanations by evolutionary theorists are too readily invoked. See a behavior and posit an evolved module for it. There are no obvious constraints on such explanations. They’re as easy and unconstrained as are many psychoanalytic explanations.
In addition to the overly facile nature of many evolutionary hypotheses and their violation of Occam’s razor, many such hypotheses are not testable by any means currently available. We aren’t obligated to pay attention to theories that are untestable. Which isn’t to say we’re not allowed to believe untestable theories—just that we need to recognize their weakness compared to theories that are. I can believe anything I want about the world, but you have to reckon with it only if I provide evidence for it or an air-tight logical derivation.
The field of psychology affords many examples of too-easy theorizing. Reinforcement learning theory taught us a great deal about the conditions that favor acquisition and “extinction” of learned responses such as a rat’s pressing a lever to get food. The theory guided important applications such as treatment of phobias and machine learning procedures. But theorists in that tradition who attempt explanations of complex human behavior in terms of presumed reinforcements sometimes make the same mistakes as many psychoanalytic and evolutionary theorists. Little Oscar does well in school because he was reinforced for conscientious behavior when he was a child, or because other people modeled conscientious behavior for him. How do we know? Because he is now so conscientious in school and does so well there. How else could he have become so conscientious other than through reinforcement for conscientious behavior or behaving like models he has observed being rewarded for such behavior? Hypotheses like that are not merely too easy and unconstrained but circular and not falsifiable by current methods.
Economists of the “rational choice” persuasion sometimes exhibit the same lack of constraint and circular reasoning as psychoanalytic, evolutionary, and learning theorists. All choices are rational because the individual wouldn’t have made the choice if he hadn’t thought it was in his best interests. We know the person thought it was in his best interests because that’s the choice the person made. The near-religious insistence that human choices are always rational leads such economists to make claims that are simultaneously untestable and tautological. The Nobel Prize–winning economist Gary Becker maintained that an individual who chooses to begin a career of drug addiction has to be considered rational if the individual’s chief goal in life is to satisfy a need for instant gratification. Facile, irrefutable, and circular. If drug addiction can be “explained” as rational behavior by a rational choice theorist, the theory is bankrupt in that person’s hands. All choices are known in advance to be rational, so nothing can be learned about the rationality of any given choice.
But of course my critique isn’t limited to scientists. Mea culpa and so are you. Many of the theories we come up with in everyday life are utterly lacking in constraints. They’re cheap and lazy, tested if at all by searching only for confirmatory evidence, and too readily salvaged in the face of contradictory evidence.
Judith, a talented young chemist who we thought was surely headed for a distinguished career in science because of her energy and intelligence, has left the field to become a social worker. She must have a fear of success. Too easy to generate that theory and too easy to apply it. And what could convince us that fear of success was not involved?
Bill, mild-mannered neighbor, erupted in rage toward his child at the big-box store. He must have an angry and cruel streak that we hadn’t previously seen. The representativeness heuristic, the fundamental attribution error, and the belief in the “law” of small numbers aid and abet one another in producing such theories willy-nilly.
Once generated, evidence that should be considered as disconfirming the hypothesis can be explained away too easily. I have a theory that start-ups supported by large numbers of small investors, even when little information about the company is available, are destined to be highly successful. This applies to the newly founded Bamboozl.com, so it’s going to have great success. Bamboozl goes bust, but I’m going to be able to come up with any number of reasons for its failure. Management was not as talented as I had thought. The competition moved much faster than could have been predicted.
I believe that announcement of a cutback of “quantitative easing” by the Federal Reserve will result in fear in the equity markets, causing a drop in stock values. The Fed announces a slowdown of quantitative easing and the markets go up. Because of … you name it.
Jennifer, disorganized in her private life, would never make a good newspaper editor, a job that requires meeting deadlines and simultaneously juggling information obtained from Internet sources, assigning tasks to copy editors, and so on. Lo and behold, she turns out to be an excellent editor. The mentoring she got from her predecessor early on must have saved her from the consequences of her fundamentally chaotic temperament.
I’m not saying we shouldn’t come up with hypotheses like the above, just that recognition of the ease with which we generate them, and the facility with which we can explain away contradictory evidence, should make us cautious about believing them.
The problem is that we don’t recognize our own strength as theorists.
Discussion of theory testing leads us to the question of just what kinds of theories can be falsified and what kinds of evidence could serve to do so.
Falsifiability
If the facts don’t fit the theory, change the facts.
—Albert Einstein
No experiment should be believed until it has been confirmed by theory.
—Arthur S. Eddington, astrophysicist
“It’s an empirical question” is a statement that ought to end far more conversations than it does.
Deductive reasoning follows logical rules, producing conclusions that can’t be refuted if the premises are correct. But most knowledge is obtained not by logic but by collecting evidence. Philosophers call conclusions that are reached by empirical means a form of “defeasible reasoning.” That essentially means “defeatable” reasoning. If you look for evidence that plausibly would support your hypothesis, and it does, you may have a reasonable belief. If the data don’t support your hypothesis, then you either have to find another way to support the hypothesis or hold on to it with appropriate tentativeness. Or, as Einstein said, show that the “facts” are mistaken.
If someone makes a theoretical claim but can’t tell us what kind of evidence would count against it, we should be especially wary of that person’s claim. As often as not, the person’s simply telling you what ideology or religion has revealed. He’s operating in the prophetic mode rather than the empirical tradition.
The falsifiability principle is now enshrined in law in several states as a criterion for teaching something purporting to be science. If it’s not falsifiable, it’s not science and can’t be taught. This is intended primarily to rule out the teaching of creation “science.” A typical creationist claim might be, “The human eye is vastly too complicated to have come about by such a cumbersome and laborious process as evolution.” The appropriate answer to that proposition is, “Who says?” Such claims are not falsifiable.
The falsifiability requirement makes me slightly nervous, though, because I’m not sure the theory of evolution is falsifiable either. Darwin believed it was. He wrote, “If it could be demonstrated that any complex organ existed, which could not possibly have been formed by numerous, successive, slight modifications, my theory would absolutely break down. But I can find no such case.”
And no one has. Or can. If the creationist says that such and such an organ could not have evolved, the evolutionist can only say, “Yes it could.” Not very convincing. And there is no way at present to test such claims empirically.
Nevertheless, the theory of evolution wins out over any other theory of the origin of life—of which there are only two others, namely God and seeding by extraterrestrials. Evolution theory triumphs, not because it’s falsifiable and has yet to be falsified, but because (a) it’s highly plausible, (b) it accounts satisfactorily for countless thousands of diverse and otherwise apparently unrelated facts, (c) it generates hypotheses that are testable, and (d) as the great geneticist Theodosius Dobzhansky said, “Nothing in biology makes sense except in the light of evolution.”
The evolutionary hypothesis and the God hypothesis are of course not incompatible. “God works in mysterious ways his wonders to perform.” Evolution is actually one of the less mysterious ways an all-powerful being might have chosen to kick-start life and keep it running all the way to us.
Dobzhansky, incidentally, was a religious man. Francis Collins, the leader of the Human Genome Project and the current director of the National Institutes of Health and (obviously) a believer in evolution, is an evangelical Christian. Collins would never pretend that his belief in evolution is of the same kind as his belief in God—which he would be the first to acknowledge is not falsifiable.
Popper and Poppycock
The Austro-British philosopher of science and London School of Economics professor Karl Popper promulgated the view that science proceeds solely by conjecture and falsification of the conjecture or failure to falsify it. Popper maintained that induction is unreliable. In his view we don’t (or shouldn’t) believe propositions simply because they’re supported by evidence from which we induce that the propositions are correct. “All swans are white” was supported by millions of sightings of swans that were white and were never any other color. Oops. Australian swans are black. Hypotheses can only be disconfirmed, not confirmed.
Popper’s injunction is logically correct. No amount of sightings of white swans can establish the truth of the generalization that all swans are white. There is an asymmetry: empirical generalizations can be refuted but can’t be proved true because they always rest on inductive evidence that could be refuted at any moment by an exception.
Though correct, Popper’s contention is pragmatically useless. We have to act in the world, and falsification is only a small part of the process of generating knowledge to guide our actions. Science advances mostly via induction from facts that support a theory.8 You have a theory based on deduction from some other theory, or from induction based on observation of available evidence, or from an inspired hunch. You then generate tests of that theory. If they support the theory, you conclude that it’s more likely that the theory is correct than it would be in the absence of such evidence. If they don’t support the theory, you reduce your confidence in the theory and either look for other tests or put the theory on hold.
Falsification is important in science, for sure. Some facts are powerful enough that they’re sufficient to utterly disabuse us of some hypothesis. Observation of chimpanzees that remained inactive and apparently asleep while being operated on after being given curare led to the hypothesis that curare obliterates consciousness. That theory went out the window when the first human was operated on under curare and reported, I assume with expletives, that he had been awake the whole time and felt the surgeon’s every excruciating maneuver. The hypothesis that the moon is made of green cheese was destroyed by Neil Armstrong in 1969.
Once you know the knockdown fact, the theory is kaput. (For the time being. Many a theory has been knocked down only to rise again in a modified version.) But mostly, research is a slog through findings that support or contradict the theory to one degree or another.
The glittering prizes in science don’t go to the people who falsified someone else’s theory, or even one of their own—though their research may have that incidental effect. Rather, the laurels are for scientists who have made predictions based on some novel theory and demonstrated that there are important facts that support the theory and are difficult to explain in the absence of the theory.
Scientists are much more likely to think they accept Popper’s anti-inductive stance than philosophers of science are to endorse it. The ones I know think it’s utterly wrong. Science advances mostly by induction.
Popper, incidentally, criticized psychoanalytic theory as unfalsifiable and insisted it could therefore be ignored. He was quite mistaken in that. I pointed out earlier that many aspects of the theory are indeed falsifiable, and some have in fact been falsified. The central claims of psychoanalytic theory about therapeutic principles have been, if not refuted, at least shown to be dubious. There is no good evidence that people get better by virtue of dredging up buried memories and working through them with the therapist. And certainly it’s the case that psychotherapeutic practices owing nothing to psychoanalytic concepts have been shown to be more effective.
I was told by an eminent philosopher of science that Popper was actually quite ignorant of psychoanalytic theory. He knew only what he picked up in café conversations.
What about Einstein’s outrageous comment that facts have to change if they don’t support a theory? Many interpretations of the comment are possible, but the one I prefer is that we’re allowed to continue to believe a satisfying theory for which there’s good support, even though there are facts that are inconsistent with the theory. If the theory is good enough, the “facts” will eventually be overturned. Eddington’s quip makes the coordinate point: we’re on shaky ground if we believe an alleged fact when there’s no plausible theory that should lead us to accept it.
Adherence to Eddington’s rule could have spared my field of social psychology a great embarrassment. Its most venerable journal published a wildly implausible claim about extrasensory perception. An investigator asked participants to predict what statement a computer would select at random from a prepared list over a large number of trials. Participants allegedly could predict the behavior of the computer accurately at a level beyond what would be achieved by random guesses. The claim was therefore for paranormal foretelling of future events produced by a machine that could not foretell the events itself. The claim is dismissible on its face; no evidence could support such a theory. Several people with time on their hands tried to replicate the findings and couldn’t.
The Hocs: Ad and Post
We have many techniques that allow us to ignore evidence that on its face would seem to contradict our predictions. One of the dodges has to do with dubiously legitimate fixes to a hypothesis. Ad hoc postulates are amendments to a theory that don’t follow directly from the theory and serve no purpose other than to keep the theory propped up. Ad hoc means literally “to this.” (Ad hoc committees are subcommittees of the whole set up to deal with a specific issue.)
Recall from Chapter 14 Aristotle’s invention of the property of “levity.” This was an ad hoc amendment to the theory that an object’s “property” of gravity causes it to fall to earth. Levity was postulated to handle the fact that some things float in water instead of sinking. The concept of levity is a special-purpose fix to Aristotle’s theory of gravity, intended to handle a problem that would otherwise wreck the theory. It doesn’t follow from the basic theory in any principled way. The theory itself was what I call “placebic.” Nothing has actually been explained. The French playwright Molière derides such explanations when he has a character attribute a sleeping potion’s effect to its “dormative virtues.”
Ptolemy’s epicycles were an ad hoc solution to the problem that heavenly bodies did not orbit the earth in the perfect circles that were presumed by his contemporaries to be the necessary pattern of motion.
Einstein’s postulation of the cosmological constant, noted in Chapter 14, was a special-purpose fix to the theory of general relativity. It was postulated just to account for the “fact” that the universe was in a steady state. Oops. It isn’t in a steady state.
An astronomer has come up with an ad hoc theory to account for the failure of Mercury to orbit the sun in the way demanded by Newton’s theory. The astronomer simply posited that the sun’s center of gravity shifts from its center to the surface—when and only when the planet in question is Mercury. A desperate (and deliberately hilarious) move to salvage a theory by a special-purpose postulate.
Ad hoc theories are typically also post hoc—literally “after this”—meaning a contrivance after the data are in to explain what was not predicted in advance. Post hoc explanations are all too easy to come up with once an anomaly is spotted. “Well, I said I was certain Joan would win the spelling contest, but I couldn’t have known that she would be upset by having done badly on a math test the morning of the contest.” “Yes, I said Charlie would fail as manager because of his social insensitivity, but I couldn’t have guessed that he would marry a woman who would trim the rough edges off him.”
In my first years as an academic, I habitually made confident predictions about how someone would function as a department chair or journal editor. When the predictions failed—as they did about as often as not—I had no trouble explaining why any particular prediction was off base. This spared me the necessity of recalibrating my theories about what leads to being a success in a particular role. I’m pleased to be able to say that I now make my predictions with far less certainty than before. Or at least I try to keep my predictions to myself. This saves me considerable embarrassment.
To this point I’ve tacitly adopted the lay view that scientific investigation and theory building are cut-and-dried procedures following clear rules about hypothesis generation, the search for evidence, and acceptance or rejection of the hypothesis. For better or worse, this is far from the case, as we’ll see in the next chapter.
Summing Up
Explanations should be kept simple. They should call on as few concepts as possible, defined as simply as possible. Effects that are the same should be explained by the same cause.
Reductionism in the service of simplicity is a virtue; reductionism for its own sake can be a vice. Events should be explained at the most basic level possible. Unfortunately, there are probably no good rules that can tell us whether an effect is an epiphenomenon lacking causal significance versus a phenomenon emerging from interactions among simpler events and having properties not explainable by those events.
We don’t realize how easy it is for us to generate plausible theories. The representativeness heuristic is a particularly fertile source of explanations: we are too inclined to assume that we have a causal explanation for an event if we can point to an event that resembles it. Once generated, hypotheses are given more credence than they deserve because we don’t realize we could have generated many different hypotheses with as little effort and knowledge.
Our approach to hypothesis testing is flawed in that we’re inclined to search only for evidence that would tend to confirm a theory while failing to search for evidence that would tend to disconfirm it. Moreover, when confronted with apparently disconfirming evidence we’re all too skillful at explaining it away.
A theorist who can’t specify what kind of evidence would be disconfirmatory should be distrusted. Theories that can’t be falsified can be believed, but with the recognition that they’re being taken on faith.
Falsifiability of a theory is only one virtue; confirmability is even more important. Contra Karl Popper, science—and the theories that guide our daily lives—change mostly by generating supporting evidence, not by discovering falsifying evidence.
We should be suspicious of theoretical contrivances that are proposed merely to handle apparently disconfirmatory evidence but are not intrinsic to the theory. Ad hoc, post hoc fixes to theories have to be suspect because they are too easy to generate and too transparently opportunistic.