D Cognitive Advantages Offered by the Hypothesis

Historically, scientists wrestled with the concept of the hypothesis, trying to balance its advantages with its real or imagined disadvantages; partly be­cause of the many meanings the word had (Chapter 2).

The comments seem unambiguous, but before placing Cajal in the anti­hypothesis camp, we read, “The hypothesis is an interpretative questioning of nature” and that “observation, explanation or hypothesis, and proof” are key elements in scientific discovery. He feels that “A hypothesis is necessary; without it phenomena cannot be explained,” and he quotes Jacob Christoph Le Blon, “he who refuses to accept hypothesis as a guide is resigned to accept chance as a master.”

Cajal resolves these contradictory points of view by distinguishing “between working hypotheses...

Karl Popper, at least, was unambivalent about the virtues of the hypothesis: in the preface to Conjecture and Refutations,¹⁸ Popper writes, “The essays and lectures of which this book is composed are variations upon one very simple theme—the thesis that we can learn from our mistakes.” His work establishes the hypothesis as the preeminent practical tool for making and learning from mistakes.

In the remainder of this chapter, I will expand on the themes introduced by Cajal, Locke, and Popper.

8. C.1 Working with Your Built-In Drive to Understand the World

The psychological literature is full of examples of our being under the influence of our own unconscious interpretations of the world. The influence of the un­conscious can be very hard to overcome, especially if we're unaware of it. These concerns touch on cognitive matters that we'll explore in greater detail later, but an example from the work¹⁹ of Daniel Kahneman and Amos Tversky helps illus­trate the point. Take a minute to think about the infamous “Linda problem” if you've never encountered it:

“Linda is 31 years old, single, outspoken, and very bright. In college she was a philosophy major and was deeply concerned with issues of discrimination and social justice. She participated in antinuclear demonstrations.” Which of the following two statements about Linda would you say is more likely? (A) Linda is a bank teller. (B) Linda is a bank teller and is active in the feminist movement.

What was your impression of Linda? Which is the correct answer? Would you believe either (A) or (B) might be correct? It's true. Depending on what you mean by “likely” and “correct”—depending on how you interpret the context of the question, you can make a plausible case for either one. The case in favor of an­swer A, “bank teller,” is that it is the larger class, it includes the smaller class, and, therefore, logically, it must be the right answer if by “more likely,” we mean “more statistically probable.” The case for answer B, “bank teller who is a feminist,” is that we're being asked for our best guess about Linda's societal identity—whether she is a “feminist” or not—given our knowledge of the world.

Thousands of people have considered and evaluated the Linda Problem, and most of us preferred answer B, the socially aware but logically weaker one. There is much more to say about the Linda Problem and its interpretations that we'll cover in Chapter 11 because it has several lessons about scientific reasoning. Here I use the problem to illustrate our universal tendency to spin a narrative when we're trying to understand the world. Working from the skimpy information in the in­itial description of Linda, we instinctively perceived her in a social context and constructed a story, a coherent narrative, about the kind of person she is.

A narrative is an account of a sequence of events or an arrangement of elements that reveals connections among the parts. A narrative makes sense. Our subliminal internal story about Linda is essentially an implicit stereotyped hypothesis about her. It puts all of the pieces of the evidence together and clarifies her character. Our hypothesis tells us what she's like and predicts that she must be a feminist.

The hypothesis works well in science in part because of this predisposal to create interpretive organizational structures and to make predictions.²⁰ But, at the same time, this drive to understand creates superstitions, gives rise to certain religious beliefs,²¹ and explains why we are so good at seeing shapes in the clouds or more generally imagining patterns in randomness.²² We impose orderly inter­pretations on nature as much as we discover them in nature.

Given that our in-born drive to understand leads both to science and super­stition, the big question is how do we, as scientists, cope with it? How do we reap the benefits of our tendency to seek hidden relationships, which is the core of the scientific urge to explain, while avoiding its pitfalls? The physicist Henri Poincare expresses the danger this way: “Some hypotheses are dangerous: first and foremost are those which are tacit and unconscious.”²³ Where do tacit and dangerous hypotheses come from? From the never-ending saga, the ongoing se­ries of stories, that we tell ourselves.

Instead of pretending that we can easily escape the traps set by our auto­matic, hypothesis generating minds and be innocently led by pure questions or curiosity, we should consciously recognize and go with our innate tendencies. We should take charge by formulating and testing our hypotheses explicitly.

As I've suggested, the difficulty in dealing with the Linda Problem stems from the narrative that we spin about her and the implicit hypothesis about her life that the narrative suggests²⁴. In contrast, when you have a problem that lacks any evocative narrative subtext, you are more likely to reason about it logically. If you were to asked to guess whether it is more likely that I have an “apple” or a “green apple” in my lunch box, youd immediately realize that the category of “apple” is the larger one and you'd have no trouble getting the logically correct answer. Formally, of course, the apple problem is identical to the Linda Problem (we'll return to these nuances in Chapter 11). Scientific thinking often profits from objectifying problems, turning them into questions about neutral, visualizable elements so that you can think clearly about their relationships.

If you deliberately make your hypothesis conscious and explicit, you can also take strategic advantage of two major safeguards against bias and self­deception: rigorous skepticism and multiple hypotheses.

As a Popperian Critical Rationalist, you are always skeptical about hypoth­eses; your own and others'. You look for flaws in them, not for ways to confirm them. You break the grip of the narrative of the hypothesis and look at it in the way a literary critic looks at a novel: skeptically, analytically, trying to determine whether the book achieves its stated purpose or how it fails.

8. C.2 Multiple Hypotheses as a Defense Against Narrative Bias

John Platt's program²⁵ of Strong Inference extends and amplifies elements of Conjectures and Refutations. According to Platt, the best defense against the in­filtration of bias into your thinking is more, not less, hypothesizing; any phe­nomenon has more than one conceivable explanation, meaning that more than one hypothesis can, conceivably, explain it. Although it's infrequently used, the technique of multiple hypotheses is invaluable. Since we inherently want explanations and interpretations, we should be actively trying to create more of them. Try to develop several, quite different, possible explanations to account for the same initial observations. If there is a danger of forming too close an attach­ment to one hypothesis, do not have one. Have two! Even three or four, if your imagination and diligence are up to it. See what you've been taking for granted, or ask if relaxing your assumptions would lead to a new hypothesis. Once you have multiple hypotheses, you can take an Olympian view of them. They are all your babies, and you can be justly proud of each one. And, if having one hypo­thesis is bad because you want it to be right, then having multiple hypotheses gives you more chances to be right.

Best of all, having multiple hypotheses not only encourages you to maintain a proper distance from your pet theory, it also helps you design better experiments. An ideal experiment acts like a razor, neatly separating results into classes that are either compatible or incompatible with a hypothesis. You can simultaneously falsify one and corroborate others.

Here's an example of how this works. As noted in Box 7.2, neuroscientists have long been fascinated by the problem of how memories are stored in the brain. Our best guess is that a phenomenon called long-term potentiation (LTP) is respon­sible, and the current consensus is that the synapse, the junction between signal sending cells (presynaptic) and signal receiving cells (postsynaptic) is where LTP occurs. Somehow the synapse becomes “strengthened” by LTP, meaning that the presynaptic cell becomes more effective in stimulating the postsynaptic cell and either of two hypotheses could explain LTP: an increased release of the chemical neurotransmitter, glutamate, from the presynaptic cell, or an increased number of glutamate receptors (and there are two kinds of glutamate receptor) on the postsynaptic cell.

I want to look at one of the classic experiments from a slightly different per­spective. Remember that investigators studied the postsynaptic response that was caused by glutamate. If LTP was caused by the release of more glutamate (regulated by the presynaptic cell), then the responses caused by both glutamate receptor types should have gotten bigger. If LTP was accompanied by an increase in the response caused by only one kind of receptor, it would mean that LTP was regulated by the postsynaptic cell. In fact, the investigators observed a highly se­lective increase in only one kind of receptor-mediated response, thereby simulta­neously falsifying the presynaptic hypothesis and corroborating the postsynaptic one. Cutting just like razor.

You have to know what the alternative hypotheses are to design good experiments. Of course, even key experiments can be asymmetrical at times: one outcome could be fatal for one hypothesis, but an alternative outcome would be inconclusive. This is not a problem; normal science advances unevenly, and if you can falsify one hypothesis by an experimental outcome, that's progress.

Encouraging you to consider competing ideas and develop multiple hypo­thesis is a major advantage to using hypothesis. Hypotheses can also help or­ganize your scientific thinking in a number of less specific ways as well.

8. C.3 The Hypothesis as an Aid to Scientific Thinking

You can be a committed hypothesis-testing scientist without being busy testing hypotheses every minute of the day, and you don't have to swear a loyalty oath to do hypothesis-based science either. Most experimental investigations, including the determinedly hypothesis-testing ones, meander at times; an experimenter may draw a tentative conclusion, try out a new idea, and, after some pilot experiments, return to the original or a modified design. The hypothesis is a cog­nitive tool, a device that helps you think clearly, and, in the following sections, I'll briefly review few of the ways in which it helps.

8. C.3.a The Hypothesis as Blueprint

A blueprint is a detailed architectural drawing; it is a definite plan with specific details—the doorframe is to be X inches wide and Y inches tall. If a blueprint were imprecise and vague, the builder wouldn't know how to proceed. Likewise, a scientific hypothesis provides a blueprint for an investigation, and it must be specific if you, its author, as well as the scientific community, are to know how to proceed.

As an explanation of a phenomenon, a hypothesis must include specific details about what it says and doesn't say for several reasons: you can only test a hypo­thesis if you know what it says. The hypothesis entails predictions, and you use deductive reasoning to bring them out. This is critical: by making predictions, a hypothesis tells you in no uncertain terms what you must do to see if your expla­nation is false. If your plan is fuzzy and indistinct, then neither you nor anyone else knows how to test it.

In this way, the framework of the hypothesis supplies built-in answers to per­ennial experimental questions: “What's next?” “What have I overlooked?” Once formulated, your hypothesis crystalizes what your investigation is about, then it indicates a course of action.

The LTP experiment constitutes a textbook example of how hypotheses pro­vide a blueprint for the investigators. The presynaptic hypothesis entailed not only the prediction of a greater release of glutamate, but also a constellation of biochemical and biophysical predictions that could cause more glutamate to be released. Similarly, the postsynaptic hypothesis entailed an analogous con­stellation of predictions having to do with glutamate receptor regulation and expression. As always, numerous known and unknown biological and other complexities conspire to keep us from being absolutely certain about the result of any one experiment, no matter how elegant and persuasive. Having done an experiment, experimenters go back to the blueprint, derive a new prediction, and test it, repeating the process until they and their colleagues decide to let “the investigation rest.”

Together, the hypothesis and the principle of falsification triage the experiments; they indicate which are the more important ones to do. You can't do everything, and there is little point in doing “confirmatory” experiments where the results are hardly in doubt and your hypothesis is left unchallenged. If you can carry out a severe test that puts the hypothesis at risk, then you're sure to learn something. And you must do it because your arch-competitor definitely will. The blueprint makes it plain to everybody what to do.

Of course, it's important to stress that blueprints are drawn on paper not engraved in granite. A hypothesis is not a rigid set of steps to follow. Just as a pro­spective home builder might look at an architect’s blueprint and request another window in the bedroom, a hypothesis is a set of guidelines, an intended plan. It is flexible and can be changed according to need or circumstance. Which tests you do will depend on practical as well as scientific or logical reasons, and they may reflect such imponderables as your esthetic sense and intuition.

8. C.3.b Hypotheses Define Success and Failure in Research

Everyone agrees that failure is an integral part of science.²⁶ Often experiments do not work (e.g., the results are obscured by noise or are uninterpretable) the first (or second, or third) time. You have to be willing to try, fail, and learn from your mistakes. Hypothesis-based reasoning is a systematic way of making mistakes and extracting learning opportunities from them. Because your hypo­thesis makes definite predictions, you know how to falsify it; hence, failure of a hypothesis is a well-defined end-point. In contrast, if you go about science in a loose, undirected way, you can’t reap the same benefit from the lessons of failure. You can’t even fail if you haven’t tried to succeed. If you seriously respect the im­portance of making mistakes in science, you will inevitably gravitate toward the hypothesis.

But we should be careful in how we talk about failure. A falsified hypo­thesis is not a failure. Eliminating a hypothesis is progress; going backward or going nowhere is failing. In Chapter 10B, I’ll review the classic example of the Michelson-Morley experiment, which falsified the prevailing “ether” theory of light propagation, and argue that it was a brilliant scientific success. Why does it matter how we label experimental outcomes? Calling successful falsifica­tion “failure” sustains the mistaken notion that only experiments that confirm predictions should be considered successful. This notion, in turn, contributes to publication bias that favors publishing only “positive” results.²⁷ Encouraging you to think objectively about your results and what they mean is, accordingly, an­other advantage of hypothesis-based science.

8. C.3.c Hypothesis Testing as a Conscious Cognitive Process

You might decide to do an experiment or launch an investigation because you have some conscious or unconscious curious thoughts or questions. We can’t say where hypotheses come from, but, even if we could, it would hardly ever matter. Having a hypothesis is not a prerequisite for starting an investigation, and once you’ve started, you quickly become absorbed in the “what” and the “how” of the plan, and you don’t worry about the “why.”

Having a hypothesis keeps you intellectually involved in the experiment. This is the “interpretive questioning” of nature that Cajal was referring to. You cannot fully carry out an investigation in a purely passive mode, and hypothesis­based science encourages active engagement at every step. One consequence of increased automatic data collection is that you can become disengaged from scientific reasoning, which has potential upsides and downsides. The ability to gather and analyze massive quantities of high-quality data mechanically and im­partially could definitely be a good thing. However, no matter what your tech­nical resources are, you need to decide which data to collect and how to analyze them. Designing and shepherding a hypothesis-testing experiment through to completion is an interactive process; you do something, see what happens, puzzle out what that means, and continue, or do something else.

Obviously, a scientist asking a series of questions may be as intellectually in­volved as a hypothesis tester is; intellectual involvement is a personal charac­teristic that, one hopes, all scientists share. My point here is that, by its nature, rigorously testing a hypothesis demands constant, conscious interpretative in­teraction in a way that other methods, question-answering, for example, may or may not do.

8. C.3.d Self-Organization

The writer Norman Mailer reported that doing the morning New York Times crossword puzzle each day was how he “combed his brain”²⁸ to get ready for the tasks ahead. Forming a hypothesis is a way of combing your brain and creating order from the occasional tangle of thoughts or the mass of unclear or incon­sistent data that can pile up. Thinking about your hypothesis helps straighten things out.

You need to exert determined efforts to make sense of observations, and yet the course of an experiment is not always (i.e., is rarely) as straightforward as you imagine it will be. Consideration of possible test outcomes and their inter­pretations is invaluable in making course corrections as required when a failed prediction falsifies your hypothesis. And making an explicit hypothesis makes it harder to for you to “fool yourself” in Feynman's often-quoted words. In other words, the cognitive act of formulating a hypothesis, independent of the hypo­thesis per se, is one of the advantages of working with hypotheses.

8. C.3.e Memory and Narrative Structure

Organized information is the easiest to recall, and the hypothesis organizes in­formation into a narrative structure that fosters memory. A dramatic nonsci- entific illustration of the power of structure to facilitate memory comes from competitions where the participants, “Memory Athletes”²⁹ try to recall large numbers of items. Most of the athletes rely on the “method of loci,” a technique of mentally placing images of items that they need to recall in various locations in a familiar physical structure, also called a “memory palace.” Mentally revisiting a location triggers retrieval of the image that was placed there. Recall of words arranged into meaningful sentences is much better than recall of random lists of words. Cognitive psychology also assures us that, when we're given compli­cated information, we tend to remember broad meanings not specific details.³⁰ And we recall stories better than lists of facts, meaningful sentences better than random word lists. Nature is complicated: biological sciences often deal with large numbers of entities interrelated in complex ways.

The narrative structure provided by the hypothesis helps organize scientific facts into logical, coherent chunks of information. We see the world as an orderly arrangement of thematically linked mental images.

8.C.3.f Communication

Not only do you, the individual scientist, gain a good sense of what is involved in your hypothesis-based inquiries, but so do your colleagues who read your reports and want to follow-up on them. The hypothesis is an exceptionally effi­cient mechanism for conveying the big picture because it puts your results into a logical, easy-to-follow context. Indeed, as a general rule, one really can't overstate the importance of readily-grasped communication in all aspects of science. The quality of your scientific publications, grant proposals, job talks, lectures, etc. are all directly related to how easily others can understand them. Superb communi­cation skills on their own do not produce excellent science, but deficient com­munication skills can most certainly obscure scientific excellence. And there is no necessary correlation between scientific expertise and communication skills. Brilliant thinkers are not necessarily brilliant communicators.

You may protest; surely a good idea will be recognized, even if it is not ideally expressed! Yes, if you are a surpassing genius, it may well be that the significance of your revolutionary contribution will instantly be perceived and rewarded no matter how inarticulately it is expressed. On the other hand, the history of science is replete with tales of key discoveries being neglected in their day and with disputes among scientific giants that often focused minutely on who said precisely what. And there is even the possibility that one of your competitors will simply appropriate you poorly articulated discovery, do a few related experiments, and rebrand it as his own (this sort of thing does, unfortunately, happen). In any event, the great majority of scientists will almost certainly ben­efit from paying close attention to what they want to say and how they say it.

Science is a social enterprise and having, or anticipating, the critical comments of others will also help improve the soundness of your thinking and uncover your unconscious biases. Stating your hypothesis is a public declaration of your thoughts and reasoning; reviewers and editors of scientific journals and others routinely provide valuable feedback, and the quantity and quality of their feedback depends on their ability to understand what you’re trying to say. You cannot benefit as much if your ideas are not readily intelligible. Many of us are familiar with the global condemnations—“diffuse” “unfocused,” and “poorly organized”—that spell doom if they appear in the review of your National Institutes of Health grant application.

Efficient, informative communication is the overriding objective of a scien­tific report, which, therefore, is not a veridical historical accounting of the actual course of an investigation. Readers expect to learn about the new information you’ve found, not contorted descriptions of your thought process, no matter how fascinating you believe them to be. You can assemble your report around your hypothesis to maximize its logical flow and scientific cogency. You might have become aware of the killer control experiment, the one that you should probably have done first, only after your project is well under way. This is not a problem: do the experiment and report the outcome in the first paragraph of the Results sec­tion if that makes the most sense. The purpose of a scientific publication is to present your data and communicate your ideas, not to build suspense or impress the reader with your literary skills. (After you’ve won the Nobel Prize, historians of science may find the exact sequence in which you did your experiments to be deeply informative, but probably not before then.)

Finally, the explicit hypothesis is especially useful for sciences such as neu­roscience and other biological sciences that rely to a large extent on natural lan­guage for communication. Hard sciences (the rigorously quantitative ones—e.g., physics, chemistry, etc.) use mathematics as a common language; chemistry also has unambiguous sets of symbols for organizing, expressing, and communicating important concepts. Neuroscience (much of which is somewhere on the con­tinuum between hard and soft sciences—it could be considered a “firm” science) uses English as the de facto official language (the lingua franca) of most large international conferences and science journals, in part because English readily assimilates new technical terms and is notoriously flexible. However, these advantages are partly offset by a tolerance of imprecision and lack of rigorous standardization; no “Academie Englaise” stands guard over its usage. Hence, any field that relies heavily on English for communication, evaluation, etc., on an in­ternational scale may be more in need of uniform linguistic conventions to foster communications than more quantitative fields. Devices such as the explicit hy­pothesis help reduce ambiguities that can otherwise arise.

8.