C Further Criticisms of Popper and Platt (Optional)

Up to this point, I've addressed the specific philosophical objections to Conjectures and Refutations that I think are the most germane to scientific thinking. A few additional concerns that I'll consider have a philosophy of sci­ence slant that not every lab scientist will find necessary, and so, depending on your interests, you may want to skip ahead to Chapter 4.

3. C.1 Popper Versus Platt?

As I noted, the philosopher, Peter Godfrey-Smith favors Platt's program over Popper's. Although Godfrey-Smith has objections to falsification, he suggests that, if we can't really rule out alternative hypotheses, a la Sherlock Holmes, maybe we can at least identify most of them as being very unlikely. Or perhaps we can find “partial support” for some of the remainder. As we noted earlier, par­tial support is not really what science is after—the Truth is. Nevertheless, we can ask how the search for partial support might be carried out.

One difficulty for John Platt's eliminative approach to hypothesis testing is that, unlike the relatively small group of suspects that Sherlock Holmes typically had to process, scientists have an unknown number of alternative hypotheses to sort through. Platt himself does not say much about the universe of alternatives— he alludes to “all” of them but doesn't worry about how many there might be. His program is open-ended and easily accommodates new ones as they crop up.

Theoretically speaking, things are not so simple. In principle, scientists have to confront an infinite number of alternative hypotheses that could account for a phenomenon. Godfrey-Smith agrees, but expresses the hope that “maybe there are ways” of getting down to a manageable number of them. You might wonder, if we are going to place our bets on hope, why not hope that scientists can find manageable ways of falsifying hypotheses? Or acknowledge that, to a large ex­tent (Section 3.B.6), that's what they already do? In short, his stance—pro-Platt, anti-Popper—seems arbitrary; he rejects falsification without explaining how to eliminate hypotheses.

3. C.2 Does Taking Rational Practical Action Demand Inductivist Justification?

When philosophers criticize Conjectures and Refutations they do it on the basis of rational argument, but, as we'll see in Chapter 11, it is not always ob­vious how we are to understand “rational.” In Chapter 2 we considered two standards for rationality: one which is consistent with the narrow demands of probability and logic, and one that is better adapted to the needs of our ancient hominid ancestors. For Popper, the essence of scientific rationality is embodied by Conjectures and Refutations. If you believe that science cannot achieve ab­solutely unchallengeable results, then the most rational state of mind is the “readiness to accept criticism,” and the soundest hypotheses are those that have been subjected to the severest criticisms and tests and therefore are, to the very best of your knowledge, true. Acting rationally means to be guided by true statements.

The philosopher Wesley Salmon³¹ thinks it is irrational to base practical actions on this reasoning because it omits what he believes is a mandatory link between a corroborated hypothesis and a predictable outcome: induction. He says that you have to make inductive predictions from theories in order to act ra­tionally. Unless a hypothesis has received, in addition to extensive corroborating evidence, the blessing of inductive confirmation, you can't rationally choose it. He concludes that induction is inescapable and therefore that Popper's thinking contains hidden inductive elements. Does it?

Popper not only dismisses the thought that “induction” can add anything, but he denies that predictions are the basisfor taking action. The denial that predictions can constitute a basis for action will strike some people as counterin­tuitive, and I'll try to make sense of it. In essence, Popper wants to ground prac­tical actions on what we know to be true; not what we predict to be true.

Popper is a realist, which, you'll recall, is a technical term for someone who believes in the existence of an external world (external to our own minds) that is “regular,” meaning that it is governed by physical laws even if we don't know what they are.³² We cannot prove that realism is true: Popper accepts it for “metaphys- ical³³ reasons, but he's not alone.

But because realism assumes that nature is regular (i.e., not utterly chaotic), Salmon believes that it incorporates a “version of a principle of the uniformity of nature”; ergo, the basis for valid induction is suddenly back in play. Therefore, since Popper accepts realism, he must also accept the tenets of induction and, ipso facto, according to Salmon, Popper is either a closet inductivist or an irra­tional man. Salmon believes that he has shown that “pure deductivism could not do justice to the problem of rational prediction in the contexts of practical decision-making.”³⁴ “Pure deductivism,” of course, was never at issue since Popper explicitly accepted realism on nondeductive grounds.

As an alternative to Salmon's critique of Popper, we could argue that Salmon has simply defined science as inherently inductivist because it depends on re­alism, and, he believes, realism implies inductivism. In this case, his argument would be a tautology which, you'll recall, tells us about language and logic but not about the world.

This is the point in the debate where the vast majority of scientists in the au­dience would quietly drift away back to their labs and do something useful, but as these issues arise in the corroboration-confirmation dilemma (Chapter 2.G.5) that is directly relevant to scientific thinking, we'll stick with it a bit more. There are two big questions still left to address: Does the past success of a hypothesis somehow strengthen it? What is it about the past success of a hypothesis that let's us base practical actions (e.g., technology) on it, if we don't believe in the power of “inductive reasoning?”

3.

Let's look again at the textbook issue of the bridge designs—you've got to choose between a well-tested one and an untested one. Popper says that the validity of a well-tested hypothesis is not strengthened when it passes experimental tests, yet he still thinks you should choose the best-corroborated hypothesis. For this, philosophers accuse him of irrationality. But Popper argues that, because you know more about the best-tested design, there is no policy more rational than choosing it. Salmon counters that, without accepting the proposition that the design has been strengthened by confirmation, there is no policy more rational than the untested design, and thus there is kind of “tie” between the policies—a stand-off and no rational way to choose between them.

To break the tie, we need to look at Popper's hypothesis-vetting program more closely. When you propose a hypothesis, you are implicitly proposing that it is true. If you have tested and not refuted it, you have found absolutely no reason to change your mind. It's still true. If you are forced to make a choice between the best-corroborated design or an entirely untested one, you have only their past records to go on. You must choose based on what you know and not on what you “predict.” You'll build the bridge according to what you know about. Predictions and reliability are concepts that go beyond what you know now, to the future which is (wait for it) uncertain, so you can't know about it. Popper is not an in- ductivist of any kind.

Popper's argument is admittedly very hard to swallow at first, in part because induction comes so naturally to us (see Chapter 11) and because we immediately imagine the social or political fallout that would occur if a bridge builder pub­licly admitted that he was not 100% certain the bridge would work perfectly. And philosophical critics say that Popper's argument is unacceptable: that without ad­ditional inductive assurances that “it will work,” they themselves could not make a choice.

3. C.2.b Hypotheses as a Basis for Practical Action

A major problem is that, as humans, our choices are generally more or less col­ored by emotion, and we have trouble separating the emotional and cognitive parts of our decisions. For instance, another philosophical critic who chides Popper for not stating that hypotheses are strengthened by experience claims that Popper gives us no reason to believe in the well-corroborated law of gravity. If you were told about an untested hypothesis that says you'll be OK if you jump off the Leaning Tower of Pisa, he says, we wouldn't know what to do without inductive reasoning, Actually, even preverbal toddlers who have never experi­enced a fall are not tempted to test the law: ecological rationality kicks in to save them. Our healthy fear of heights is embedded in our genome. Confusingly, so is our need to justify our behavior, so when we instinctively decide against testing the hypothesis of gravity with our own bodies, we create a reason—induction!— for not doing so.

Even a coldly analytical scientist wouldn't be tempted to jump and wouldn't have to rely on inductive reasoning in making her decision. Why? As a scien­tist, she would understand the law of gravity has been postulated to be a true statement (within the terrestrial realm of falling bodies), and, as it has been well- corroborated by past experience, she has no reason to think that it isn't true. Since the law of gravity applied to falling human bodies predicts a bad outcome if she were to jump, she'd be unlikely to do so. Enumerative induction plays no part in her reasoning.

Philosophers such as Salmon think the difficulty is in finding a link between the search for valid hypotheses—what I've been calling “basic science”—and technology, “applied science.” He is correct in thinking that you behave differ­ently in the two realms, but he mischaracterizes the practical problem.

3. C.2.c How Do Popper's Methods Mesh with “Levels” of Scientific Explanation?

It may have struck you that the concept of “levels” of scientific analysis (Chapter 1) pose a potentially serious challenge for Karl Popper's philosophy. How can we square the concept of “Truth” in science with the possibility that questions have distinct answers at several levels of inquiry? Is there a whole bat­tery of truths, one for each level, and, if so, where does that leave the overarching search for Truth and the falsification method?

Let's start with falsification: Can a hypothesis be falsified at one level and not at another? Obviously, in a practical sense it can: we've noted the case of Newton's law of gravity. At the deepest levels of understanding, the law has been falsified by the revelations of modern physics beginning with Einstein. But Newton's law still works well for countless purposes: if you want to launch a satellite into orbit around earth, Newton's law is what you need. It works, so it must be true³⁵ at that level. Has it been falsified by modern physics or not?

Once again, the answer depends on whether you're doing basic (Popper's word is “pure”) science or applied science. When the Nobel Prize-winning phys­icist Werner Heisenberg remarked that “we do not say [Classical] mechanics is false,” but rather that “Classical mechanics... is everywhere exactly ‘right' where its concepts can be applied.”³⁶ At first, Popper bristles. Among other sins, he feels that Heisenberg's reasoning would be a disaster for pure science because you could always turn to “ad hoc for rescuing a physical theory that was in danger of being falsified”; for example, you could decide that the theory was not actu­ally false, just inapplicable to the case in question. Nevertheless, at the end of his argument Popper, I think grudgingly, allows Heisenberg's reasoning, which is “like that of applied science” when it comes to the “success of applications.” Even though he still doesn't like Heisenberg's approach, as long as it is confined to the domain of “application,” away from the search for Truth, he can live with it.

In a way, the dispute about levels of science is linked to the inadequacy of pure “prediction” to act as a test of a theory. A bad theory might happen to make a good prediction, and so success in predictions doesn't guarantee that a theory is correct. Meanwhile, in the real world of action, good predictions can undeniably be useful wherever they come from, even rejected hypotheses. Our theory of the slipperiness of ice may be ever so flawed, but as it predicts that Aunt Minnie will be safer if we put sand down on the ice, the hypothesis is useful. When it's time to take action, Popper allows a standard that appeals to the utility of a corroborated hypothesis—“try it and see”—in place of severe testing, and his pragmatic stance connects his thinking to levels of science.

3. C.2.d Does the Observation of Black Swans Prove Anything?

Popper's famous, ifoverly simplistic, argument against induction is that, although observing any number of white swans cannot prove the truth of the statement, “all swans are white,” you can disprove (i.e., falsify) that statement by observing one black swan (repeatedly with proper controls, etc.). Deborah Mayo, who feels that Popper “came up short” in his logical analyses, argues that that we can ex­tract more information from observations of black swans and analogous falsi­fying tests of hypotheses than the mere fact that our hypothesis was incorrect.

Assume, she says, that you're testing the hypothesis, H1, “all swans are white,” and an alternative hypothesis, H2 “some swans are not white.” Mayo asserts that observing a black swan not only falsifies H1 but “proves” H2³⁷ deftly turning Popper's reasoning inside out: if we can't prove a hypothesis by obtaining confirming evidence for it, can we at least prove a contrary hypothesis by disconfirming the first one? If her argument were universally applicable, it would knock the foundations right out from under Popper's philosophy! His main premise would be wrong—science could prove some hypotheses after all, we'd have more than corroborated hypotheses, etc.,—and the rationale for Conjectures and Refutations would collapse like a house of cards.

However, before giving up on Popper, we need to examine Mayo's argument. Superficially, it seems valid: the existence of a black swan would certainly mean that some (in logic, “some” means “at least one”) swans are not white. The ques­tion is, can this conclusion benefit scientists generally? I am afraid that it can't. The statement about swans as it stands is a description (a “there is” statement in Popper's terms) not a hypothesis, and Mayo's statistical argument depends on its being a description. However, reasoning about descriptions doesn't generalize to scientific hypotheses, which are explanations. If you interpret the statement about swans as a hypothesis, then her argument is no longer valid. We'll examine both points.

Let's assume that the implicit assumption about classifying swans that both Popper and Mayo make is valid; namely, you can classify swans unambiguously as “white” or “nonwhite” and that these two categories are mutually exclusive and exhaustive: each swan can be put into only one color bin, and all swans can go into one of the two bins. Mayo's argument is sound: if you can falsify H1 (“all swans are white”) for one swan, you simultaneously prove H2 (“not all swans are white”). So far so good, but what does this argument hold for science?

Scientific hypotheses are explanations for phenomena, and we never know if any two genuine scientific hypotheses are mutually exclusive and exhaustive as possible explanations for a phenomenon. Hence, if you're a fallibilist, then even falsifying one hypothesis does not entitle you to conclude that you've “proved” another one; there will always be other possible explanations and unforeseen complexities.

Let's look at the swan problem as a scientific hypothesis. We first have to ask what swans are. After all, if the large black bird is not a swan, all bets are off; finding one tells you nothing about swans. If you guessed that all things called swans are members of the same species of bird, you'd be wrong; there are six or seven³⁸ species in the genus Cygnus. A familiar white swan would be the trum­peter swan, Cygnus buccinator, while the Australian black swan is Cygnus atratus. So white swans and black swans are not classified as members of the same spe­cies! We reevaluate the hypothesis: Was it about swans as a species or swans as a genus? (Actually, white and black swans are classed as members of different “subgenuses,” but never mind.)

You can't begin to answer these questions without knowing precisely what “species” and “genus” mean. At this point, you might be surprised to learn that these terms aren't unambiguously defined despite decades of attempts to do so. A variety of properties ranging from reproductive compatibility to behav­ioral traits, ecological niches, and genetic make-up have all failed to yield a con­sistent, uniform picture of what a species is. And the notion of “genus” is fuzzier still; there is an active debate as to whether a genus is a natural category at all. (The discussions of the biological complexities involved by, for example, Ernst Mayr³⁹ and Peter Godfrey-Smith,⁴⁰ are fascinating and well worth reading.) Nevertheless, without going further, we can see that falsifying, “All swans are white” is not trivial, and it does not entitle you to conclude that you've proved anything.

3.