G Karl Popper and John Platt

The middle of the twentieth century saw the development of two programs that have had lasting effects on practicing scientists: Critical Rationalism, a program most famously associated with Karl Popper¹⁶ and Strong Inference, by John Platt.¹⁷ Popper was a philosopher who was outside of the philosophical main­stream of his day.

Although philosophers of science acknowledge Popper's sway, by and large, they reject his ideas. In fact, his continued appeal to working scientists frankly puzzles them.²⁰ Hence, philosophers, who mostly ignore scientists' dabblings in philosophy or treat them with polite indulgence, would not be sympathetic to my stance on the hypothesis. As noted in the Introduction, I am a Popperian who strays from strict Popperian orthodoxy from time to time. I think that Popper got the key issues right and that there are many good reasons for scientists to learn about his thinking despite the nuances and objections that philosophers fixate on (Chapter 3 lays out and responds to some of the objections). Popper's program is not as straightforward as scientists often assume it is. I'll bring out a few aspects of his thinking that tend to be overlooked and suggest how to understand them.

The positions of Popper and Platt are especially congenial to doing science, as opposed to contemplating science. For me, the following, which I offer as a fun­damental tenet, essentially an axiom, gets to the heart of the matter.

Scientists will never accept a scientific statement as true if they can think of a fea­sible test that might show that it is not true.

If you believe this, then you, too, may be a Popperian.

2. G.1 Critical Rationalism (Conjectures and Refutations)

The philosophical school that Popper founded is called Critical Rationalism, al­though Popper referred to his program as Conjectures and Refutations, which describes its essential principles.²¹ Conjectures and Refutations is a form of trial-and-error reasoning that encapsulated the fundamental procedure of sci­ence: you conjecture a hypothesis to explain an aspect of the world; try to refute, or falsify, it; and, if you can't refute it, then you may tentatively conclude that you've discovered a fact² While we need to be able to tell if a hypothesis is false, there is no such requirement to tell if it is true. To a nonscientist, this may seem like a glaring omission: If the hypothesis is an important element of scientific reasoning, and if science seeks the Truth, why not do something to prove that a hypothesis is True?

Popper's best- known example, slightly embellished, of how his program works is this: one day, while observing a group of swans in the park, you notice that they are all white, and you wonder why. The hypothesis occurs to you that they are all white because “all swans are white.” (It is a stretch to call this a genuine hypo­thesis, but it is classical and serves the immediate purpose: I'll have more to say about it in Chapter 3.C.2.d.) You realize that you can't test this hypothesis to see if it's true by finding more white swans because, no matter how many you find, you won't ever know if all swans are white unless you observe “all swans” (i.e., all things that were, are, or ever will be, swans).

In short, rather than trying to solve the problem of induction, Popper revolted against the very concept of philosophical induction. He was impressed that scientists were able to arrive at powerful “laws” that sustained practical actions and made further research advances possible, even though the present laws might eventually need to be replaced by improved laws. And Popper stressed that inductive observations themselves could not “lead to” knowledge; know­ledge comes from people applying their minds to understand the world, not from the world itself. In glossing over the role of the mind, inductivists missed the main story.

Popper considered the question of how knowledge could increase in the ab­sence of induction. The Logical Positivist philosophers had argued that science should only admit or include statements that can be shown to be true,²³ they sought justification (or confirmation or verification or validation) for scientific statements. While at first glance their objective seems eminently worthwhile, the impossibility of guaranteeing that scientific statements are True (i.e., of demon­strating their validity beyond any conceivable doubt) is exactly the fundamental philosophical problem of science: we remain uncertain that our “truths” are True. Strict justification^ programs are doomed to disappointment, and Popper worked to find a way of explaining the growth of knowledge without them.

2. G.2 Elimination of Induction

One of Popper's chief objectives was to “eliminate induction from science,” but his goal is frequently mischaracterized.²⁴ He was not so much concerned with in­duction as a means of generating scientific hypotheses.²⁵ The origin of concepts was a problem for psychologists—or, nowadays, cognitive scientists—to solve, not philosophers.

Popper was a committed fallibilist who took the concept that we are eternally uncertain seriously, even though he recognized that it can be awkward to do so. For instance, many of us have trouble discussing scientific findings that we “know” to be true (global warming is real!), together with the uncertain state of our knowledge. (The discussion of levels of explanation and Aunt Minnie and the ice in Chapter 1 show one way to resolve this paradox. We'll come back to it in Chapter 3.) Nevertheless, Popper wanted to ensure to the greatest extent possible that the contents of science (i.e., the facts as we know them) were true.²⁶ If you seek Truth, but are dead set against induction, what do you do?

2. G.3 The Truth About Falsification

Popper's solution was radical: if we can't guarantee the Truth of scientific know­ledge, then we should stop trying to guarantee it and stop worrying that we can't guarantee it. A valid hypothesis must predict at least one experimental outcome that, if it actually occurred, would mean that the hypothesis was wrong (i.e., it would be falsified). Falsifiability, not provability, is the standard that we live by. Testing is not easy though; it must be severe: we have to demand that our hypo­thesis demonstrates its “mettle” by surviving the tests.²⁷ In Chapter 3, we'll dis­cuss fall-back positions (e.g., “probabilistic truth”) and see why, although it has its place in science, it is unsatisfactory as a final standard for basic science.

We must, says Popper, forget trying to prove that a hypothesis is True and ac­cept that the best we can hope for is to avoid accepting hypotheses that are false. We have to be satisfied if a hypothesis is intended to be True, is falsifiable, and withstands attempts to falsify it. After the testing, we must decide whether to ac­cept the test results (i.e., what they mean, whether they are reliable and unambig­uous).

For Popper, observations must be robust and repeatable (or “reproducible”). Repeatability is crucial for two reasons. You can't test a hypothesis about a one­time observation, such as “a miracle,” because you don't know if it will ever occur again, let alone subject it to tests. And repeatability is important because it allows for intersubjective testing—that is, more than one person can make the observation—and “intersubjectivity” is another definition of “objectivity.” If sci­ence is to be objective, then its observations must be repeatable.

Popper is also clear that observations of all kinds are influenced by our ex­isting hypotheses (they are “theory laden”²⁸). When we examine tissue under the microscope, we do so on the assumption that its microscopic structure holds important secrets. We also assume that we know how microscopes work, that the theories that underlie the microscopes are correct and, therefore, that what they reveal is correct. Finally, we interpret what we see through the microscope in light of what we already know and what we expect to see. And, of course, if we detect an anomaly at any stage of the process—something quite unexpected happens—then we become alert because it may signal that an opportunity to make a big discovery is lurking nearby. Or that we're making a mistake.

Some philosophers ask why being able to find out whether a hypothesis is wrong is such a good thing?^29,30 Philosophy aside, falsifiability is not as coun­terintuitive as it is sometimes made to seem. Remember our car mechanic—call him Bob—uses the same logic in diagnosing the reason that your car engine did not make a sound when you pushed the Start button (or turned the ignition key if it's an older car).

Falsification also solved the “demarcation problem” of the philosophy of science. The question is how can we tell scientific statements apart from non- scientific ones or, more generally, how distinguish science from nonscience? According to Popper, you apply the falsifiability standard. A scientific statement can, in principle, be falsified, it is subject to the possibility of disproof. Therefore, statements that are not disprovable, such as religious statements, are not scien­tific. We'll explore the role of falsifiability and the demarcation problem in the context of the unity of science, which we'll go into in Chapter 4.

The falsifiability criterion removes ambiguity, preventing the intellectual hocus-pocus that allows astrologers or cult leaders to continue to claim that their belief systems are correct despite massive evidence to the contrary. Because they do not set out unambiguous predictions that could falsify their hypotheses, charlatans can rescue their claims no matter what happens. (Unless they give up: when noted end-of-the-world oracle Harold Camping missed in his predic­tion of a global cataclysm on September 6, 1994, he changed the date, first to September 29, 1994; then to October 2, 1994; March 31, 1995; May 21, 2011, and finally, October 21, 2011. After suffering an unexpected brain stroke, Camping admitted that he couldn't get it right and quit the predicting business.³²)

2. G.4 Revising Versus Rejecting Hypotheses

If a hypothesis makes a false prediction, then the hypothesis is falsified and must be rejected. For many people, “rejection” seems harsh and insensitive. They'd like a milder, more nurturing way of treating a clever idea that hasn't lived up to its promise, and they have difficulty in accepting Popper's program because of it. Their attitude may stem from overly negative connotations of rejection. An issue that we'll return to is whether one test is sufficient to falsify a hypothesis—it isn't—and we'll see why later.

In any case, “rejecting” a falsified hypothesis does not mean that you have to purge your brain of all vestiges of it. Assume that your testing was appro­priately severe and repeatable and that you falsified the hypothesis. This only means that, as it was literally stated, the hypothesis was wrong; hence, you re­ject it and need to find a fresh one. The fresh one may be a little or a lot different from the old one; you may start from scratch or build on elements of the falsified hypothesis. If for example, the data can be explained by a revised hypothesis, then it is okay revise it, to derive new predictions, and test them. If Bob tested your battery and found that it was not strictly speaking “dead” (i.e., no longer rechargeable) but merely had a low voltage, then he'd revise his hypothesis to “discharged battery,” recharge it, and begin trying to discover why it ran down. Bob's revision would be legitimate because it would explain the original and the new data in a natural way, and, in fact, Bob could have advanced it originally as an alternative hypothesis.

However, you can't legitimately revise a falsified hypothesis simply by doing an ad hoc patch to rescue it. You can't introduce extraneous elements that were not suggested by any of the data. A good way to avoid making ad hoc patches is to ensure that your revised hypothesis makes predictions that the falsified one did not make. Bob's revised hypothesis predicts that he'll be able to recharge the battery, but he wouldn't be able to recharge a truly dead battery. This procedure is the opposite of what the defenders of the Ptolemaic planetary system were doing by tweaking their epicycle hypothesis with ad hoc fixes, trying to catch up with and explain new anomalies as they cropped up. That is, they retained and embellished their original hypothesis with new details to “rescue” rather than reject it.

It's true that you have to be careful in revising hypotheses and be aware of the slippery slope: if Snowball's biweekly sessions at Puppy School did not help, this would argue that your hypothesis about his hyperkinetic behavior was false. You might then guess that Snowball's problem is that his training did not take place in his normal environment (i.e., your apartment). Technically, this would be a revised hypothesis; it makes a new prediction, and you could go ahead and spend the money to have Snowball's trainer come over and do the training on site. Obviously, you could continue down this road indefinitely. If Snowball's at- home training regimen were unsuccessful, you might be tempted to revise the training deficit hypothesis again: maybe he needs to be trained at home, but only between the hours of 5 and 7 p.m., which is when you usually get home from work. There is, literally, no limit to the number of hypotheses that you can gen­erate. Poor revisions of hypotheses are narrower, less streamlined, and have more ties to a particular case than good revisions. In the end, however, there are only guidelines and no hard and fast rules for deciding when to stick with a falsified hypothesis and revise it or toss it out and start over from scratch.

2. G.4 Tested-and-Not-Falsified? (True, as Far as We Know)

A peculiar sense of dismay seems to grip almost everyone who first encounters Popper's philosophy. How are we supposed to think about hypotheses that have been severely tested and not falsified? Did passing the severe test affect its va­lidity? For Popper, if a hypotheses is tested and not rejected, nothing special happens to it. “Wait a minute!,” you may object, “it seems stronger somehow,” and almost everyone feels the same. Despite our official allegiance to the principle of eternal uncertainty, we expect tested hypotheses to be truer. We'll take up the complicated idea of “probable truth” in the next chapter; for the moment let's remember that the ideal goal of science is the Truth, not “probable truth,” even if we knew what that meant. In the meantime, if we look closely at Popper's anal­ysis, we'll see that the situation is not so complicated.

Popper says that, while you can't know if a tested hypothesis actually is closer to the Truth, it is entirely reasonable to act as if it is. Again, scientific facts (the “contents of science”) are hypotheses that have been severely tested but not falsi­fied.³³ They represent the best information we have.

Putting the matter this way seems awkward, though. Another Critical Rationalist philosopher, David Miller,³⁴ chides Popper for being “squeamish” and not coming right out and saying that we can classify our “tested-and-not- falsified” hypotheses as actually “true.” It would be much easier if he had, and saying so would make sense. Why would this be justified? Miller points out that there is a big difference between “classifying” a hypothesis as “true as far as we know” and “certifying” that it is actually True. We can classify, but not certify. Recall that when you proposed a hypotheses, you were proposing that it is a true explanation; you provisionally classified it as being true before you tested it, and the testing didn't make it false. Ergo, to the best of your knowledge, the hypo­thesis is true.

2. G.5 Corroboration Versus Confidence

Now another confusing point. We can't say that tested-and-not-rejected hypoth­eses are “confirmed” or “verified” (Chapter 1 L.4.), but we do need a way to talk about them. Popper presses a different word into service; he says that

tested-and-not-rejected hypotheses have been corroborated. “Corroborated” is not a synonym for “confirmed” or “verified.” We don't know if a corroborated hypothesis is true; we only know that, at this point, it's not untrue. Corroborated thus merely means “tested severely and not falsified.”

Corroboration is not a trivial concept. We know more about a corroborated hypothesis than we do an untested hypothesis. Popper toys with the notion that we could rank order hypotheses according to the degrees to which they have been corroborated, although nothing much comes of this.³⁵ In any case, corrob­oration refers only to our present state of knowledge; it does not, in a sneaky in- ductivist way, hint anything about the future fitness of the hypothesis, which is, of course, what “confirmation” and the like do. A well-corroborated hypothesis may turn out to be falsified tomorrow. We don't know. In the meantime, we treat it as being provisionally true.

While we might—and I think should—accept this reasoning intellectually, there is no denying that it goes much against our instincts. Intuitively, we feel that a tested-and-not-falsified-hypothesis is better somehow. And we feel an ir­repressible surge of triumph (“Yes! [fist pump], we did it!”) when our nifty hypo­thetical explanation passes a tough experimental test with flying colors. There is nothing wrong with these emotions, as long as we don't mix them up with our objective knowledge. We are not actually any more sure of the Truth of the hypo­thesis today after the experiment than we were yesterday, before the experiment.

Popper's thinking provides the basis of the modern hypothesis. Before fleshing it out more, we should cover the basics of John Platt's program, which adds prac­tical elements to Popper's abstract program.

2.