Theory and Reality

by Peter Godfrey-Smith

  For Popper, theories have the form of generalizations, and they take risks by prohibiting certain kinds of particular events from being observed. If we believe that all pieces of iron, of whatever size and shape, expand when heated, then our theory forbids the observation of something that we know to be a piece of iron contracting when heated. A problem may have occurred to you: how sure can we be that, if we see a piece of "iron" contracting when heated, that it is really iron? We might also have doubts about our measurements of the contraction and the temperature change. Maybe the generalization about iron expanding when heated is true, but our assumptions about the testing situation and our ability to know that a sample is made of iron are false.

  This problem is a reappearance of an issue discussed in chapter 2: holism about testing. Whenever we try to test a theory by comparing it with observations, we must make a large number of additional assumptions in order to bring the theory and the observations into "contact" with each other. If we want to test whether iron always expands when heated, we need to make assumptions about our ability to find or make reasonably pure samples of iron. If we want to test whether the amounts of the bases C and G are equal and the amounts of A and T are equal in all samples of DNA (Chargaff's rules), we need to make a lot of assumptions about our chemical techniques. If we observe an unexpected result (iron contracting on heating, twice as much C as G in a sample of DNA), it is always possible to blame one of these extra assumptions rather than the theory we are trying to test. In extreme cases, we might even claim that the apparent observation was completely misunderstood or wrongly described by the observers. Indeed, this is not so uncommon in our attempts to work out what to make of reports of miracles and UFO abductions. So how can we really use observations to falsify theories in the way Popper wants?

  This is a problem not just for Popper's solution to the demarcation problem, but for his whole theory of science as well.

  Popper was well aware of this problem, and he struggled with it. He regarded the extra assumptions needed to connect theories with testing situations as scientific claims that might well be false-these are conjectures too. We can try to test these conjectures separately. But Popper conceded that logic itself can never force a scientist to give up a particular theory, in the face of surprising observations. Logically, it is always possible to blame other assumptions involved in the test. Popper thought that a good scien tist would not try to do this; a good scientist is someone who wants to expose the theory itself to tests and will not try to deflect the blame.

  Does this answer the holist objection? What Popper has done is to move from describing a characteristic of scientific theories to describing a characteristic of scientific behavior. In some ways this is a retraction of his initial aim, which was to describe something about scientific theories themselves that makes them special. That is a problem. Then again, this shift to describing scientific modes of thought and behavior, rather than theories, may well be a step forward. This will be discussed in more detail in section 4.6.

  Popper also accepted that we cannot be completely certain about the observation reports that we use to falsify theories. We have to regard the acceptance of an observation report as a "decision," one that is freely made. Once we have made the decision, we can use the observation report to falsify any theory that conflicts with it. But for Popper, any falsification process is based, in the end, on a decision that could be challenged. Someone might come along later and try to show, via more testing, that the observation report was not a good one-that person might investigate whether the conditions of observation were misleading. That testing has the same conjecture-and-refutation form described earlier. So this investigation into the controversial observation ultimately depends on "decisions" too.

  Is this bad news for Popper? Popper insisted that making these decisions about single observations is very different from making free decisions directly about the theories themselves. But what sort of difference is this? If observation reports rest on nothing more than "decisions," and these determine our choice of theories, how is that better than directly choosing the theories themselves, without worrying about observation? Or why couldn't we just "decide" to hang onto a theory and reject the observation reports that conflict with it? I am not saying that we should do these things, just that Popper has not given us a good reason not to do them. I believe that we should not do these things because we have good reason to believe that observation is a generally reliable way of forming beliefs. As I will argue in chapter io, we need to make use of a scientific theory of perception at this point in the story. But that argument will have to come later. Popper himself does not try to answer these questions by giving an argument about the reliability of perception.

  This point about the role of decisions affects Popper's ideas about demarcation as well as his ideas about testing. Any system of hypotheses can be held onto despite apparent falsification, if people are willing to make certain decisions. Does that mean that Popper's theory fails to differentiate between science and pseudo-science after all? The answer is "yes and no." The yes comes from the fact that scientific theories can be handled in a way that makes them immune to falsification, and nonscientific theories can be rejected if people decide to accept claims about particular matters that are incompatible with the theory. But there is a "no" part in the answer as well. A scientific theory is falsifiable via a certain kind of decision-a decision about an observation report. A pseudo-scientific theory, Popper says, does not clash with any possible observations. So if a pseudo-scientific theory is to be rejected, some different kind of decision must be made. We can accept, with Popper, that this is a significant difference. But Popper has not told us why this way of doing things, the scientific way, is more rational than some other way.

  I have been fairly tough on Popper's views about falsification in this section, and there is another problem to discuss as well. The problem is bad for Popper, but I should emphasize that it is bad for many others as well.

  What can Popper say about theories that do not claim that some observation 0 is forbidden, but only that it is very unlikely? If I believe that a certain coin is "fair," I can deduce from this hypothesis various claims about the probabilities of long "all heads" or "all tails" sequences of tosses. Suppose I observe zoo tosses turning up heads zoo times. This is very unlikely according to my hypothesis about the coin, but it is not impossible. Any finite stretch of heads tosses is possible with a fair coin, although longer and longer runs of heads are treated by the theory as more and more unlikely. But if a hypothesis does not forbid any particular observations, then, according to Popper, it is taking no risks. That seems to entail that theories that ascribe low probabilities to specific observations, but do not rule them out altogether, are unfalsifiable and hence unscientific for Popper.

  Popper's response was to accept that, logically speaking, all hypotheses of this kind are unscientific. But this seems to make a mockery of the important role of probability in science. So Popper said that a scientist can decide that if a theory claims that a particular observation is extremely improbable, the theory in practice rules out that observation. So if the observation is made, the theory is, in practice, falsified. According to Popper, it is up to scientists to work out, for their own fields, what sort of probability is so low that events of that kind are treated as prohibited. So probabilistic theories can only be construed as falsifiable in a special "in practice" sense. And we have here another role for "decisions" in Popper's philosophy of science, as opposed to the constraints of logic.

  Popper is right that scientists reject theories when observations occur which the theory says are highly improbable (although it is a complicated matter which kinds of improbability have this importance). And Popper is right that scientists spend a good deal of time working out "how improbable is too improbable." Complex statistical methods are used to help sci entists with these decisions. But in making this move, Popper has badly damaged his original picture of science. This was a picture i
n which observations, once accepted, have the power to decisively refute theoretical hypotheses. That is a matter of deductive logic, as Popper endlessly stressed. Now Popper is saying that falsification can occur without its being backed up by a deductive logical relation between observation and theory.

  4.5 Objections to Popper on Confirmation

  As described earlier, Popper believed that theories are never confirmed by observations, and he thought inductive arguments are never justified. Popper thought that a theory of the rational choice of theories could be given entirely in terms of falsification, so he thought that rejecting induction and confirmation was no problem.

  In the previous section I discussed problems with Popper's views about falsification. But let us leave those problems aside now, and assume in this section that we can use Popperian falsification as a method for decisively rejecting theories. If we make this assumption, is Popper's attempt to describe rational theory choice successful? No, it is not.

  Here is simple problem that Popper has a very difficult time with. Suppose we are trying to build a bridge, and we need to use physical theories to tell us which designs are stable and will support the weight that the bridge must carry. This is a situation where we must apply our scientific theories to a practical task. As a matter of fact, engineers and scientists in this situation will undoubtedly tend to use physical theories that have survived empirical testing; they will use "tried and true" methods as far as possible. The empiricist approach to the philosophy of science holds that such a policy is rational. The problem for an empiricist philosophy is to explain in more detail why this policy is the right one. That task is hard, as I hope became clear in chapter 3. But let us focus on Popper, who wants to avoid the need for a theory of confirmation. How does Popper's philosophy treat the bridge-building situation?

  Popper can say why we should prefer to use a theory that has not been falsified over a theory that has been falsified. Theories that have been falsified have been shown to be false (here again I ignore the problems discussed in the previous section). But suppose we have to choose between (z) a theory that has been tested many times and has passed every test, and (2) a brand new theory that has just been conjectured and has never been tested. Neither theory has been falsified. We would ordinarily think that the rational thing to do is to choose the theory that has survived testing. But what can Popper say about this choice? Why exactly would it be irrational, for Popper, to build the bridge using a brand new theory that has never been tested?

  Popper recognized and struggled with this problem too. Perhaps this has been the most common objection to Popper from other empiricist philosophers (e.g., Salmon 119811). Popper is not able to give a very good reply.

  Popper refuses to say that when a theory passes tests, we have more reason to believe that the theory is true. Both the untested theory and the welltested theory are just conjectures. But Popper did devise a special concept to use in this situation. Popper said that a theory that has survived many attempts to falsify it is "corroborated." And when we face choices like the bridge-building one, it is rational to choose corroborated theories over theories that are not corroborated.

  What is "corroboration"? Popper gave a technical definition and held that we can measure the amount of corroboration that a theory has at a particular time. The technicalities do not matter, though. We need to ask, What sort of property is corroboration? Has Popper just given a new name to confirmation? If so, he can answer the question about building the bridge, but he has given up one of his main differences from the logical empiricists and everyone else. If corroboration is totally different from confirmation-so different that we cannot regard corroboration as any guide to truth-then why should we choose a corroborated theory when we build the bridge? This issue has been much discussed (see Newton-Smith 1198r). Popper's concept of corroboration can be interpreted in a way that makes it different from confirmation, but Popper can give no good answer to why we should choose corroborated theories over new ones when building bridges.

  To understand corroboration, think of the difference between an academic transcript and a letter of recommendation. This distinction should be vivid to students! An academic transcript says what you have done. It measures your past performance, but it does not contain explicit predictions about what you will do in the future. A letter of recommendation usually says something about what you have done, and it also makes claims about how you are likely to do in the future. Confirmation, as understood by the logical empiricists, is something like a letter of recommendation for a scientific theory. Corroboration, for Popper, is only like an academic transcript. And Popper thought that no good reasons could be given for believing that past performance is a reliable guide to the future. So corroboration is entirely "backward-looking." Consequently, no reason can be given for building a bridge with a corroborated theory rather than a noncorroborated but unfalsified one.

  I think the best thing for Popper to say about the bridge-building situation is to stick to his inductive skepticism. He should argue that we really don't know what will happen if we build another bridge with a design that has worked in the past. Maybe it will stay up and maybe it won't. There might also be practical reasons for choosing that design if we are very familiar with it. But if someone comes along with a brand new untested design, we won't know whether it's a bad design until we try it.

  Popper liked to say here that there is no alternative policy that is more rational than using the familiar and welltested design, and we do have to make some decision. So we can go ahead and use the established design. But as Wesley Salmon (1981) replied, this does not help at all. If confirmation does not exist, then it seems there is also no policy that is more rational than choosing the untested design. All we have here is a kind of "tie" between the options.

  For most people, this is an unsatisfactory place for a philosophy of science to end up. Inductive skepticism of this kind is hard to take seriously outside of abstract, academic discussion. However, the efforts of the last two hundred years have shown how extremely hard it is to produce a good theory of induction and confirmation. One of the valuable roles of Popper's philosophy is to show what sort of theory of science might be possible if we give up on induction and confirmation.

  In the first chapter of this book, I said that few philosophers still try to give descriptions of a definite "scientific method," where this is construed as something like a recipe for science. Popper is something of an exception here, since he does come close to giving a kind of recipe (although Popper insists there is no recipe for coming up with interesting conjectures). His view has an interesting relationship to descriptions of scientific method given in science textbooks.

  In many textbooks, one finds something called the "hypotheticodeductive method." Back in chapter 3, I discussed a view about confirmation that is often called "hypotheticodeductivism." Now we are dealing with a method rather than a theory of confirmation. Science textbooks are more cautious about laying out recipes for science than they used to be, but descriptions of the hypotheticodeductive method are still fairly common. Formulations of the method vary, but some are basically a combination of Popper's view of testing and a less skeptical view about confirmation. In these versions, the hypotheticodeductive method is a process in which scientists come up with conjectures and then deduce observational predictions from those conjectures. If the predictions come out as the theory says, then the theory is supported. If the predictions do not come out as the theory says, the theory is not supported and should be rejected.

  This process has the basic pattern that Popper described, but the idea that theories can be "supported" by observations is not a Popperian idea. The term "support" is vague, but I think discussions of the hypotheticodeductive method generally assume that if a theory makes a lot of successful predictions, we have more reason to believe that the theory is true than we had before the successful predictions were made. We will never be completely sure, of course. But the more tests a theory passes, th
e more confidence we can have in its truth. The idea that we can gradually increase our confidence that a theory is true is an idea that Popper rejected. As I said at the start of this chapter, some of Popper's scientific admirers do not realize that Popper's view has this feature, because some of Popper's discussions were misleading.

  Other formulations of the hypotheticodeductive method include a first stage in which observations are collected and a conjecture is generated from these observations. Popper disagreed with this picture of scientific procedure because he argued that fact-gathering can only take place in a way guided by a conjecture. But this is a fairly minor point.

  Another term that some textbooks use in discussing scientific method (though not so much any more) is "strong inference." This term was introduced by a chemist named John Platt (1964). Strong inference is roughly a Popperian kind of testing plus another further assumption, which Popper rejected. This assumption is that we can write down all the possible theories that might be true in some area, and test them one by one. We find the true theory by eliminating the alternatives-it's a kind of "Sherlock Holmes" method. For Popper, this is impossible. In any real case, there will be an infinite number of competing theories. So even if we eliminate ten or one hundred possibilities, there is still the same infinite number remaining. According to Popper, all we can do is to choose one theory, test it, then choose another, and so on. We can never have confidence that we have eliminated all, or most, of the alternatives. (More recent attempts to make use of this "Sherlock Holmes" method will be discussed in chapter 14.)

  I have not discussed objections to Popper's theory of scientific change (section 4.3) yet, but I will do so in the next few chapters.