Theory and Reality

by Peter Godfrey-Smith

  The version of holism that Quine defended in "Two Dogmas" was an extreme one. It included an attack on the one idea in the previous section that you might have thought was completely safe: the analyticsynthetic distinction. Quine argued that this distinction does not exist; this is another unjustified "dogma" of empiricism.

  Here again, some of Quine's arguments were directed at a version of the analyticsynthetic distinction that the logical positivists no longer held. Quine said that the idea of analyticity was intended to treat some claims as immune to revision, and he argued that in fact no statement is immune to revision. But Carnap had already decided that analytic statements can be revised, though they are revised in a special way. A person or community can decide to drop one whole linguistic and logical framework and adopt another. Against the background provided by a given linguistic and logical framework, some statements will be analytic and hence not susceptible to empirical test. But we can always change frameworks. By the time that Quine was writing, Carnap's philosophy was based on a distinction between changes made within a linguistic and logical framework, and changes between these frameworks.

  In another (more convincing) part of his paper, Quine argued that there is no way to make scientific sense of a sharp analyticsynthetic distinction. He connected this point to his holism about testing. For Quine, all our ideas and hypotheses form a single "web of belief," which has contact with experience only as whole. An unexpected observation can prompt us to make a great variety of possible changes to the web. Even sentences that might look analytic can be revised in response to experience in some circumstances. Quine noted that strange results in quantum physics had suggested to some that revisions in logic might be needed.

  In this discussion of problems for logical positivism, I have included some discussions that started early and some that took place after World War II, when the movement had begun its U.S.-based transformation. Let us now look at some central ideas of logical empiricism, the later, less aggressive stage of the movement.

  2.5 Logical Empiricism

  Let's see how things looked in the years after World War II. Schlick is dead, and other remnants of the Vienna Circle are safely housed in American universities-Carnap at Chicago, Hempel at Pittsburgh and then Princeton, Reichenbach at UCLA (via Turkey), Feigl at Minnesota. Many of the same people are involved, but the work is different. The revolutionary attempt to destroy traditional philosophy has been replaced by a program of careful logical analysis of language and science. Discussion of the contributions that could be made by the scientific worldview to a democratic socialist future have been dropped or greatly muted. (Despite this, the FBI collected a file on Carnap as a possible Communist sympathizer.)

  As before, ideas about language guided logical empiricist ideas about science. The analyticsynthetic distinction had not been rejected, but it was regarded as questionable. The logical empiricists felt the pressure of Quine's arguments. The verifiability theory, which had been so scythe-like in its early forms, was replaced by a holistic empiricist theory of meaning. Theories were seen as abstract structures that connect many hypotheses together. These structures are connected, as wholes, to the observable realm, but each bit of a theory-each claim or hypothesis or conceptdoes not have some specific set of observations associated with it. A theoretical term (like "electron" or "gene") derives its meaning from its place in the whole structure and from the structure's connection to the realm of observation.

  Late in the logical empiricist era, in z97o, Herbert Feigl gave a pictorial representation of what he called "the orthodox view" of theories (see fig. z.z). A network of theoretical hypotheses ("postulates") is connected by stages to what Feigl calls the "soil" of experience. This anchoring is the source of the network's meaning. Feigl used this picture to describe a single scientific theory. For the more extreme holism of Quine, a person's total set of beliefs form a single network.

  The logical positivist distinction between observational and theoretical parts of language was kept roughly intact. But the idea that observational language describes private sensations had been dropped. The observational base of science was seen as made up of descriptions of observable physical objects (though Carnap thought it might occasionally be useful to work with a language referring to sensations).

  Logical positivist views about the role of logic in philosophy and about the sharp separation between the logic of science and the historicalpsychological side of science were basically unchanged. A good example of the kind of work done by logical empiricists is provided by their work on explanation in science (see especially Hempel and Oppenheim 1948; Hempel 1965). For Hempel, to explain something is to show how to infer it using a logical argument, where the premises of the argument include at least one statement of a natural law (see chapter 13 below). This illustrates the idea, common to logical positivism and logical empiricism, that logic is the main tool of philosophy of science.

  Fig. 2.2

  Feigl's picture of the logical empiricist view of theories (From Feigl 1970; reproduced courtesy of University of Minnesota Press)

  We saw that logical positivism held that the sole aim of science is to track patterns in experience. For logical positivism, when a scientist seems to be trying to describe unobservable structures in the world that give rise to what we see, the scientist must instead be seen as describing the observable world in a special, abstract way. Scientific language is only meaningful insofar as it picks out patterns in the flow of experience. Now, does logical empiricism make the same claim? Does logical empiricism claim that scientific language ultimately only describes patterns in observables?

  The answer is that logical empiricists agonized over this. In their hearts their answer was yes, but this answer seemed to get harder and harder to defend. Carl Hempel wrote a paper in 1958 called "The Theoretician's Dilemma;' which was the height of logical empiricist agony over the issue. As a fairly traditional empiricist, Hempel was attracted to the idea that the only possible role for those parts of language that seem to refer to unobservable entities is to help us pick out patterns in the observable realm. And if the parts of theories that appear to posit unobservable things are really any good, this "goodness" has to show up in advantages the theory has in its handling of observables. So there is no justification for seeing these parts of scientific language as describing real objects lying beyond experience. But Hempel and the logical empiricists found themselves forced to concede that this view does not make much sense of actual scientific work. When scientists use terms like "electron" or "gene," they act as if they are doing more than tracking complex patterns in the observable realm. But the idea that the logical empiricists were being pushed toward-the idea that scientific theories are aimed at describing unobservable real structureswas hard to put on the table and defend. Empiricist philosophy of language seemed implacably opposed to it.

  Empiricists were familiar with bad versions of the idea that behind the ordinary world of observables there is a special and superior realm, pure and perfect. This "layered" view of reality seemed to empiricists a source of endless trouble, right from the time of the ancient Greek philosopher Plato, who distinguished the illusory, unstable world of "appearances" from the more perfect and real world of "forms." Empiricists have rightly been determined to avoid this kind of picture. But much of science does appear to be a process in which people hypothesize hidden structures that give rise to observable phenomena. These hidden structures are not "pure and perfect" or "more real" than the observable parts of the world, but they do lie behind or beneath observable phenomena. Of course, unobservable structures posited by a theory at one time might well turn out to be observable at a later time. In science, there is no telling what kinds of new access to the hidden parts of the world we might eventually achieve. But still, much of science does seem to proceed by positing entities that are, at the time of the research in question, truly hidden. For the traditional empiricist philosopher, understanding scientific theorizing in a way that posits a layer
of observable phenomena and a layer of hidden structure responsible for the phenomena takes us far too close to bad old philosophical views like Plato's. We are too close for comfort, so we must give a different kind of description of how science works.

  The result is the traditional empiricist insistence that, ultimately, the only thing scientific language can do is describe patterns in the observable realm. In the first published paper that introduced logical positivism, Carnap, Hahn, and Neurath said: "In science there are no `depths'; there is surface everywhere" ([19291 1973, 306). This is a vivid expression of the empiricist aversion to a view in which the aim of theorizing is to describe hidden levels of structure. Science uses unusual theoretical concepts (which look initially like attempts to refer to hidden things) as a way of discovering and describing subtle patterns in the observable realm. So the logical positivists and the logical empiricists talked constantly about prediction as the goal of science. Prediction was a substitute for the more obviouslooking-but ultimately forbidden-goal of describing the real hidden structure of the world.

  Twentiethcentury empiricism made an important mistake here. We can make sense of science only by treating much of it as an attempt to describe hidden structures that give rise to observable phenomena. This is a version of scientific realism, an idea that will be discussed later in this book. In science there are depths. There is not a simple and fixed distinction between two "layers" in nature-the empiricists were right to distrust this idea. Instead there are many layers, or rather a continuum between structures that are more accessible to us and structures that are less accessible. Genes are hidden from us in some ways, but not as hidden as electrons, which in turn are not as hidden as quarks. Although there are "depths" in science, what is deep at one time can come to the surface at later times, and there may be lots of ways of interacting with what is presently deep.

  2.6 On the Fall of Logical Empiricism

  Logical empiricist ideas dominated much American philosophy, and they were very influential elsewhere in the English-speaking world and in some parts of Europe, in the middle of the twentieth century. But by the mid-196os the view was definitely under threat; and by the middle or late 1970s, logical empiricism was near to extinction. The fall of logical empiricism was due to several factors, all of which I have either introduced in this chapter or will discuss in later chapters. One is the breakdown of the view of language that formed the basis of many logical positivist and logical empiricist ideas. Another is pressure from holist arguments. A third is the frustrating history of attempts to develop an inductive logic (chapter 3). A fourth is the development of a new role for fields like history and psychology in the philosophy of science (chapters 5-7). And eventually there was pressure from scientific realism. But this was only possible after logical empiricism had begun to decline.

  Further Reading

  For much more on the empiricist tradition in general, see Garrett and Barbanell, Encyclopedia of Empiricism (1997).

  Schlick's "Positivism and Realism" (1932-33) and Feigl's "Logical Empiricism" (1943) are good statements of logical positivism by original members of the Vienna Circle. (Feigl uses the term "logical empiricism," but his paper describes a fairly strong, undiluted version of the view.) Ayer's Language, Truth, and Logic (1936) is readable, vivid, and exciting. Some see it as a distortion of logical positivist ideas.

  The Routledge Encyclopedia of Philosophy (1998) has an interesting collection of articles, especially in the light of new debates about the history of logical positivism. The article on logical positivism is by Friedman and reflects his somewhat unorthodox reading (de-emphasizing the empiricist tradition). Stadler's entry on the Vienna Circle gives a more traditional view. See also Creath's entry on Carnap. On all these issues, see also the essays in Giere and Richardson 1997.

  Peter Galison's AufbauBauhaus" (1990) is a wonderful account of the artistic, social, and political interests of the logical positivists and the links between these interests and their philosophical ideas. Passmore 1966 is a good and accessible survey of philosophical movements and trends in the late nineteenth and early twentieth centuries, including absolute idealism.

  Hempel, Aspects of Scientific Explanation (1965), is the definitive statement of logical empiricism. His Philosophy of Natural Science (1966) is the easy version. Carnap's later lectures have been published as Introduction to the Philosophy of Science (1995).

  An attempt to revive some logical positivist ideas has recently begun; see, for example, Elliott Sober's forthcoming book Learning from Logical Positivism.

  3.1 The Mother of All Problems

  In this chapter we begin looking at a very important and difficult problem, the problem of understanding how observations can confirm a scientific theory. What connection between an observation and a theory makes that observation evidence for the theory? In some ways, this has been the fundamental problem in the last hundred years of philosophy of science. This problem was central to the projects of logical positivism and logical empiricism, and it was a source of constant frustration for them. And although some might be tempted to think so, this problem does not disappear once we give up on logical empiricism. The problem, in some form or other, arises for nearly everyone.

  The aim of the logical empiricists was to develop a logical theory of evidence and confirmation, a theory treating confirmation as an abstract relation between sentences. It has become fairly clear that their approach to the problem is doomed. The way to analyze testing and evidence in science is to develop a different kind of theory. But it will take a lot of discussion, in this and later chapters, before the differences between approaches that will and will not work in this area can emerge. The present chapter will mostly look at how the problem of confirmation was tackled in the middle of the twentieth century. And that is a tale of woe.

  Before looking at twentiethcentury work on these issues, we must again look further into the past. The confirmation of theories is closely connected to another classic issue in philosophy: the problem of induction. What reason do we have for expecting patterns observed in our past experience to hold also in the future? What justification do we have for using past observations as a basis for generalization about things we have not yet observed?

  The most famous discussions of induction were written by the eighteenthcentury Scottish empiricist David Hume (117391 1978). Hume asked, What reason do we have for thinking that the future will resemble the past? There is no contradiction in supposing that the future could be totally unlike the past. It is possible that the world could change radically at any point, rendering previous experience useless. How do we know this will not happen? We might say to Hume that when we have relied on past experience before, this has turned out well for us. But Hume replies that this is begging the question-presupposing what has to be shown. Induction has worked in the past, sure, but that's the past! We have successfully used "past pasts" to tell us about "past futures." But our problem is whether anything about the past gives us good information about what will happen tomorrow.

  Hume concluded that we have no reason to expect the past to resemble the future. Hume was an "inductive skeptic." He accepted that we all use induction to make our way around the world. And he was not suggesting that we stop doing so (even if we could). Induction is psychologically natural to us. Despite this, Hume thought it had no rational basis. Hume's inductive skepticism has haunted empiricism ever since. The problem of confirmation is not the same as the classical problem of induction, but it is closely related.

  3.2 Induction, Deduction, Confirmation, and Explanatory Inference

  The logical empiricists tried to show how observational evidence could provide support for a scientific theory. The idea of "support" is important here; there was no attempt to show that scientific theories could be proved. Error is always possible, but evidence can support one theory over another.

  The cases that were to be covered by this analysis included the simplest and most traditional cases of induction: if we see a m
ultitude of cases of white swans, and no other colors, why does that give us reason to believe that all swans are white? But obviously not all cases of evidence in science are like this. The observational support for Copernicus's theory that the earth goes around the sun, or for Darwin's theory of evolution, seems to work very differently. Darwin did not observe a set of individual cases of evolution and then generalize.

  The logical empiricists wanted a theory of evidence, or "theory of confirmation," that would cover all these cases. They were not trying to develop a recipe for confirming theories. Rather, the aim was to give an account of the relationships between the statements that make up a scientific theory and statements describing observations, which make the observations support the theory. You might wonder, at this point, what use there could be for a theory with so distant a relationship to actual scientific behavior. Who cares whether a logical analysis of this kind exists or not? In defense of logical empiricism, we might say this: although scientific behavior is not being directly described by the theory of confirmation, nonetheless scientific procedures might be based on assumptions described in the theory of confirmation. Perhaps scientists do many things that cannot be justified if confirmation does not exist.