Philosophical Issues in Scientific Practice

In this first major section, we look briefly at some ways philosophical issues arise in science, with particular emphasis on some issues relevant to the workings of science. There are a large number of such issues, of which we will see just a few.

2.1 Confirmation and disconfirmation

It is a fairly common belief that science proceeds by observing some class of phenomena - for example, the observed movement of the stars and planets across the night sky. And then one generates a theory, perhaps the theory that the earth and other planets move about the sun in elliptical orbits, with the sun occupying one focus of the ellipse. Then one finds ways to test that theory, generally by making predictions about what we would expect to observe assuming the theory is on the right track - for example, that the point of light we call ‘Mars' should be in such and such a position in the night sky on a particular day at a particular time. Then one checks to see if the predictions are correct, and if they are, this provides confirming evidence for the theory, and if not, this disconfirms (or falsifies) the theory.

There is not much question that this process - observation, theory gener­ation, designing tests, and confirmation/disconfirmation - plays an important role in science. But what is not widely appreciated is just how subtle this simple-sounding process is. In fact, the process just outlined is often pre­sented, and I think often (wrongly) viewed, as relatively straightforward, almost a cookbook approach to science. Sure, the details and ingredients might change from one recipe to the next, but overall, a common view is that the scientific process is not that much different from the relatively straightforward recipe-following process most of us employ in our kitchens.

This science-as-recipe-following view is a substantial misconception about the workings of science. When we probe into this process just a bit, we quickly find some difficult, and largely philosophical, issues lurking just beneath the surface. For example, let us look into just the confirmation/ disconfirmation aspect of the process; that is, the stage where one checks to see if predictions made by a theory are accurate. As a concrete example, consider again the theory mentioned above, that the earth and other planets move about the sun in elliptical orbits, with the sun occupying one focus of an ellipse. This theory (or, at least, the part involving the motion of Mars) was first put forward by Johannes Kepler (1571-1630) in 1609.^[13] In the centuries following the publication of Kepler's theory, predictions based on the theory - such as when and where a planet will appear in the night sky - were quite accurate, and certainly far more accurate than any competing astronomical theory. This large body of accurate predictions provided a vast amount of confirming evidence for Kepler's theory.

On the recipe-following view of science, one might think that centuries' worth of accurate predictions would show that a theory is correct beyond any reasonable doubt. But is this right? Do all these accurate predictions show that Kepler's theory is definitely (or, at least, almost definitely) correct? The answer is clearly no. One general reason (not the only reason, but an important one) is because this sort of reasoning - that is, inferring that a theory is correct based on the theory making accurate predictions - is a species of inductive reasoning. And the inductive nature of this reasoning has important implications for theory confirmation.

In general, what characterizes inductive reasoning is that, even in the strongest sort of such reasoning, and when all the premises (that is, the reasons given in support of the conclusion) given are correct, it is still possible for the conclusion to be wrong.

So again, inductive reasoning can at best provide support for a theory but can never show that a theory is definitely correct. And it is worth empha­sizing that in the case of Kepler's theory, after over 200 years of accurate predictions, increasingly accurate measuring devices suggested that predic­tions about planetary motion, based on Kepler's theory, were not entirely accurate (the amount of error was small, but measurable). So in spite of 200 plus years of confirming evidence for Kepler's theory, it turns out that the theory is not quite right after all.

The example above is typical of what happens in science (and in everyday life, for that matter). We might have years or even centuries of confirming evidence for a theory, in the form of accurate predictions made from the theory. Yet it might (and often does) turn out that even­tually new evidence arises (often as a result of new technology and new inventions) that is no longer compatible with the original theory.^[14]

What about situations in which a prediction made by a theory does not turn out to be accurate, that is, cases of disconfirmation? If we cannot, by inductive confirmation reasoning, show that a theory is definitely correct, can we at least, with disconfirming evidence, show that a theory is definitely (or at least almost definitely) not correct?

At first glance, it might appear so.

But wait. There are any number of other explanations for the lack of the gift. Perhaps your sweetheart did send a gift, but the delivery van crashed. Or perhaps the van was hijacked. Or had a flat tire. Or maybe the gift was delivered to the wrong address. Or perhaps a rival for your sweetheart’s attention stole the gift before you could find it, or your roommate is hiding it as a joke, or any of an indefinite number of other possibilities.

In this context, such other possibilities are often referred to as “auxiliary hypotheses.”^[15] More specifically, auxiliary hypotheses are all the additional factors (in many cases there are a large number of such factors) needed to get the prediction from the theory. That is, one can expect to observe the predicted observation only if the theory, and all the auxiliary hypotheses, are correct. So if no gift arrives for you on Valentine’s Day, the only conclu­sion you can reasonably draw is that either your sweetheart does not love you, or the van broke down, or the van was hijacked, or had a flat tire, or a rival stole the gift, or your roommate is hiding it as a joke, and so on for all the other possible auxiliary hypotheses. In short, in most situations in which a predicted observation does not occur, there are countless explan­ations, other than the falsity of the main theory, for why the prediction was not observed. So when faced with an incorrect prediction, one typically cannot have great confidence that the theory in question is incorrect.

For an instance of this in a scientific context, consider again the example above of Kepler's theory. At about the same time as Kepler was publishing his views, the telescope was invented and was being used (mainly by Galileo) for astronomical observations.^[16] As a new instrument, the telescope provided new data relevant to the issue of whether the sun moves around the earth (as in the traditional system), or whether the earth moves around the sun.

Some of this new evidence seemed incompatible with the theory that the sun moves around the earth. For example, through a telescope Venus can be seen to go through a range of phases much like the moon, which for reasons we need not go into here initially seems problematic for the earth-centered view. That is, it seems that if the earth-centered view is correct, then Venus should not be observed to go through phases.

But as usual, auxiliary hypotheses are lurking under the surface. It turns out that the phases of Venus (again for reasons we will skip over here) are only a problem if both the sun and Venus revolve around the earth. So, in this case, the real reasoning is “if the sun moves around the earth, and Venus does also, then Venus should not be observed to go through phases.” And as a matter of fact, even before the telescope was invented one of the most prominent astronomers of the time had proposed a theory in which the earth was the center of the universe, the moon and sun orbited the earth, and all the other planets orbited the sun. The system is now referred to as the Tychonic system (after its proposer, Tycho Brahe, 1546-1601).^[17] Moreover, the phases of Venus are exactly what would be expected if the earth-centered Tychonic system is correct (and so the phases of Venus actually provide confirming evidence for the earth-centered Tychonic system).

In short, the auxiliary hypotheses lurking under the surface mean that the discovery of the phases of Venus did not in any way rule out an earth­centered system.

In general, it is difficult to rule out, definitively, any theory. There may be no better way to reinforce this point than by noting that, even 400 years later, there is no shortage of advocates of the Tychonic system, that is, advocates of the view that the earth is the center of the universe. In fact, if one takes the original Tychonic system, and modifies it to include Kepler's contribution of elliptical orbits (and a few other modifications from twentieth-century physics), the resulting system (call it the modernized Tychonic system) is an earth-centered system that makes the same major predictions of the vastly more widely accepted sun-centered view of our solar system. So the bottom line is that, even with a seemingly outdated theory such as the earth-centered theory, it is difficult to find straightforward observational evidence that definitively shows the theory to be incorrect.

2.2 Karl Popper and falsificationism

A topic closely related to the confirmation/disconfirmation issues dis­cussed above concerns a common view on what distinguishes scientific from non-scientific theories. For example, there is widespread agreement that, say, Kepler's theory on planetary orbits is a scientific view, whereas, say, astrology is not. (One can make a case that astrology at one time could have been considered properly scientific, but the general consensus is that astrology is no longer a genuinely scientific theory.)

What distinguishes scientific theories from non-scientific theories? There is no widespread consensus on the answer to this question, but one of the most popular answers, especially among working scientists (as well as a good number of philosophers of science), has its roots in the work of Karl Popper (1902-94).^[19]

In outline, Popper's approach is to emphasize the disconfirmation (or at least attempted disconfirmation) of theories, rather than focusing on confirmation. Part of the reason for this is because confirmation often seems too easy. For example, if I check my astrological forecast for this week, I find that I should not expect straight answers at work, leading to frustra­tion with some work projects. And indeed, as I reflect on this past week, I have been a bit frustrated with some committee work, partly because of a sense of not receiving information as complete as I would like.

If we look again at the pattern of confirmation reasoning from the previous section, we can note that this astrology example fits the pattern. But there is broad consensus that this sort of example provides no real confirming evidence for astrology. The reason seems clear: the astrological prediction is much too vague and broad, so there is almost no chance of not observing something this week that would fit the prediction.

Situations similar to those illustrated by the astrology example arise in mainstream science as well. Consider again the examples discussed in the previous section, namely Kepler's theories on planetary orbits versus the (modernized) Tychonic view of planetary orbits. As noted, the vast majority of predictions generated by each of these theories are identical. So even though these are competing theories, we can easily generate thousands of predictions for each of these theories (for example, predictions about where all the planets, moon, sun, stars, etc. should appear in the sky), and these predictions, for both theories, will all be accurate.

Partly because of the relative ease with which one can generate confirm­ing evidence for a theory, Popper argued that we should de-emphasize confirmation and instead look to disconfirmation as a key criterion for whether a theory should be considered scientific. Moreover, according to Popper, the riskier a theory, the better it is. That is, the best examples of scientific theories will be those that make specific, straightforwardly testable predictions, such that if those predictions turn out to be incorrect, this will strongly suggest that the theory is incorrect. In short, the best theories are those that expose themselves to being falsified. (By way of contrast, note that because of the vagueness of its predictions, astrology would not meet this criterion, and thus, on Popper's view, astrology should not count as a legitimate scientific theory.)

As a corollary to this, our best theories, according to Popper, are not those that have piled up tons of confirming predictions; rather, our best theories are those that have passed these sorts of tests in which the theory was exposed to the possibility of being falsified. An example (one Popper himself used) might help clarify this.

In the decades and century following the publication of Kepler's theory of planetary orbits, that theory was fleshed out by various advances, most notably, Newton's physics.^[20] We need not go into the details here, but, broadly speaking, Newton's physics provided a broader theoretical framework, of which Kepler's earlier theory was a sort of subset. So, for the remainder of this example, I will focus on Newton's theory, with the understanding that it contains Kepler's account of planetary motion.

Newton's theory was capable of making quite specific predictions - for example, predictions as to the position of planets. And coupled with the increasingly accurate means of measuring planetary positions, Newton's theory seems a good example of a theory at risk.

And indeed, in the 1800s astronomers were able to determine that the orbit of Uranus did not match the predictions of Newton's theory. But, as it turns out, an additional planet (Neptune) was discovered, and the influence of Neptune on Uranus nicely explains, using Newton's theory, the earlier observed perturbations in the orbit of Uranus. Moreover, Newton's theory had been used to predict the possible existence of an unknown planet, exactly because such a planet would explain the oddities with the orbit of Uranus.

So, Newton's theory is one that makes specific predictions that are reason­ably straightforward to test, and, as such, Newton's theory is, according to Popper, a good example of a theory at risk. And moreover, Newton's theory is a good example of a theory that has survived such tests, including initially problematic situations such as illustrated by the case involving the perturbations of the orbit of Uranus.

In summary, on Popper's view, our best theories are not those that have endless cases of confirming predictions, but rather, our best theories are those that have survived specific cases where they could have been shown to be wrong. And in addition, it is primarily this characteristic - that is, being at risk in the sense of being able to be shown to be incorrect - that distinguishes theories that are genuinely scientific (for example, Newton's physics) from those that are not (for example, astrology).

The above discussion is a fairly rough sketch of Popper's views on these matters. But suffice it to say that this approach of Popper's - primarily, view­ing attempts at falsifying theories as more important to science than confirm­ing evidence, and viewing potential falsification as the key characteristic of a scientific theory - is a fairly common view among working scientists.

As noted, Popper's views have been quite influential among working scientists and philosophers of science. But such views have also played important roles in key court cases. For example, in an important case in Arkansas in 1982 (McClean v. Arkansas Board of Education 1982), the judge ruled that creation science was not science, and Popper's views (especially the importance of a theory being capable of being falsified) played an important role in that decision. Similar comments hold for a more recent ruling in a Dover, Pennsylvania case (Kitzmiller v. Dover Area Sch. Dist. 2005), in which intelligent design was found not to be a legitimate scientific theory, again largely on Popperian grounds.

But in spite of the popularity of Popper's views, the views outlined above are a bit more problematic than they might at first seem. The basic com­plicating factors are largely those discussed in the preceding section. Recall that in any case where a theory appears to be falsified (that is, any case in which an expected prediction is not observed), there are always auxiliary hypotheses involved. For example, consider again the issues described above involving the orbit of Uranus. When observations indicated that the orbit was not what Newton's theory predicted, the reaction was not to jump to the conclusion that Newton's theory was incorrect. Instead, researchers did the reasonable thing when faced with problematic evidence from a key and well established theory such as Newton's - they looked to auxiliary hypotheses as an alternative explanation for the failed predictions. And in this case, they found an alternative explanation, namely, a new planet.

Moreover, as argued most notably by another prominent figure in twentieth-century philosophy of science (as well as physics and mathematics), Imre Lakatos (1922-74), even if a new planet had not been discovered to account for the failed predictions, there is essentially no chance we would have abandoned Newton's theory (Lakatos 1980). Newton's theory had become so key to science, and so well established, that scientists would almost certainly have continued to look into alternative auxiliary hypotheses rather than rejecting the Newtonian theory. Part of the view defended by Lakatos, now generally accepted, is that long-term, well entrenched theories are quite resistant to any sort of straightforward falsification, and that theory change in such cases is a much more complex process (involving, for example, complex interactions within and between competing research programs).^[21]

Issues such as those raised by Lakatos provide problems that go beyond what we can thoroughly look into here. There is no doubt that falsification is an important tool in science, but there is likewise not much doubt that the process of falsification, and of theory acceptance and rejection, is a more complex process than that outlined above. Suffice it to say that, as usual, the situations involving scientific theories, and issues in the philosophy of science, tend to be much more complex and difficult than they often appear to be at first glance.

2.3 The underdetermination of theories

Another topic that ties in closely with those discussed above is the notion of the underdetermination of theories. Underdetermination is a common topic in the philosophy of science, and so worth taking a moment to explore. However, issues surrounding underdetermination are, as usual, complex, and in this brief section we will just take a brief look at this subject.

The key ideas behind underdetermination are most closely associated with the work of Pierre Duhem (1861-1916; Duhem was a physicist, though one with a lifelong interest in philosophical issues related to science), and also with W.V.O. Quine (1908-2000).^[22] As usual, an example will help illus­trate this notion. Consider again the Newton/Kepler sun-centered view, and the modernized Tychonic (earth-centered) view. As mentioned above, even though these are competing theories (one being an earth-centered the­ory; the other a sun-centered theory), both make the same predictions about observations such as the phases of Venus, where and when planets should be observed in the night sky, the positions and movements of the stars, times of solar and lunar eclipses, times of sunrise and moonrise, and so on.

In short, most (arguably all) of the straightforward observational evi­dence is compatible with both the earth-centered, modernized Tychonic system, and the more widely accepted Kepler-style sun-centered view of the solar system. In general, this is what is meant by underdetermination. That is, the basic idea behind underdetermination is that often (maybe typically, or even always, depending on your views on the subject), the available evidence will be compatible with two or more competing theories. Or, as it is often phrased, theories are underdetermined by the available data.

I should note again that the issues involved in underdetermination are more complex than the quick sketch above might suggest. For example, there are a variety of ways, some very controversial and some not contro­versial at all, of construing underdetermination. As an example of an uncontroversial construal, there is little question that, at times, there have in fact been two or more competing theories equally supported by the available evidence. At the much more controversial end of the spectrum, some argue that such considerations support the view that scientific theories are entirely social constructs, such that there are no objective reasons to support one theory over another any more than there are objective reasons to support one ice-cream preference over another.

To review quickly §§2.1-2.3: we began by noting that the process of observation, theory generation, testing, and confirmation/disconfirmation was an important process in science. And moreover, it seems to be a com­mon conception that this process is a more or less recipe-following pro­cess, with few or no philosophical or conceptual issues involved. Thus far we have discussed primarily the confirmation/disconfirmation part of this process, but as we have seen, just a little probing under the surface reveals some difficult, and largely philosophical, issues involved in this process.

Though we do not have time to consider them in detail here, similar issues arise when one looks into the other aspects of the process. For example, consider the observation part of the process. Presumably, one observes data before generating possible theories to explain that data. But it is far from clear that it is possible to gather data in a theory-neutral way. (For example, can we be theory-neutral about any subject, and can we gather data with­out already having some theoretical framework to help make sense of that data?) So it is far from clear that one can objectively gather data prior to having theories about that data.

Theory generation is likewise a complex process, and no one has ever developed anything remotely resembling a recipe for producing theories. When one looks at historical examples and tries to reconstruct the process by which a scientist arrived at a theory, the range of processes we find are astounding. In short, there does not appear to be any sort of easily articulated process or procedure by which one generates theories.

And it is likewise for the testing aspect of the process. Designing a test of a theory is almost never simple or straightforward. Some of the reasons for this involve the role of auxiliary hypotheses, discussed above, and the sorts of issues discussed above involving inductive reasoning. But also there is simply the raw complexity of many tests of theories. Over the past few centuries, theories in science have tended to become increasingly complex, and, not surprisingly, many of the tests of such theories have likewise become increasingly complex. For example, when Einstein proposed his theory of relativity early in the 1900s, one of the unusual predictions one could make from the theory was that starlight should be bent when passing near a large object such as the sun. But putting this prediction to the test was complex. To note just one complexity, the mathematics involved were complex enough that, in order to solve the equations, a good number of simplify­ing assumptions, known to be incorrect, had to be made (for example, that the sun is perfectly spherical, uninfluenced by outside forces, and non­rotating - none of which is correct).^[23]

In summary, none of the features of the processes outlined - observation, theory generation, testing, or confirmation/disconfirmation - are in any way straightforward or any sort of recipe-following procedure. As we are seeing, it does not take long to uncover, in science, substantive philosophical issues.

2.4 Instrumentalism and realism

The discussions above largely concerned issues involved in theory confirm­ation and disconfirmation. In this section we will look at a slightly different topic, one involved, roughly, with the question of what we want from scientific theories.

One answer to this question, which almost everyone agrees on, is that we want scientific theories to be able to handle the relevant data. In par­ticular, we want theories to be able to make accurate predictions, and we evaluate theories in large part based on how accurate their predictions are.

So, for example, there is general agreement that Newton's physics (developed mainly in the late 1600s) makes more accurate predictions than does Aristotle's physics, and that modern physics (for example, Einstein's special and general theories of relativity, developed early in the 1900s) makes more accurate predictions than does Newton's physics (for a brief overview of Aristotle, Newton, and Einstein, see DeWitt 2004).

But suppose we ask the question of whether Newton's theories provided a better model, or picture, of reality, than did Aristotle's theories, and whether Einstein's theories in turn provide a better model or picture of reality than do Newton's theories. That is, in general, is one theory better than another not just because it gives more accurate predictions, but, in addition, because it provides a better model, or picture, of reality? Is reality the business of science; that is, should we expect, and even require, a good theory to provide an accurate picture of the way things really are?

11

With respect to this question, we find a substantial amount of dis­agreement. On the one hand are those who maintain that science ought to attempt to provide theories that not merely handle the data, but that reflect the way things really are. Stephen Weinberg, a leading physicist of recent decades and winner of the 1979 Nobel Prize in physics, is not hesitant to express sentiments such as that the job of science “is to bring us closer and closer to objective truth” (Weinberg 1998), and that there is “a necessity built into nature itself” (Weinberg 1992), and that in large part the job of a scientist is to find theories that capture the necessity inherent in nature. Einstein likewise often seemed to embrace such sentiments. For example, Einstein objected all his life to one of the most predictively successful theories we have ever had, namely quantum theory. Einstein fully recognized that quantum theory was enormously successful in terms of making accurate predictions, but he objected to the theory because he was convinced the theory could not possibly reflect the way the universe really is.

The view expressed by Weinberg and Einstein (and plenty of others, both scientists and philosophers of science), that theories in science ought to reflect the way the universe really is, is generally referred to as realism. And again, the basic idea behind a realist approach to science is that a good theory ought not only to make accurate predictions, but should in addition be required to model accurately, or picture, the way the universe really is.

In contrast to realism is a view usually referred to as instrumentalism (also sometimes referred to as operationalism). Those in this camp believe the main job of a theory is to handle the data (by, for example, making accurate predictions), and whether a theory accurately models or pictures reality is of no importance. This view is also found prominently among both working scientists and philosophers of science. For example, an instrumentalist attitude toward science is advocated in almost every standard college-level text on quantum physics, in which the student of quantum physics is generally encouraged to focus on making accurate predictions, and to leave worries about reality to the philosophers.

Notice a key aspect of realism and instrumentalism: namely, instru­mentalism and realism are clearly philosophical views. They certainly are not themselves scientific theories - they are not, for example, designed to handle any body of observed data, nor can they be used (not in any straight­forward sense) to generate predictions about what might be observed. Nor does it make much sense to speak of putting instrumentalism and realism to any sort of experimental test. That is, some of the key hallmarks of scientific theories - the ability to use them to make predictions, being able to put them to experimental tests, and so on - do not apply to instru­mentalism and realism.

Nor are instrumentalism and realism parts of scientific theories. This is easy enough to see. One and the same theory (e.g., quantum theory) might well be taken with an instrumentalist approach by some scientists and a realist approach by others. In each case the theory is the same, so the instru­mentalism and realism must not be part of the theory itself.

So instrumentalism and realism are neither scientific theories, nor parts of scientific theories. Instead, instrumentalism and realism are attitudes toward scientific theories, and philosophical attitudes at that.

Notice also that whether a particular scientist is more of an instru­mentalist or a realist might well affect the approach he or she takes toward science. To give one example: in the beginning of a classic 1905 paper by Einstein (the paper in which he introduces what is now called the special theory of relativity), Einstein indicates that part of his motivation for pursuing this project stemmed from what he viewed as oddities in the application of the usual equations in physics used to handle cases where an electrical current is induced in a wire coil, either by moving the coil past a stationary magnet or by moving the magnet past a stationary coil (Einstein 1905). Notably, there were no problems with the predictions one got from the existing theory. So from an instrumentalist view (that is, if one primarily wants a theory to provide accurate predictions), there were no problems in this case for the existing theory.

The only puzzle (and this is part of what Einstein was interested in) stemmed from taking a realist perspective. Roughly, the worry was that the existing theory treated the two cases (magnet moving past a stationary coil versus coil moving past a stationary magnet) differently. And although Einstein did not phrase it this way, one way to put the problem is by asking how could the magnet and coil “know” which was stationary and which was moving? They obviously cannot, and so it seems strange to have dif­ferent parts of the existing theory apply depending on which is stationary and which is moving.

Again, note that this worry is only a worry from a realist perspective - again, the predictions from the existing theory were accurate, so there are no worries from an instrumentalist perspective. It is not difficult to find other examples of this sort, in which whether a working scientist takes more of a realist or an instrumentalist approach affects the way he or she approaches science. And it is worth noting that in the example above, Einstein’s realist leanings were part of what led to one of the most import­ant theories of the twentieth century.

So once again we see it does not take long to find philosophical issues - in this case, instrumentalist and realist attitudes - cropping up in science. And as with the issues involved in the previous section, these philosophical issues are not of interest only to philosophers, but rather, they often influence the actual doing of science.

3.