Reading Assignment No. 1 (Philosophy of Science).pdf

CHAPTER 1 Philosophy of Science Bjørn 1 Hofmann1 and Søren Holm2 Section for Health, Technology and Society, University College of Gjøvik, Gjøvik, Norway and Centre for Medical Ethics, Institute of Health and Society, University of Oslo, Blindern, Oslo, Norway 2 Centre for Social Ethics and...

CHAPTER 1 Philosophy of Science Bjørn 1 Hofmann1 and Søren Holm2 Section for Health, Technology and Society, University College of Gjøvik, Gjøvik, Norway and Centre for Medical Ethics, Institute of Health and Society, University of Oslo, Blindern, Oslo, Norway 2 Centre for Social Ethics and Policy, School of Law, The University of Manchester, England, Centre for Medical Ethics, Institute of Health and Society, University of Oslo, Blindern, Oslo, Norway and Department of Health Science and Technology, Aalborg University, Denmark 1.1 INTRODUCTION The sciences provide different approaches to the study of man. Man can be scrutinized in terms of molecules, tissues and organs, as a living crea- ture, and as a social agent and a cultural person. Correspondingly, the philosophy of science investigates the philosophical assumptions, founda- tions, and implications of the sciences. It is an enormous field that covers the formal sciences, such as mathematics, computer sciences, and logic; the natural sciences; the social sciences; and the methodologies of some of the humanities, such as history. Discussion in the current chapter is limited to the natural sciences (Section 1.2) and the social sciences (Section 1.13), and comprises a brief overview of the philosophical aspects salient to research in the medical and biological sciences. 1.2 PHILOSOPHY OF THE NATURAL SCIENCES What does it mean when one says that “smoking is the cause of lung cancer”? What counts as a scientific explanation? What is science about, e.g., what is a cell? How does one obtain scientific evidence? How can one reduce uncertainty? What are the limits of science? These are but a few of the issues discussed in the philosophy of the natural sciences that will be presented in this chapter. The traditional philosophy of science has aimed to put forth logical analyses of the premises of science; in particular the logical analysis of the syntax of basic scientific concepts. In the following sections, the principal traditional issues regarding the rationality, method, evidence, and the object of science (the world) are discussed. But first the core concepts of science, knowledge, and truth will be addressed: What is science? What is scientific knowledge, and what constitutes a scientific fact? Research in Medical and Biological Sciences © 2015 Elsevier Ltd. DOI: http://dx.doi.org/10.1016/B978-0-12-799943-2.00001-X All rights reserved. 1 2 Research in Medical and Biological Sciences 1.3 WHAT IS SCIENCE? DIFFERENTIATING SCIENCE FROM NONSCIENCE Science is traditionally defined as “the systematic search for knowledge,” meaning that science has an aim (knowledge) and it is an activity (search) with certain qualifications (systematic). However, not everyone who car- ries out a systematic search for knowledge can obtain research funding. For example, though a religious man may search for knowledge through highly systematic meditation techniques, it is unlikely that this will be considered a project in need of research funding. Indeed, much of the time even well-funded scientists do not perform systematic searches for knowledge. They struggle with experimental designs, analyze data, pres- ent research results, argue with other researchers over conflicting results, and write funding proposals. Therefore, a more complete definition of science may be: “Scientific research is the systematic and socially organized (a) search for, (b) acquisi- tion of, and (c) use or application of knowledge and insight brought forth by acts and activities involved in (a) and (b).”1 This definition better reflects what scientists actually do. They search for new knowledge, e.g., by investigating the possibility of using biomolecular tests of cell-free fetal DNA/RNA in the blood of pregnant women to find defects in the fetus (noninvasive prenatal testing, NIPT). They acquire knowledge by testing, accepting, or rejecting a hypothesis, e.g., that NIPT is better than com- bined ultrasound and serum tests for detecting fetuses with trisomy 13, 18, and 21. Finally, scientists apply knowledge when they argue that a certain study is either appropriate or flawed, and therefore its results either valid or invalid. Although this definition of science is closer to what scientists do, it may still be difficult to differentiate those doing science from those doing nonscience. Throughout history a series of criteria has been used to demarcate science from nonscience (sometimes also referred to as pseudoscience), and thereby also to define science. Francis Bacon (1561 1626) defined science as a specific method, i.e., performing a systematic analysis of data without preconception. Data analysis framed by preconception was considered non- science. However, in reality it can be very difficult to analyze data without preconception. To illustrate this fact, if you study the figure below (Figure 1.1), what do you see? A rabbit, a duck, or perhaps both? When viewed on paper (or on a screen) the figure has black dots on a white background. So where are the duck and the rabbit? Are they  Philosophy of Science 3 Figure 1.1 The duck/rabbit. figments of the imagination? Are they preconceptions? Hence, to say that science is the systematic analysis of data without preconception is too restrictive; it would rule out most of what is called science today. Indeed, all observations and analyses are based on preconceptions. Another way to differentiate science from nonscience is to say that preconception is acceptable in the context of discovery but not in the context of justification, i.e., when the data are tested. The basic idea is that the pattern of nature is neutral, and will stand out in the end. However, this does not solve the problem of preconception when the hypothesis is tested, as testing presupposes observations, and observations presuppose preconceptions. A third classical demarcation criterion is that a scientific hypothesis or theory can be contested with possible observations, i.e., it can be falsified (or refuted).2 However, theories are seldom really falsified,3 and to falsify a theory presupposes that the researcher has an idea about what will hap- pen. As scientists we are seldom ready to give up our theories, and instead add specifications or modifications. It has been argued that scientists are preoccupied with puzzle-solving, i.e., solving puzzles within a given mode of thought (paradigm), when they should be concentrating on falsification.3 Until the 1960s it was commonly thought that science progressed in a linear and piecemeal fash- ion, in which new knowledge added to existing knowledge. However, Thomas Kuhn (1922 1996) and others argued that science evolved through anomalies. Hard cases that could not be explained within the given paradigm challenged existing theories and resulted in a scientific revolution. A new set of theories established a new paradigm, and scien- tists turned to puzzle-solving within this new paradigm. The shift from Newton’s mechanics to Einstein’s theory of (special) relativity is a key 4 Research in Medical and Biological Sciences BOX 1.1 Selected Characteristics of Nonscience 1. Belief in authority: Some person or persons have a special ability to determine what is true or false, and others have to accept their judgment. 2. Nonrepeatable experiments: Reliance on experiments with outcomes that cannot be reproduced by others. 3. Handpicked examples: Examples that are not representative of the general category to which the investigation refers are considered decisive. 4. Unwillingness to test: A testable theory is not tested. 5. Disregard of refuting information: Ignoring or neglecting observations or experiments that conflict with a theory. 6. Built-in subterfuge: Arranged the testing of a theory so that it can only be verified but never falsified by the outcome. 7. Explanations abandoned without replacement: Giving up tenable explanations without replacing them, so that more is left unexplained in the new theory than in the previous one.4,5 example. However, if science is defined by a given paradigm, this poten- tially makes everything science, as long as scientists define it as a paradigm and are preoccupied with solving its small problems. The above mentioned four demarcation criteria are but a few of many. None of them are flawless, and it has turned out to be quite difficult to dif- ferentiate science from nonscience. In order to remedy this, several sets of more pragmatic criteria to identify nonscience have evolved (Box 1.1). 1.4 KNOWLEDGE AND TRUTH: WHAT IS KNOWLEDGE AND WHAT CONSTITUTES A SCIENTIFIC FACT? The great tragedy of science, the slaying of a beautiful theory by an ugly fact. T. H. Huxley (1825 1895) The standard definition of knowledge is: “justified true belief.” Hence, one key criterion of knowledge is that it can be justified. This is done all the time when researchers do experiments, present results, and publish papers. If a statement about events in the world cannot be justified (by the- ory, analogies, or experiments), it is not accepted as knowledge. Another criterion is truth. Traditionally, scientists have thought that knowledge is true. If one knows that individuals with Huntington’s disease have more than 39 repetitions of the Huntington gene, one holds this to be true. But what about the third criterion of knowledge: belief ? Many think that science and belief belong to different realms. However, if scientists say that  Philosophy of Science 5 BOX 1.2 Statistical Methods and Advanced Technology May Muddle Truth In a study some researchers showed photographs of people in various social settings to a salmon and measured the difference in brain activity. The salmon “was asked to determine what emotion the individual in the photo must have been experiencing.” The results showed that the salmon could read human emotions. However, the salmon was dead. The point was to show that brain researchers can use complicated instruments and simple statistics to show anything they want, even meaningful brain activity in a dead salmon.6 they know something, they must believe that it is true, and they must be able to justify it to their peers. To say that one knows something but does not believe it is known as Moore’s paradox. For example, “I know that type I diabetes results from the autoimmune destruction of the insulin-producing beta cells in the pancreas, but I do not believe it (Box 1.2).” So how does one decide what is true? There are vast debates on whether mammography screening reduces breast cancer mortality and whether it results in overdiagnosis. How should this be decided? As with the demarcation criteria for science, there are a large number of criteria for assessing the truth of statements, theories, and hypotheses about events and objects in the world. The correspondence theory of truth says that some- thing is true if it corresponds with events or things in the world; but how does one know the nature of the world if not through one’s observations? If one is to decide whether a certain phenomenon can be seen in a microscope, a hypothesis must compare what one expects to see with what one actually sees through the ocular. However, as seen in Figure 1.1, observations also strongly depend on preconceptions. Hence, when corre- spondence is verified, only the hypothesis and the conceptions of the observations are compared. In other words, it is hard to speak of corre- spondence with events or things in the world without having direct access to the world (independent of our observations). This problem is avoided by the coherence theory of truth, which contends that a scientific hypothesis or theory is true if it coheres with other hypotheses or theories. If it is completely out of the realm of established knowledge, it is most likely untrue. However, take the Australian scientists Barry Marshall and Robin Warren, who identified Helicobacter pylori in patients with chronic gastritis and gastric ulcers and hypothesized that it 6 Research in Medical and Biological Sciences BOX 1.3 Facts and Truths Facts are simple and facts are straight. Facts are lazy and facts are late. Facts all come with points of view. Facts don’t do what I want them to. Facts just twist the truth around. Facts are living turned inside out. David Byrne may be a causal factor. In 2005 Barry Marshall was awarded the Nobel Prize for Physiology for this research (Chapter 4). However, their hypoth- esis certainly did not cohere with the established medical knowledge of the time, which held that ulcers were caused by stress, too much acid, and spicy foods. Indeed, according to the coherence theory of truth, Marshall and Warren were wrong, which shows how theory can prevent new and important insights. Therefore, it has been argued that scientific truths should be established through consensus, e.g., at consensus conferences. If the best experts in a field, after thorough analyses of the evidence, open debates, and careful deliberation, decide on something, it is the closest thing to scientific truth in this field. However, scientists can be wrong, even when they agree, as is illustrated by the case of Helicobacter pylori. Another theory, called the pragmatic theory of truth, argues that truth is what is useful or what works. Truth can be decided through inquiry by a community of scientific investigators7 or by whether it is trustworthy and reliable.8 The problem with a pragmatic theory of truth is that it tends to be either circular or relativistic. It is circular when truth depends on what works, and researchers’ assessment of what works depends on their concep- tion of what is true. It is relativistic if anything can be defined as truth as long as it is part of scientists’ common beliefs or decisions (Box 1.3). Example Sol Spiegelman (1914 1983) was a molecular biologist who developed the technique of nucleic acid hybridization. During the 1970s he worked to establish that retroviruses cause human cancers, and he identified them in human leukemia, breast cancer, lym- phomas, sarcomas, brain tumors, and melanomas. His persistent work and optimistic statements impelled enthusiasm, which fueled funding, and so on, in a self-perpetuating manner. However, he saw viruses where nobody else could; nobody was able to repli- cate his experiments. Hence, there are many sources of facts, and we do well in reflect- ing critically on their existence and their emergence.  Philosophy of Science 7 BOX 1.4 Why Most Research Results in Emerging Fields are False 1. The studies conducted in an emerging scientific field are small. 2. The effect sizes in an emerging scientific field are small. 3. The number of tested relationships is great, but the selection of tested relationships is small. 4. The flexibility in designs, definitions, outcomes, and analytical modes in an emerging scientific field are great. 5. The financial and other interests and prejudices in a scientific field are great. 6. The scientific field is hot (more scientific teams are involved). In an article published in PLOS Medicine, John Ioannidis9 argues that when it comes to emerging fields of knowledge, much of what medical researchers conclude in their studies is misleading, exaggerated, or flat-out wrong. In this article, which is one of the most frequently downloaded from PLOS Medicine, he concluded that in such studies it is difficult to get a positive predictive value above 50% that a research result published in a peer-reviewed journal is true. The reason for this is given in Box 1.4. 1.5 THE GLUE THAT HOLDS THE WORLD TOGETHER: CAUSATION Another core concept of science is causation. A pivotal task of the bio- medical sciences is to find the causes of phenomena such as disease, pain, and suffering. However, what is the implication of saying that something is the cause of a disease? What does it mean when one says that smoking causes lung cancer? According to Robert Koch (1843 1910), who was awarded the Nobel Prize for Physiology or Medicine in 1905 further to his discovery of tubercle bacillus, a parasite can be considered the cause of a disease if it can be shown that its presence is not random. Such random occurrences may be excluded by satisfying what have been called the (Henle-) Koch postulates:10,11 1. The organism must be found in all animals suffering from the disease but not in healthy animals. 2. The organism must be isolated from a diseased animal and grown in pure culture. 8 Research in Medical and Biological Sciences 3. The cultured organism should cause disease when introduced into a healthy animal. 4. The organism must be reisolated from the experimentally infected animal. The Koch postulates require that there be no disease without presence of the parasite, and no presence of the parasite without disease. That is, the parasite is both a necessaryi and sufficientii condition for the disease. Please note that conditions are not the same as conditionals, although there are affinities.iii However, as Koch realized when he discovered asymptomatic carriers of cholera, the requirement of both necessary and sufficient conditions for causation is overly rigorous. Indeed, if such pos- tulates were used as the general criteria for something to be considered a cause in the biomedical sciences, causation would be quite rare. 1.5.1 Necessary Conditions What if only the necessary conditions are known, i.e., necessary but not suffi- cient conditions—can they be called causes? The cholera bacterium (Vibrio cholera) is necessary for one to develop cholera, but it is not sufficient, as not all people with the bacterium develop cholera. One would normally say that Vibrio cholera causes cholera. However, the presence of an intesti- nal wall (where the bacteria thrive and produce toxins) is also a necessary condition for developing cholera. If one does not have intestinal walls, one does not develop cholera. This does not mean it is correct to say that the intestinal wall causes cholera. Indeed, there may be many necessary conditions for an event that are not considered causes. However, neces- sary conditions are germane to health care, as without them the event (e.g., disease) will not occur. Hence, necessary conditions are relevant i A necessary condition here means a condition without which an event would not occur, i.e., sine qua non. Without HIV, a person does not develop AIDS. Necessary conditions work through their absence, because if you remove a necessary condition, you also remove or prevent the event. ii A sufficient condition here means a condition that is sufficient for an event to occur, i.e., the event occurs every time the condition occurs. Sufficient conditions work through their presence, because if you provide the condition, then you also provide the event. iii A conditional is a statement of the form “If p, then q.” Conditional statements are not statements of causation, as the relationship between p and q is logical, and not temporal. E.g., the conditional statement “If Barack Obama is president of the United States in 2014, then Norway is in Europe” is true, although we would not say that Barack Obama causes Norway to be in Europe.  Philosophy of Science 9 through their absence; e.g., tuberculosis can be eradicated by eliminating one of its necessary conditions: Mycobacterium tuberculosis. 1.5.2 Sufficient Conditions What then is a researcher to say when there are sufficient but not necessary conditions? For example, when a person develops skin burns after being exposed to ionizing radiation known to be of the sort and strength that results in burns, one would tend to claim that the ionizing radiation caused the skin burns. Of course, the burns could have been caused by other factors (sufficient conditions) in the absence of ionizing radiation. Nevertheless, if a sufficient condition is present, the researcher knows that the event (the effect) will occur. Hence, sufficient conditions for an event are usually acknowledged as causes. Contrary to necessary conditions, suffi- cient conditions work through their presence; however, for how many dis- eases are the sufficient conditions actually known? Indeed, if causes are restricted to known sufficient conditions, not many causes are known. For instance, being infected with Yersinia pestis is not sufficient to get the plague. 1.5.3 Combination of Conditions that Together are Necessary and Sufficient Another situation is when there are two conditions that, individually, are insufficient for a certain event, but combined they can make an event occur. For example, an anaphylactic reaction cannot be caused simply by being stung by a bee, or simply by being hypersensitive to bee venom. However, acting jointly in certain circumstances, both of these conditions are necessary and sufficient for an anaphylactic reaction, therefore both conditions are said to cause the event. In short, each of the conditions is an insufficient but necessary part of a necessary and sufficient condition for the event. Again the question must be asked, for how many diseases are com- bined sufficient and necessary conditions known? It is more difficult to find examples than one would like to think, which may be due to the complicated characteristics of nature itself. 1.5.4 Combination of Conditions that Together are Sufficient It has been argued that cause usually refers to an insufficient and necessary part of an unnecessary but sufficient condition (INUS condition),12 and is char- acterized by a combination of conditions that together are sufficient but not necessary. For instance, a person who is not exposed to the sun and 10 Research in Medical and Biological Sciences does not take vitamin D will have osteomalacia. Here the lack of sun exposure and lack of vitamin D are necessary but insufficient conditions. However, together they are sufficient but not necessary conditions for osteomalacia, as this disease may also result from other conditions. Correspondingly, having blood is a necessary condition for having sepsis, as no person without blood has sepsis; this does not mean one can say that blood causes sepsis. One way to visualize INUS causation is through the use of “causal pies.” Figure 1.2 illustrates the conditions that are considered to be suffi- cient for cardiovascular disease. Indeed, one can use this model to find and fill in various combined sufficient conditions for any disease, as there may be several conditions that are sufficient for a disease to occur (Figure 1.3). Figure 1.2 Conditions considered sufficient for cardiovascular disease. Causal pies were suggested by Kenneth Rothman in 1976.13 Figure 1.3 There may be several sets of sufficient conditions for a disease. Note that component A is a necessary condition because it appears in every pie.13  Philosophy of Science 11 Take the example of smoking and lung cancer: are the sufficient conditions for lung cancer known, or is smoking an INUS condition for lung cancer? An INUS condition accommodates the fact that not all smokers develop lung cancer, and not all people with lung cancer have been smokers. However, it requires a concert of conditions for which lung cancer follows when smokers, but not nonsmokers, are subjected to them. Approaches that define causation in terms of combined necessary and sufficient conditions commonly hinge on scientific determinism: the idea that complex phenomena can be reduced to simple, deterministic mechanisms, and therefore in principle can be predicted. In the case of smoking and lung cancer, all the conditions are not known. Hence, the existence of hidden conditions must be assumed in order to retain scien- tific determinism. The belief in unidentified conditions, as well as the difficulty in explaining a dose response relationship, has challenged the conception of the different versions of causation based on components of sufficient conditions (sufficient condition, insufficient but necessary part of a necessary and sufficient condition, and INUS). 1.5.5 Probabilistic Causation Rather than satisfying an INUS condition, the observation is that smokers develop lung cancer at higher rates than do nonsmokers.14 This leads to the belief that the increased probability of lung cancer among smokers constitutes causation, i.e., probabilistic causation. The central idea in probabilistic causation is that causes raise the probability of corresponding effects.15 Acknowledging that overly stringent criteria for causation minimize the chances of identifying any causes of disease, the British medical statis- tician Austin Bradford Hill (1897 1991) outlined tenable minimal condi- tions germane to establishing a causal relationship between two entities.16 Nine criteria were presented as a way to determine the causal link between a specific factor (such as cigarette smoking) and a disease (such as emphysema or lung cancer). Bradford Hill called them “viewpoints” on causation (Box 1.5). However, they are frequently referred to as criteria (see also Chapter 9). The Bradford Hill criteria are less exacting than the Koch postulates. Nonetheless, there are many cases where one might refer to “the cause of the disease,” but where the Bradford Hill criteria do not apply. Despite their plausibility, probabilistic approaches to causation have been challenged on several levels. First, association is not causation. 12 Research in Medical and Biological Sciences BOX 1.5 Bradford Hill’s “Viewpoints” on Causation 1. Strength of Association—the stronger the association, the less likely the relationship is due to chance or a confounding variable. 2. Consistency of the Observed Association—has the association been observed by different persons, in different places, circumstances, and times (similar to the replication of laboratory experiments)? 3. Specificity—if an association is limited to specific persons, sites, and types of disease, and if there is no association between the exposure and other modes of dying, then the relationship supports causation. 4. Temporality—the exposure of interest must precede the outcome by a period of time consistent with any proposed biologic mechanism. 5. Biologic Gradient—there is a gradient of risk associated with the degree of exposure (dose response relationship). 6. Biologic Plausibility—there is a known or postulated mechanism by which the exposure might reasonably alter the risk of developing the disease. 7. Coherence—the observed data should not conflict with known facts about the natural history and biology of the disease. 8. Experiment—the strongest support for causation may be obtained through controlled experiments (clinical trials, intervention studies, animal experiments). 9. Analogy—in some cases, it is fair to judge cause effect relationships by analogy—“With the effects of thalidomide and rubella before us, it is fair to accept slighter but similar evidence with another drug or another viral disease in pregnancy.”16 The rooster’s crow does not make the sun rise. Down syndrome is strongly associated with birth rank, but birth rank is not considered to be a cause of Down syndrome (because the effect of birth rank is medi- ated by the association between Down syndrome and maternal age). Correspondingly, there is a significant association between chocolate con- sumption in a country and the number of Nobel laureates per million inhabitants,17 though most people would not say that chocolate consump- tion creates Nobel laureates. Second, not all associations (even strong ones) are causal, so it is not easy to say how strong an association must be in order to classify it as a causal relationship. Where does one set the limit on causation? It is argued that aspirin “causes” Reye’s syndrome in children, and that certain tampons “cause” toxic shock syndrome, though the probabilities are very low indeed. The limits of what can be classified as causation appear to be  Philosophy of Science 13 rooted in social values, not scientific values. Indeed, social commitments appear to play a role in probabilistic causation. For instance, the associa- tion between exposure to low-dose ionizing radiation (such as from med- ical X-rays) and cancer is very low for most types of cancer in adults. Nevertheless, radiation protection is organized and based on a causal rela- tionship, as there is a strong social commitment to protect people under- going medical investigations. If, as seems to be the case, the limits of what can be called causation depend on social values, causation diffuses from the realm of science into the realm of society. 1.5.6 Counterfactual Conditions Another approach to causation highlights whether the presence or absence of a cause “makes a difference.” If Mr. Hanson had not been exposed to the hepadnavirus, he would not have got hepatitis B. If Mrs. Jones had taken two aspirins instead of just a glass of water an hour ago, her headache would now be gone. A counterfactual draws on the contrast between one outcome (the effect) given certain conditions (the cause), and another outcome given alternative conditions. C causes E if the same condition except C would result in a condition different from E, when all other conditions are equal. The last premise is called ceteris paribus (all other conditions kept equal). Counterfactual conditions seem similar to necessary conditions, as without the necessary condition, the event will not occur. However, counterfactuals support or undermine suppositions, and are acceptable or not acceptable, while necessary conditions are either true or false.12 Counterfactual conditions are often considered to be deterministic, but they can also be probabilistic, as in the following counterfactual: “if Mrs. Jones had taken two aspirins instead of just a glass of water an hour ago, she would be much less likely to have a headache now.” One of the challenges with counterfactuals is that in practice it is not easy to know or to assess what the case would have been if the situation had been different. Although epidemiological data and experiments may give clues about differences between groups, it is hard to know whether Mrs. Olsen would have avoided lung cancer if she had not smoked. Hence, it is difficult to satisfy the ceteris paribus condition, because the same individual cannot be observed in the exact same situation as both a smoker and a nonsmoker. Table 1.1 sums up the deterministic and probabilistic conceptions of causation commonly referred to in the life sciences, which are discussed in this chapter. 14 Research in Medical and Biological Sciences Table 1.1 Deterministic and probabilistic conditions of causation Deterministic conceptions of causation Probabilistic conceptions of causation Sufficient condition for an event Probabilistic Insufficient but necessary part of a Raised probability for an event sufficient and necessary condition Counterfactual INUS condition If the condition, C, had not been Counterfactual present, then it is less likely that If the condition, C, had not been the event, E, would have occurred, present, the event, E, would not i.e., C makes a difference with have occurred, i.e., C makes a respect to the probability of E. difference with respect to E. Altogether, this means that when scientists say that X causes Y, many meanings can be inferred, and much confusion could be avoided if these meanings were clearer. However, the fact that there is no one clear-cut concept of causation should not result in disappointment or despair; even in his time Aristotle identified four kinds of causes: material cause, formal cause, efficient cause, and final cause. Moreover, there are many other meanings of “cause,” of which researchers must also be aware. For exam- ple, consider the following five statements: 1. The heart’s capacity to pump causes the blood to circulate. 2. A gene causes Huntington’s disease. 3. Smoking causes lung cancer. 4. Aspirin causes Reye’s syndrome. 5. Professional interests cause (Norwegian) radiologists to be negative to expanding radiographers’ tasks (towards diagnostics). Here the word cause is used in many ways. In statement 1, “cause” denotes a mechanism, in statement 2 a sufficient condition, in state- ment 3 a significantly raised probability, in statement 4 a slightly raised (but socially important) probability, and in statement 5 “cause” denotes a reason. Correspondingly, causation may have many levels. If a small child dies in Nigeria, what was the cause of his death? Was it the diarrhea, or was it that the diarrhea was not properly treated? Was it the malnutrition, or was it that his parents did not have access to clean water? Was it the low income, the lack of education of the population, the poor infrastructure, international politics, or the history of colonial exploitation? In fact, all of these factors may be relevant (Figure 1.4).  Philosophy of Science 15 Figure 1.4 Web of causation. There are several levels to consider. Much confusion, and also the adverse effects of hype, could be avoided if scientists were more explicit and clear when they claim that they have “solved the riddle of cancer” or that “X causes (disease) Y.” 1.6 SCIENTIFIC EXPLANATION Causation, and the various accounts of causation discussed above, are inherent in scientific explanations. From the time of Aristotle, philoso- phers have realized that a distinction could be made between two kinds of scientific knowledge—roughly “knowledge that” and “knowledge why.” It is one matter to know that myocardial infarction is associated with certain kinds of pain (angina pectoris); it is a different matter to know why this is so. Knowledge of the former type is descriptive; knowl- edge of the latter type is explanatory, and it is explanatory knowledge that provides a scientific understanding of the world.18 How, then, can phenomena studied in the biomedical sciences be explained? For example, how does one explain the change in hematopoi- etic cell growth in a medium when the temperature changes? What criteria exist for something to be considered an acceptable scientific explanation? The standard answer to such questions is that something is 16 Research in Medical and Biological Sciences explained by showing how it is expected to happen according to the laws of nature (nomic expectability).19 Hematopoietic cell growth is explained by the laws that govern this growth as well as the initial conditions, including the type of medium, humidity, temperature, and pressure. Accordingly, a singular event is explained if (a description of) the event follows from law-like statements and a set of initial conditions. When a phenomenon is explained by deducing it from laws or law- like statements, the sequence of deductive steps is said to follow a deductive-nomological model (DNM), which turns an explanation into an argument where law-like statements and initial conditions are the pre- mises of a deductive argument (Box 1.6). In other words, a phenomenon is explained by subsuming it in a law. For this reason DNM is often referred to as “the covering law model of explanation.” One reason for the prominent position of DNM is its close relation to prediction. A deductive-nomological explanation of an event amounts to a prediction of its occurrence. However, DNM does present some challenges. One is that DNM allows for symmetry. For instance, certain conditions of a growth medium for cells (temperature, humidity, light, etc.) can be explained by the growth rate of hematopoietic cells (in this medium) given the same laws. This is in contrast with the desire for there to be asymmetry between cause and effect; that is, what is considered to be a cause leads to an effect, and not the other way round. Moreover, if the biomedical sciences can provide explanations only when phenomena subsume under deterministic laws of nature, it means that there are innumerable phenomena that cannot be explained. For instance, it is often stated that lung cancer can be explained by smoking, although there is no strict law stating which smokers will develop lung cancer. There is a straightforward solution to this problem, which entails replacing deterministic laws with probabilistic statements. This engenders BOX 1.6 The DNM of Explanation Premise 1: Initial conditions Type of medium, humidity, light, temperature Premise 2: Universal law(s) Laws of hematopoietic cell growth Conclusion: Event or fact to be explained Greater growth due to temperature increase  Philosophy of Science 17 the deductive-statistical model (DSM) of explanation, the form of which is shown in Box 1.7. DSM is a version of DNM that supports explanations of statistical regularities by deduction from more general statistical laws (instead of deterministic laws). However, DSM cannot explain singular events, such as Mr. Hanson recovering from bacterial meningitis after taking antibio- tics. DSM can only explain why persons taking antibiotics will recover (in general). In order to explain singular events in terms of statistical laws, one can refer to the inductive-statistical model (ISM) of explanation, which explains likely events inductively from statistical models (Box 1.8). Table 1.2 summarizes the traditional models of explanation, DNM, DSM, and ISM. Common to all these models is the idea that explanations are arguments (deductive or inductive) based on initial conditions and on BOX 1.7 The DSM of Explanation Premise 1: Initial conditions Having bacterial meningitis Premise 2: Statistical laws Taking antibiotics probably leads to recovery Conclusion: Event or fact to be explained Persons taking antibiotics will recover BOX 1.8 The ISM of Explanation Premise 1: Initial conditions Mr. Hanson has meningitis and takes antibiotics Premise 2: Probability (r) of event, given 1 The probability of recovery in such cases 5 r  1 Induction: Event or fact to be explained Mr. Hanson will recover Table 1.2 Models of explanations according to Salmon18 Laws Singular events General regularities Universal laws DNM DNM Statistical laws ISM DSM DNM, deductive-nomological model; ISM, inductive-statistical model; DSM, deductive-statistical model. 18 Research in Medical and Biological Sciences law-like statements, be they deterministic or statistical (nomic expectancy). The standard form of each such argument is: Premise 1: Initial conditions Premise 2: Law-like statement Implication: Event or fact to be explained Most explanations in the biomedical sciences appear to fit these models. Nevertheless, all these models of explanation present challenges. One is that arguments with true premises are not necessarily explanatory. For instance, if Mr. Hanson is a man and takes oral contraceptives (initial conditions), and if no man who takes oral contraceptives becomes pregnant (law), it leads deductively to the conclusion that Mr. Hanson will not become pregnant. According to DNM, oral contraceptive use explains why Mr. Hanson does not become pregnant, but this is intui- tively wrong, as the premises are explanatorily irrelevant. As already indicated, DNM permits symmetry. For example, DNM enables the use of plane geometry and the elevation of the sun to find the height of a flagpole from the length of its shadow, as well as to predict the length of the shadow from the height of the flagpole. However, as the length of its shadow clearly does not explain the height of the flag- pole, DNM does not present a set of sufficient conditions for scientific explanation. DNM, DSM, and ISM are the principal models relevant to the biomedical sciences, but represent only three of the many models that can be used for scientific explanation. The challenges they present in the way of relevance and symmetry have made some philosophers of science argue that explanations should be based on causation; to explain is to attribute a cause. According to a causal model of explanation, one must follow specific procedures to arrive at an explanation of a particular phenome- non or event: 1. Compile a list of statistically relevant factors. 2. Analyze the list by a variety of methods. 3. Create causal models of the statistical relationships. 4. Test the models empirically to determine which is best supported by the evidence. However, these procedures reveal some of the core challenges of demonstrating causation. Moreover, although stating that to explain a phenomenon is to find its cause is intuitively correct, it is not necessarily the case in practice. Indeed, David Hume (1711 1776) argued that  Philosophy of Science 19 causation entails regular association between cause and effect, though the conception of causation as regularity adds nothing to an explanation of why one event precedes another. Accordingly, Bertrand Russell (1872 1970) claimed that causation “is a relic from a bygone age, surviving, like the monarchy, only because it is erroneously supposed to do no harm.”20 Indeed, defining explanation in terms of causation would enhance our ability to predict, but not to understand phenomena.21 Accordingly, expla- nation entails more than referring to a cause—it invokes understanding, and thus one could argue it must include the laws of nature. 1.7 MODES OF INFERENCE The biomedical sciences tend to employ three modes of inference first set forth in 1903 by Charles Sanders Pierce (1839 1914): deduction, induction, and abduction. Deduction is inference from general statements (axioms, rules) to particular statements (conclusions) via logic. If all persons with type I (insulin-dependent) diabetes are known to have deficiencies in pancre- atic insulin production (rule), and Mr. D has type I diabetes (case), then Mr. D has deficiencies in pancreatic insulin production (conclusion). Induction is inference (to a general rule) from particular instances (cases). If all persons with deficiencies in pancreatic insulin production have symptoms of type I diabetes, and these persons are from the gen- eral population (i.e., they were not included in the study sample because of other deficiencies that could cause these symptoms), one can conclude that all persons with deficiencies in pancreatic insulin production have symptoms of type I diabetes. Abduction infers the best explanation. When a certain observation (case) is made, a hypothesis (rule) can be found that makes it possible to deduce a conclusion. If Mr. D has deficiencies in pancreatic insulin production, and all persons with type I diabetes have deficiencies in pancreatic insulin production, then Mr. D has type I diabetes. The crucial question is whether these modes of inference are valid. Deductive inference is knowledge-conservative; if the axioms are true, the conclusion is true. However, the ultimate issue is whether the axioms hold. Inductive and abductive inferences are both knowledge-enhancing, and are therefore called ampliative inference. The challenge with induc- tion and abduction is to justify the knowledge enhancement. In induction, inference is made from some cases (conclusion) to the general rule, and in 20 Research in Medical and Biological Sciences Figure 1.5 Modes of inference. abduction there could of course be other rules that could explain what is observed even better. Figure 1.5 illustrates the differences among these three modes of inference. 1.8 WHAT SCIENCE IS ABOUT The biomedical sciences are about this world and its biomedical phenom- ena. This raises the philosophical question, what is this world? Many scientists find this question odd, even irrelevant. Scientists deal with viruses, cells, molecules, and the effects of interventions, and it is clear to most researchers that cells exist and that they more or less correspond to scientific theories. However, there are innumerable examples throughout history of situations where convictions of the reality of the entities con- tained in different theories, such as phlogiston, proposed by the German physician and alchemist Johann Joachim Becher (1635 1682); ether; miasms; and the “cadaver poison,” identified by Ignaz Semmelweiss (1818 1865) have been replaced by new convictions and new entities. Therefore, how can one be sure that the world is as science portrays it, and how can changes in theories be explained? Scientific realists hold that successful scientific research enhances knowledge of the phenomena of the world, and that this knowledge is largely independent of theory. Furthermore, scientific realists hold that such knowledge is possible even when the relevant phenomena are not observable. According to scientific realism, there is good reason to believe what is written in a good, contemporary medical textbook, because the authors had solid scientific evidence for the (approximate) truth of the  Philosophy of Science 21 claims put forth about the existence and properties of viruses and cells and the effects of interventions. Moreover, there is good reason to think that such phenomena have the properties attributed to them in the text- book, independent of theoretical concepts in medicine. Consequently, scientific realism can be viewed as the scientists’ own philosophy of science. On the other hand, scientific antirealism holds that the knowledge of the world is not independent of the mode of investiga- tion. A scientific antirealist might say that photons do not exist, and that theories about them are tools for thinking. These theories explain observed phenomena, such as the light beam of a surgical laser. Of course, the energy emitted from a laser exists, as well as the coagulation, but the photons are held not to exist. The point is that there is no way to know if the world is independent of our scientific investigations and theories. There are several levels of scientific realism. A weak notion of scientific realism holds that a real world exists that is independent of scientific scrutiny, without advancing any claim about what this real world is like. A stronger notion of realism argues that not only does the real world exist independent of scientific inquiry, but it also has a structure that is indepen- dent of this inquiry. An even stronger notion of scientific realism holds that certain things, including entities in scientific theories such as photons and DNA, exist independent of humans and our scientific inquiry of the world. Accordingly, the scientific realist claims that when phenomena, such as entities, states, and processes, are correctly described by theories, they actually exist. Scientific realism is common sense, and certainly “common science,” as a researcher does not doubt that the phenomena he/she studies exist independent of their investigations and theories. However, how can this intuition be justified? This is where the philosophical challenges start. Three arguments justifying scientific realism are commonly advanced: tran- scendental, high-level empirical, and interventionist. The transcendental argument asks what the world must be like to make science possible. Its first premise is that science exists. Its second premise is that there must be a structured world independent of human knowl- edge of science. There is no way that science could exist, considering its complexity and extent, if the things science describes did not exist.22 Hence, the argument reasons from what one believes exists to the pre- conditions for its existence. Even if science is seen as a social activity, the same question applies. How can this activity exist without the precondi- tion that the world actually exists? Science is intelligible as an activity only if scientific realism is assumed. One premise of the transcendental 22 Research in Medical and Biological Sciences argument is that science expands our knowledge of the world and corrects errors, but how does one know this? Furthermore, how can one reason from what one believes to exist to the preconditions for its exis- tence? The answer is through thought experiments. The effects of certain microbiological events could not be pondered without the existence of DNA. Based on this one can argue that the existence of DNA is a neces- sary condition for microbiological events. But how can one be sure that the reason microbiological events cannot be pondered without the exis- tence of DNA is not simply due to the limits of scientific imagination? The high-level empirical argument contends that scientific theories are (approximately) true because they best elucidate the success of science. The best way to explain progress and success in science is to observe that (1) the terminology of mature sciences typically refers to real things in the world, and (2) the laws of mature sciences are typically approximately true.23 However, this is an abductive inference, in which one argues from the conclusion (science has success) and the rule (if science is about real things, then it has success) to the case (science is about real things). Abductive arguments are knowledge-expanding, therefore there may be other, better explanations that have not yet been discovered. The interventionist argument holds that one can have well-grounded beliefs about what exists based on what one can do.24 Intervention can be employed to test whether the entities contained in scientific theories actually exist. If an intervention on an entity from a theory does not work, the entity does not exist, but “if you can spray them, then they are real.”24 Hence, one can test whether something is real. One problem with the interventionist argument is that it is not robust with respect to explanation. If one were to test whether ghosts are real by spraying them with red paint, one may conclude that ghosts are not real. However, how could a person know that this is the correct method to show that ghosts are real? Could it not be that red paint does not adhere to ghosts, whereas yellow paint does? Scientific realism, which most scientists consider common sense, is irritatingly difficult to justify. We could, of course, dismiss the whole question about the existence of the entities in our theories by arguing that observable results are what matters, and the validity of the existence of said entities, be they photons or arthritis, does not matter. However, at a certain point a scientist may need to reflect upon the nature of being of the entities studied.  Philosophy of Science 23 1.9 SCIENTIFIC RATIONALITY Rationalism is the position that reason takes precedence over other ways of acquiring or justifying knowledge. Traditionally, rationalism has been contrasted with empiricism, which claims that true knowledge of the world can only be obtained through sensory experience. In antiquity, rationalism and empiricism referred to two schools of medicine, the former relying primarily on theoretical knowledge of the concealed work- ings of the human body, the latter relying on direct clinical experience. One might argue that the demarcation between rationalism and empiricism has become irrelevant in science, yet remains relevant in clini- cal practice. There are many examples of treatments established on rationalistic grounds, such as the ligation of arteria mammaria interna as a treatment for angina pectoris, that have been revealed by empirical studies to have no effect (beyond placebo). Hormone replacement therapy in postmenopausal women was said to prevent cardiovascular disease. Laying babies on their stomach was thought to prevent sudden infant death syn- drome because it would avoid suffocation from vomit that could occur if babies were put on their back. Some established treatments prescribed based on a physician’s experience have also been revealed to be without effect, or even detrimental. However, modern biomedical scientists tend to rely on rationality as well as experience. For example, hypotheses may be generated on rationalistic grounds (the substance S should have the effect E because it has the characteristics X, Y, and Z), and are then tested empirically, such as in animal models or in randomized clinical trials. Therefore, the enduring rationalism-empiricism debate still seems rele- vant in the biomedical sciences given the limitations of scientific method- ology. There may be ethical reasons, such as reluctance to use placebo surgery, that limit empirical research; there may be lack of knowledge of mechanisms that limits a rationalistic approach, such as when one wishes to test a substance that appears to have promise in eliciting a desired effect, but for which one lacks the knowledge of how it works. 1.10 HYPOTHESIS TESTING Whenever a theory appears to you as the only possible one, take this as a sign that you have neither understood the theory nor the problem which it was intended to solve. Karl Popper 24 Research in Medical and Biological Sciences The author of one of the most prominent Hippocratic writings, The Art (of medicine), identified three challenges in medical treatment and research: (1) the obtained effects may be due to luck or accident (and not intervention), (2) the obtained effect may occur even if there is no intervention, (3) the effect may not be obtained despite intervention. In the terminology of causation, researchers are faced with the challenges that the intervention is not a necessary condition (as in case 2), nor a sufficient condition (as in case 3) for the effect, and that there may be either a probabilistic relationship between the intervention and the effect, or other (unknown) conditions for the effect (as in case 1). Almost three millennia later, scientists still struggle with the same kind of question: how can one be certain that the theories and hypotheses of the world are true, given the large variety of possible errors? The standard solution to this problem is to test the hypothesis according to the hypothetical- deductive method. 1.10.1 Hypothetical-Deductive Method The hypothetical-deductive methodiv is the scientific method of testing hypotheses by predicting particular observable events, then observing whether the events turn out as predicted. If so, the hypothesis is verified, and if not, the hypothesis is falsified. The steps in the hypothetical- deductive method are: 1. State a clear and experimentally testable hypothesis. 2. Deduce the empirical consequences of this hypothesis. 3. Perform empirical experiments (in order to compare the results with the deduced empirical consequences). 4. If the results concur with the deduced consequences, one can con- clude that the hypothesis is verified, otherwise it is falsified. According to the traditional interpretation of the hypothetical- deductive method, hypotheses can be verified and scientific knowledge is accumulated through the verification of ever more hypotheses (Table 1.3). However, as Karl Popper (1902 1994) showed, the verification approach to hypothesis testing is flawed.2 First, the verification of a hypothesis presupposes induction. However, as already pointed out, reasoning from some (group) to many (the population) is not warranted. Second, the logical form of the model is not sound (Table 1.4). iv This is also frequently called the hypothetico-deductive method or model.  Philosophy of Science 25 Table 1.3 Simplified comparisons between the structure of verification and falsification Verification Falsification 1. Hypotheses A is better than B B is better than A 2. Deduced empirical If A is better than B, one If B is better than A, one consequences must observe that A must observe that B gives better results gives better results than B in the than A in the empirical setting empirical setting 3. Experiments and We observe that instances We observe that instances observations where A is used obtain where A is used obtain better results than B better results than B 4. Conclusion The experiment The experiment refutes confirms the the hypothesis and hypothesis lends support to the alternative hypothesis (A is better than B) Table 1.4 Logical comparisons between the structure of verification and falsification Verification Falsification Logical structure If p, then q If p, then q q Not q p (Confirming the Not p (Modus tollens) consequent) Mode of knowledge Accumulation of Exclusion of false acquisition knowledge, confirming hypotheses, narrowing hypotheses down and delimiting knowledge Moreover, Popper was critical of the lack of standard criteria for estab- lishing scientific truth that existed in the early twentieth century, and of the corresponding trend to employ (scientific) authority to decide what was true, which made it difficult to differentiate science from other social activities. Popper’s radical solution was to avoid stating explicit (authori- tative) criteria for truth and to provide stringent procedures for testing hypotheses. Furthermore, he broke with the ideal that truth could ultimately be determined, and instead strove to provide a scientific knowledge base of nontruths, or falsified hypotheses. For Popper, scien- tific knowledge progressed through enlarging the graveyard of falsified hypotheses. The method of falsification rather than verification makes all 26 Research in Medical and Biological Sciences truth provisional, conjectural, and hypothetical. According to Popper, experiments cannot determine theory, only delimit it, nor can theories be inferred from observations. Experiments only show which theories are false, not which theories are true (Figure 1.6) (Box 1.9). In empirical fields, the hypothetical-deductive method is used almost daily and often without a thought. The control experiment is a typical example. Can a possible effect or an absent effect have a trivial explana- tion? Might changes over time or in titrations of solvents produce effects, or might the cells have failed to respond? Control experiments are included to rule out such trivial explanations. Figure 1.6 Knowledge generation in verification versus falsification. BOX 1.9 Popper on “The Success of Refutation” “Refutations have often been regarded as establishing the failure of a scientist, or at least of his theory. It should be stressed that this is an inductivist error. Every refutation should be regarded as a great success; not merely a success of the scientist who refuted the theory but also of the scientist who created the refuted theory and who thus in the first instance suggested, if only indirectly, the refuting experiment. Even if a new theory (such as the theory of Bohr, Kramers, and Slater) should meet an early death, it should not be forgotten; rather its beauty should be remembered, and history should record our gratitude to it—for bequeathing to us new and perhaps still unexplained experimental facts and, with them, new problems; and for the services it has thus rendered to the progress of science during its successful but short life.”2  Philosophy of Science 27 In clinical trials involving new drugs, patient symptoms may be strongly influenced by the treatment situation, and a placebo may cause an effect; a placebo group is included to rule out (falsify) the placebo effect. Correspondingly, tests are conducted in a double-blind manner to falsify the hypothesis that the observed effect of a treatment is due to the expectations of the researcher. Correspondingly, statistical tests are performed to investigate whether an obtained result is due to selection bias, or other types of bias. These tests assess whether recorded differences between groups are random. This is done by formulating a null hypothesis, denoted H0, that states there is no difference between the groups, and thereafter assessing the probability that H0 is true. If that probability is very small, the null hypothesis is rejected, which strengthens the principal hypothesis that there is a real, instead of a random, difference. A hypothesis must have testable implications if it is to have scientific value. Popper contended that if a hypothesis it is not testable, and thus not falsifiable, it is not science. The lack of adequate methods often hinders scientific progress, because limited testability restricts the scope of topics that can be subjected to scientific inquiry. Therefore, great leaps in science are often made thanks to new, more powerful methods that open up new areas of research. Outstanding instances include Kary Mullis’s development of the polymerase chain reaction (PCR) in molecular biology, which was recognized by a Nobel Prize in 1993, and the devel- opment of the patch clamp in neurobiology by Erwin Neher and Bert Sakmann, which was recognized by a Nobel Prize in 1991. The development of hypotheses is closely associated with the develop- ment of models and the planning of experiments. Many hypotheses are too imprecise and ambiguous to be rejected, and consequently cannot be challenged as Popper requires. Formally, there should be two alter- native hypotheses that are mutually exclusive, and an experiment should be designed to distinguish between them. If a hypothesis is falsified, this may lead us to the development of new hypotheses, which in turn can be tested (Figure 1.7). Moreover, a scientific hypothesis should have the power to explain. It should relate to an existing, generally accepted theoretical basis of the field. There must be good grounds to reject established theories, such as an accepted law of nature. Whenever newer observations so indicate, it is advisable to modify a hypothesis. Modifications are acceptable if they make the hypothesis more testable. However, sometimes a theory that 28 Research in Medical and Biological Sciences Figure 1.7 Sketch of the hypothetical-deductive method. flounders on the grounds of falsifying experiments is defended by its remaining adherents by proposing ad hoc hypotheses that save the favored hypothesis by nullifying the negative observational evidence, not by mak- ing it more testable. While modified hypotheses are part of the ordinary scientific process, ad hoc hypotheses are not, as they hinder rather than promote scientific development. Although falsification has become common in empirical biomedical research, its strengths and weaknesses are not always appreciated. According to Popper, a theory or hypothesis should be bold and far-reaching. Its empirical content should be high; that is, it should have great predictive power. Furthermore, the hypothesis should be testable. If the results of empirical tests support the hypothesis, it is corroborated (but not verified); if not, it is falsified. Moreover, if one can choose among several hypotheses, it is better to select the simplest, according to what has been called Ockham’s razor, after the British Franciscan monk William of Ockham (1288 1347) (see also Chapter 4). Regardless of how influential Popper’s approach has been and continues to be, falsification in empirical research in the biomedical sciences has been severely criticized. Five of the most cited challenges of falsification are: 1. When one falsifies theories, their prospective robustness is not tested. They are only tested on past evidence. 2. A severe test is one which is surprising and unlikely based on present evidence. However, this means that knowledge of what is likely is used to set up a test that is unlikely, which necessitates the application of induction. Accordingly, if one really defies induction, there is no reason to act on corroborated theories or hypotheses, because doing so would be based on induction. 3. Falsification of a theory is based on empirical observations. However, observational statements should also be fallible. Hence, the falsification  Philosophy of Science 29 of a theory may be erroneous if the observational statements are not true. 4. Popper’s method can lead to falsification of robust and fruitful theories with high empirical content, e.g., due to errors in the test procedure. 5. In practice one does not falsify single theories but rather groups or whole systems of hypotheses. One reason why one cannot test single hypotheses in isolation is that a series of background assumptions is made to test the hypothesis (the Quine-Duhem thesis). Any of these background assumptions may fail when one tries to falsify a hypothesis. 6. Moreover, in practice one does not falsify a potentially fruitful theory based on a single observation or study. Instead, new experiments are conducted and ad hoc hypotheses generated to investigate or explain the falsifying observation. 1.11 THE AIM OF SCIENCE: REDUCING UNCERTAINTY The primary aim of science is to increase knowledge in order to explain, understand, and intervene. Scientific knowledge is necessary to reduce our uncertainty. It is practical to differentiate among four kinds of uncer- tainty: risk, uncertainty, ignorance, and indeterminacy (Table 1.5). Risk is when the system behavior is basically well known, and the chances of different outcomes can be defined and quantified by structured analysis of mechanisms and probabilities. One of the tasks of science is to find the outcomes of a given situation or intervention, as well as the probability of these outcomes. An example is the incidence of cardio- vascular disease among patients with type II (noninsulin-dependent) diabetes practicing prophylactic statin use. Uncertainty is characterized by knowledge of the important system parameters but not of the probability distributions. In this case the major outcomes of a certain intervention may be known, but not their respec- tive probabilities. There may be many sources of uncertainty in a study, Table 1.5 Modes of uncertainty25 Outcome Known Unknown Known Risk Indeterminacy (Ambiguity) Probability Unknown Uncertainty Ignorance 30 Research in Medical and Biological Sciences including uncertainty in reasoning (i.e., how to classify a single case with regards to general categories) or uncertainty in biomedical theory (i.e., when all mechanisms in a certain field are not known in detail, or because of multifactor causation). Moreover, diseases are complicated, and it can be difficult to know and understand all their causes (see Figure 1.4 above for an example). In the case of uncertainty, the main task of science is to provide the probability distributions, to estimate risks, and to provide sensitivity analyses. Ignorance describes a situation in which neither the outcomes nor their probabilities are known. One of the tasks of science is, of course, to find both, but when it comes to ignorance this is difficult, as a scientist does not know what he does not know. For example, the effect of thalidomide taken by pregnant women on fetuses and children (severe limb deformation) was difficult to foresee. The role of science is to reduce ignorance while being observant of and attentive to the unexpected. Indeed, it took an unnecessar- ily long time to discover the detrimental effects of thalidomide. Even if scientists are successful in reducing ignorance to uncertainty and all cases of uncertainty to risk, they might still be subject to indeterminacy. It is not always a question of uncertainty due to imprecision (which is assumed to be narrowed by further research) but also a question of the properties or cri- teria used to classify things. When myocardial infarction is classified accord- ing to a set of clinical criteria, the resultant perspective will be different than if it is classified according to the level of troponin in the blood. Likewise, if pain is being investigated in terms of neural activity or according to a visual analogue scale, the risk, uncertainty, and ignorance may differ. Hence, pain may be ambiguous. Accordingly, processes may not be subject to predictable outcomes from given initial conditions, due to imprecise classifi- cation. Table 1.5 summarizes and compares the four modes of uncertainty. 1.12 THE EMPIRICAL TURN IN THE PHILOSOPHY OF SCIENCE: SCIENCE IN SOCIETY Although many of the previously mentioned challenges in the traditional philosophy of science have been addressed and progress has been made, interesting and fruitful contributions have been fueled through empirical studies of science and scientific practices. Close empirical studies have revealed social aspects that are typical in science. In particular, the norms and activities of scientists have been shown to be basically similar to the norms and activities of other societal groups.26 Science is but one of  Philosophy of Science 31 many socially organized activities that generates knowledge; studying science as a socially organized activity and comparing it to other such activities has given important insights into the field. Where the philosophy of science has traditionally been theoretical and has focused principally on the products of science, that is, knowledge and its conceptual preconditions, more recent approaches are empirical and focus on the social processes of science (and its interaction with material matters). A seminal and famous study of scientific activity showed that knowledge is not accumulative and that science does not develop in a linear manner.3 Instead, it evolves in an abrupt way (scientific revolutions) with intervening quiescent periods. Inspired by Kuhn’s paradigmatic conception of scientific progress, and by Wittgenstein’s theories on rule-following and language games,27 a series of studies of science, termed the Sociology of Knowledge (SoK) movement, emerged. The key goal was to show that science is a social activity that follows the same social patterns as other activities in society. The question of how things are in the world cannot be addressed without questioning how the social group comprising scientists conceives of and handles these things. It follows that things, be they photons or DNA, cannot be attributed a role in our world independent of symbols and meaning. Hence, while the traditional philosophy of science had proce- dural criteria for demarcating science from nonscience, such as Popper’s criteria for falsification, the SoK movement applies social criteria. Whereas the normative aim of traditional studies of science was to free science from the authority and power inherent in the social structures among scientists and in society, the SoK movement strives to disclose power within the scientific society and to emancipate. In many respects, the key issue in the traditional philosophy of science has been the relationship between scientific theories and nature. In the SoK movement the focus is on the relationship between theory and culture: in what way do scientific theories reflect social structures (instead of the structures of the world)? Nevertheless, what appears to be similar in both the traditional philosophy of science and the SoK movement is the focus on epistemological issues. In both cases the key question is: what do scientific theories represent? In the first case, scientific theories represent patterns in nature; in the second, they represent social structures. In the former case theories about DNA represent biomolecular entities in cells, whereas in the latter case they represent social activities. In both cases there is something behind the theories; something that the theories represent. 32 Research in Medical and Biological Sciences Later studies of science have tried to avoid this representational pattern, i.e., the belief that there is something behind a theory. Instead of investi- gating the relationship between theories and the social processes and struc- tures in science, the scientific process itself was studied, including its material premises. How do scientists behave, and how do they produce the facts of science? For example, Bruno Latour studied how the daily activities and processes of scientists at the Salk Institute contributed to the establish- ment of scientific facts about the thyrotropin-releasing factor (TRF), e.g., that it is a peptide.28 This can be referred to as a processual approach, accord- ing to which science is the change as it restructures, makes new, and stabi- lizes things and theories. What characterizes the social process of science is an interaction of methods, material, activities, and processes, where negotia- tions lead to the stabilization and generation of facts. When species of the Helicobacter pylori bacteria were found to be associated with gastric and peptic ulcers, scientific debate ensued on the bases of the embedded, previ- ously accepted theories, until negotiations based on continued empirical work established Helicobacter pylori as a key factor. It is not a question of what the theory represents (either nature or culture) but rather a question of negotiation between different scientific groups with regards to what will be considered fact. Hence, according to the processual approach the issue is not the relationship between theory and nature/culture (epistemological and representational) but what scien- tists regard and treat as real (ontological and processual). It is not a ques- tion of what is behind prions in nature or in culture but the scientific practices and processes that constitute facts about prions. 1.13 PHILOSOPHY OF THE SOCIAL SCIENCES A significant part of the overall spectrum of health-care problems consti- tutes matters that are not principally biological. Should one wish to find out why patients do not take prescribed medications, why incorrect med- ications are sometimes administered in hospitals, or why it is difficult to attain fully informed consent for trials or treatment, one cannot search for answers in human biological research; instead one must turn to the methods of the social sciences. For this reason it is essential to know how the philosophies of the social sciences and the biological sciences differ, so one does not errone- ously use the criteria for one area to judge another. In the social sciences, many different methods are used, and there are various schools of theory.  Philosophy of Science 33 The following comprises only a brief introduction and does not cover the broad scope of methods or schools of theory. 1.14 INTERPRETATION, UNDERSTANDING, AND EXPLANATION The social sciences differ from the biological sciences in two respects: 1. They entail greater elements of overt interpretation that often enter into the collection of data. 2. In many cases, a research result is an understanding, not an explanation. 1.14.1 Explanation and Understanding The principal goal in the biological sciences is to elicit causal explanations of the phenomena studied, for instance, the cause of a particular manifes- tation of a disease. Some projects in the social sciences also seek causal explanations of social phenomena, but many seek an understanding instead. Understanding is a form of knowledge that enables us to know why a person or a group behaves in a particular way, why and how they experience a specific situation, how they themselves understand their way of life, etc. We attain understanding through interpretation. The distinction between explanation and understanding was first expressed by the German philosopher, psychologist, and educator Wilhelm Dilthey (1833 1911), who believed that these two ways of viewing the world were characteristic of the natural sciences and the human sciences (Geisteswissenschaften), respectively. However, the differ- ence between explanation and understanding is not as distinct as many believe. Often theories of the social sciences include elements of both causal explanation and noncausal understanding. For instance, when one explains why the poor have greater morbidity than the rich, one usually refers to both causal factors, like the greater physical risks present in poor areas, and noncausal factors, like the influence of the working-class culture on lifestyle choices. 1.14.2 Interpretation All content-bearing objects and statements can be interpreted. People express themselves not just in speech, writing, and deeds but also in archi- tecture, garden design, clothing, etc. If, for example, one investigates where and why institutions for psychologically ill patients were built, one will find that the history reflects varying understandings of psychological 34 Research in Medical and Biological Sciences illness. The architecture of the asylum is also content-bearing. However, in this chapter the focus will be on the interpretation of texts and other linguistic statements, as it is germane in the discussion of the theory of interpretation, often called hermeneutics. Interpretation may have many goals, but in general, through interpre- tation one seeks to understand the information put forth in content- bearing material. The various theories of interpretation are based on differing concepts of the nature of content and how it should be identi- fied. Is there content in a statement itself, in the thoughts of the person making the statement, in the social structure in which the statement is made, etc.? These differences are germane when analyzing the validity of specific methods of the social sciences but are of lesser importance here in our general discussion of interpretation. The question of whether one ever obtains a true interpretation of a text is an old one. All written religions have sets of hermeneutic rules for interpreting the content of their holy texts. For example, in Christian theology, biblical exegesis concerns the interpretation of the scriptures. In modern times, interest arose in the interpretation of secular state- ments, first as part of literary and historical research, and then as a part of research in the social sciences. The goals of the various hermeneutic methods that have been developed are to arrive at an understanding of content that can be defended as a valid, intersubjective understanding; that is, an understanding that can be substantiated and discussed rationally. As Popper pointed out, the elements of interpretation enter into all observations and thereby into all forms of science. Humans lack direct access to the world “as it is” through our senses. We always view the world through a theoretical filter, and all observations are theoretically loaded. For example, to say that the Sun rises is to reflect the influence of the old geocentric worldview in which the Sun circled the Earth. And the “description” that a pathologist gives of a histological preparation seen through a microscope is to a large extent an interpretation based on theories of cells, of inflammation, etc. Biological research findings are interpreted in a similar manner; no P-value speaks for itself. 1.15 THE HERMENEUTIC CIRCLE, HORIZON OF UNDERSTANDING, AND “DOUBLE HERMENEUTICS” The hermeneutic interpretation of a text rests on each individual part, as well as on the understanding of how these individual parts are related to  Philosophy of Science 35 the whole. Neither an individual part nor the whole text may be inter- preted without reference to each other. Therefore, interpretation is circu- lar; the hermeneutic circle. In principle, this circle cannot lead to certain closure, as one will never know if a deeper analysis of the text may change its interpretation. The problem of attaining valid, intersubjective interpretations has long been and continues to be discussed, and optimis- tic interpretation theoreticians speak of a hermeneutic spiral, which implies that interpretation gets better and better. At the pragmatic level, the problem of the hermeneutic circle is less worthy of attention, since agreement on the meaning of a text can usually be attained. The concrete interpretation is also influenced by the interpreter’s “horizon of understanding,” a concept introduced in Wahrheit und Methode, the principal work of German philosopher Hans-Georg Gadamer (1900 2002). Gadamer argues that before one has begun a conversation with another person or begun to interpret a text, one already has pre- conceptions about it based on one’s horizon of understanding, a collec- tive term for one’s worldview. The horizon of understanding is built throughout life and comprises one’s understanding of particular words, the connotations that particular words and concepts hold, and so on. For a resident of London, the word city connotes financial affairs, while for people elsewhere it simply connotes an urban concentration of popula- tion. This difference in connotations means that two people engaging in a conversation may believe that they have understood each other without actually having done so. Full understanding is possible only when the two people in the conversation have acquired each other’s horizons of under- standing, or have “fused horizons.” This explains the problem of interpretation, as in interviews in the social sciences that often are too short for the interviewer to understand the interviewee’s horizon of understanding. Consequently, a vital part of the interview comprises an effort to learn how the interviewee uses and understands words and concepts in the area being discussed. Furthermore, English sociologist Anthony Giddens (1938 ) pointed out that social science research comprises “double hermeneutics.”29,30 In reality, social science research interprets interviewees’ interpretations of their own understandings, and parts of their understandings arise through concepts that they have acquired from the theories of the social sciences (such as the Marxist concept of class or the incest taboo of psychology). Hence, there is a complex interaction between the interpretations of the researcher and the interviewee, which is why an additional level of 36 Research in Medical and Biological Sciences interpretation may often be needed to focus on how an interviewee’s self- image is affected by the theories of social science. An interviewee may be misunderstood if the interviewer does not take such reflections into account. 1.16 POWER, IDEOLOGY, AND INTERESTS One’s interpretations of the statements and deeds of others are influenced by aspects other than one’s horizon of understanding. The German philosopher Jürgen Habermas (1929 ) pointed out that power, ideology, and interests also play leading roles. Usually, one is not a neutral or objec- tive observer; instead, one interprets according to the power of one’s position, ideology, and the interests one wishes to further.31 In Habermas’s view, ideology is not restricted to political ideology. An ideology is simply a set of assumptions that furthers the interests of a particular group in society. For example, the assertion that “an extensive hospital system is essential in health care” is an apolitical ideology that, in addition to safeguarding the interests of patients, furthers the interests of doctors and other health-care professionals. One of the difficulties of ideologies is that they are often concealed, as researchers are neither aware that they have them, nor of where they came from. So ideologies can influence actions and interpretations “behind our backs,” so to speak. Consequently, Habermas maintains that the princi- pal task of the critical social sciences is to identify prevailing ideologies so one may be freed from them. 1.17 VALIDITY In the above, widely recognized problems have been discussed: that an interpretation and the understanding attained through interpretation can never be a final truth about the meaning of a particular statement, unless the statement is extremely simple. Therefore, one is obliged to ask how one can judge the validity of a scientific interpretation. The simple answer is that if a researcher is aware of these problems and has taken the best possible steps to avoid or avert them, such as by trying to identify which ideologies and interests have influenced the various elements of the research process, there are grounds to rely on the interpretation—not because it is necessarily true, but because it constitutes a well-founded hypothesis without significant sources of error in the research process.  Philosophy of Science 37 1.18 REDUCTIONISM AND EMERGENCE Some biological researchers contend that there is no need for social scientific interpretation, as in the final analysis all knowledge can be reduced to facts about physical conditions. Social phenomena can be reduced to group psychology, which in turn can be reduced to individual psychology, which in turn can be reduced to neurology, which in turn can be reduced to cellular biology, and so on, until one reaches the physical level at which pre- vailing physical laws provide explanations for all phenomena observed at higher levels. This view, called reductionism, is in strong dispute. It is crucial to distinguish between methodological reductionism and general reductionism. In some research projects, methodological reasons may dictate the exploration of one or more factors that can influence the phenomenon of interest, without indicating that other factors are unim- portant. There are no methods that can acquire data and at the same time investigate “the whole.” Our attention must be focused on something more specific. Methodological reductionism can be meaningful and nec- essary, whereas general reductionism has been refuted. If, for instance, one wishes to examine a biological relationship, it may be necessary to ignore an ancillary social relationship. Conversely, if one examines a social relationship, it may be necessary to ignore a biological relationship. Methodological reductionism in itself is straightforward, as long as the factors examined are sensible. It becomes problematic only when a set of factors is systematically excluded, such as the correlation between poverty, social deprivation, and disease. There are many arguments against reducing social phenomena to physical phenomena. Two of them will be summarized here. The first problem confronting the reductionist is that it is doubtful that individual psychology can be reduced to neurophysiological processes. Dispute per- sists on the precise description of the relationship between psychological phenomena and cerebral activity; scientists are no closer to solving the “mind-brain” riddle today than they were a century ago. If this link in the reductionistic chain fails, reductionism as a whole cannot be carried out. The other problem for the reductionist is that many social phenom- ena are emergent, that is, they are socially not reducible as they occur at particular social levels and have no meaning when reduced to lower levels (individual psychology, neurology, etc.). Paper money, for example, is an emergent social phenomenon. A d10 banknote has no value in itself (save the value of the portrait of scientist 38 Research in Medical and Biological Sciences Charles Darwin). It cannot be exchanged for gold or other objects of value at the central bank. But it is integrated in social relationships that enable it to be exchanged for goods or services worth ten pounds. Otherwise, it is just a small, rectangular scrap of paper. Emergence at the social level may also be ascribed to a particular set of social conventions or formalized laws. For instance, most societies have the institution of marriage, but the concrete implications of being married and the social effects of it vary from society to society. The human penchant to form pair relationships might be reduced to the biological level, but the con- crete institution of marriage in a particular society cannot be similarly reduced. However, it is clear that the concrete, nonreducible institution of marriage affects human actions and considerations, so a full description of these actions and considerations is possible only at the social level. Similarly, in health care the professions of doctor, nurse, etc. have differ- ent roles. These roles are not naturally given but are a result of social con- vention. Despite this, it is important to understand these specific roles if, for instance, one is doing health services research aimed at improving the function of a particular type of hospital ward. If the antireductionists are right, scientific effort in the social sphere is useful and may employ methods that differ from those applicable at lower levels. 1.19 GENERALIZATION Statistics are often useful in research projects that employ quantitative methods. Whenever a study sample is selected from a well-defined source population, statistical inference to the source population is performed, for instance, by calculating the statistical confidence. Inferences to a wider population, the target population, are then made. In principle, this implies that results from research conducted in the United States may be directly applicable to treatment choices in Norway. However, it is worth noting that such generalization of results is only acceptable when there are grounds to assume that the populations are in fact similar. In the pre- vious example this would mean assuming that there is no biological difference between Americans and Norwegians. Generalization may be employed in much the same way in quantita- tive social science research, but statistical methods cannot be employed in research that is not quantitative. Does this imply that understanding  Philosophy of Science 39 attained in social science research cannot be generalized? The answer would be yes if statistical generalization were the only form of generaliza- tion available. But there is a form of generalization that is not quantitative and is frequently employed across all the sciences. It is called theoretical or conceptual generalization, sometimes called transferability. We often generalize, not in exact numbers, such as the cure rate for a particular drug, but rather within a conceptual or a theoretical frame of understand- ing. For instance, when teleological explanations based on the theory of evolution are used in biology, they rest upon a theoretical generalization of the theory of evolution, not upon a statistical generalization. Social scientific concepts and theories may be generalized in the same manner. In all forms of generalization, both statistical and conceptual, it is impor- tant to keep in mind that conditions change with time. Generalizations that were once valid can be rendered invalid if there are changes in the support- ing biological conditions, such as the resistance patterns in bacteria or the structures of families. QUESTIONS TO DISCUSS 1. What makes your research scientific and subject to funding above other social knowledge-generating activities? Mention and assess four demarcation criteria for science. 2. How do you know that DNA actually exists? 3. Can you be more or less sure that RNA exists compared with that pain exists? Why? 4. Discuss four types of uncertainty mentioned in this chapter. Do they apply to your research project, and what can you as a scientist do to reduce them? What should you do if you are not able to reduce them? 5. Describe the type of causality that is relevant to your research project. What are the pros and cons of this conception of causality? 6. Which kind of scientific explanation is prominent in your discipline? 7. Give an example of abduction from your field of research. 8. What is the problem with verification, according to Karl Popper? 9. What are the challenges of falsification? 10. Mention four criteria for truth. Discuss the pros and cons of each of them. Which conception of truth is prevalent in your field of research?