Introduction:the nature of science 9 justified by the evidence,but this justification is inductive only in the former,broader sense.We need to distinguish between these two senses:I am happy to keep using the word "induction"with the broader sense;the narrow sense I will indicate by the phrase "Humean induction",for reasons that will become clear shortly. We saw earlier on that some forms of knowledge seem to be obtainable just by pure thinking while others require the making of observations and the gathering of data. Philosophers mark this distinction by the use of the terms a priori and a posteriori.A priori knowledge is knowledge that can be gained without the need for experience.(To be precise,a priori knowledge is knowledge that can be gained without any experience beyond that reguired to acguire the relevant concepts.This additional proviso is required because,without some experience,one may not be able to have the thought in question.) A posteriori knowledge is knowledge that is not a priori.So to have a posteriori knowledge experience is required.Pure mathematics and logic are usually taken to be examples of a priori knowledge,whereas most of chemistry and biology are a posteriori. Another concept used in this connection is that of an empirical proposition.Empirical propositions are those the truth or falsity of which can only be known a posterifori. If we take a typical generalization of science-all mammals possess a muscular diaphragm-we can see that it is empirical.Knowing what this proposition means is not sufficient for knowing whether it is true or false.To know that,we would have to go and examine at least some mammals.Then we would have to infer from our observations of a limited range of mammals that all,not just the ones we have seen,possess a muscular diaphragm.This is an inductive inference. It is often said that a priori knowledge is certain,and by implication empirical propositions like those discussed cannot be known with certainty."Certainty"is a slippery concept and is related to an equally slippery concept,"probability",which I will discuss in Chapter 6.One reason why a priori knowledge is thought to be certain in a way that a posteriori knowledge of empirical generalizations is not,is that the former is gained through deductive reasoning while the latter requires inductive reasoning.As we saw above,if the premises of a deductively valid argument are true,the conclusion must be true too,while the premises of an inductive argument might be true yet the conclusion still false.But we need to be careful here,for as we use the word "certain",we are certain of many empirical propositions:that we are mortal,that the water at 30C in my bath tonight will be liquid,and that this book will not turn into a pigeon and fly away.If someone asks why we are certain we may appeal to our past experience.But past experience does not entail these claims about the future.Whatever the past has been like, it is logically possible that the future will be different.One might argue from some natural law.It is a law of nature that people will die and that water at 30C is liquid.How are we certain about these laws of nature?Here again it is because we have reasoned on the basis of our experience that these are among the laws of nature.But this experience does not entail the existence of a law.For it may have been sheer chance that so many people have died,not a matter of law.So to reason that there is such a law and that we too are mortal is to reason inductively,not deductively.Yet it is still something of which we are certain,and in this sense induction can give us inductive certainty. So far I have said nothing to impugn the integrity of inductive argument;all I have done is to draw some rough distinctions between deductive and inductive reasoning.The fact that deductively valid arguments entail their conclusions in a way that inductive
justified by the evidence, but this justification is inductive only in the former, broader sense. We need to distinguish between these two senses: I am happy to keep using the word “induction” with the broader sense; the narrow sense I will indicate by the phrase “Humean induction”, for reasons that will become clear shortly. We saw earlier on that some forms of knowledge seem to be obtainable just by pure thinking while others require the making of observations and the gathering of data. Philosophers mark this distinction by the use of the terms a priori and a posteriori. A priori knowledge is knowledge that can be gained without the need for experience. (To be precise, a priori knowledge is knowledge that can be gained without any experience beyond that required to acquire the relevant concepts. This additional proviso is required because, without some experience, one may not be able to have the thought in question.) A posteriori knowledge is knowledge that is not a priori. So to have a posteriori knowledge experience is required. Pure mathematics and logic are usually taken to be examples of a priori knowledge, whereas most of chemistry and biology are a posteriori. Another concept used in this connection is that of an empirical proposition. Empirical propositions are those the truth or falsity of which can only be known a posteriori. If we take a typical generalization of science—all mammals possess a muscular diaphragm—we can see that it is empirical. Knowing what this proposition means is not sufficient for knowing whether it is true or false. To know that, we would have to go and examine at least some mammals. Then we would have to infer from our observations of a limited range of mammals that all, not just the ones we have seen, possess a muscular diaphragm. This is an inductive inference. It is often said that a priori knowledge is certain, and by implication empirical propositions like those discussed cannot be known with certainty. “Certainty” is a slippery concept and is related to an equally slippery concept, “probability”, which I will discuss in Chapter 6. One reason why a priori knowledge is thought to be certain in a way that a posteriori knowledge of empirical generalizations is not, is that the former is gained through deductive reasoning while the latter requires inductive reasoning. As we saw above, if the premises of a deductively valid argument are true, the conclusion must be true too, while the premises of an inductive argument might be true yet the conclusion still false. But we need to be careful here, for as we use the word “certain”, we are certain of many empirical propositions: that we are mortal, that the water at 30°C in my bath tonight will be liquid, and that this book will not turn into a pigeon and fly away. If someone asks why we are certain we may appeal to our past experience. But past experience does not entail these claims about the future. Whatever the past has been like, it is logically possible that the future will be different. One might argue from some natural law. It is a law of nature that people will die and that water at 30°C is liquid. How are we certain about these laws of nature? Here again it is because we have reasoned on the basis of our experience that these are among the laws of nature. But this experience does not entail the existence of a law. For it may have been sheer chance that so many people have died, not a matter of law. So to reason that there is such a law and that we too are mortal is to reason inductively, not deductively. Yet it is still something of which we are certain, and in this sense induction can give us inductive certainty. So far I have said nothing to impugn the integrity of inductive argument; all I have done is to draw some rough distinctions between deductive and inductive reasoning. The fact that deductively valid arguments entail their conclusions in a way that inductive Introduction: the nature of science 9
Philosophy of science 10 arguments do not,may suggest that induction is inferior to deduction.This would be misleading.Consider the deductive inference:all colitis sufferers are anaemic,therefore this colitis sufferer is anaemic.You can see that the conclusion "this colitis sufferer is anaemic"says nothing more than is already stated in the premise "all colitis sufferers are anaemic".To say all colitis sufferers are anaemic is to say,among other things that this colitis sufferer is anaemic.So logical certainty goes along with saying nothing new.But in saying something about the future we will be adding to what we have already said about the past.That is what makes induction interesting and valuable.There would be no point in wanting induction to give us logical certainty,for this would be tantamount to wanting it to tell us nothing new about the future but only to restate what we already knew about the past. Hume's problem There is nonetheless a well-known difficulty associated with induction.The problem is this:Can these inductive arguments give us knowledge?This dilemma was first identified by the great Scottish historian and philosopher David Hume (1711-76)whose views on causation and induction,as we shall see,have had a very strong influence on the development of the philosophy of science.Although the dilemma is framed in terms of Humean induction,it is in fact perfectly general and raises problems for any form of induction.(Hume himself posed the problem in terms of causation.)It proceeds as follows.If an argument is going to give us knowledge of the conclusion,then it must justify our belief in the conclusion.If,because we have observed all colitis patients hitherto to have anaemia,we argue that all colitis sufferers are anaemic,then those observations along with the form of the inductive argument should justify the belief that all colitis sufferers are anaemic. How does this justification come about?On the one hand,it cannot be the sort of justification that a deductive argument gives to its conclusion.If it were,the conclusion would be a logically necessary consequence of the premises.But,as we have seen,this is not the case;in inductive arguments it is logically possible for the conclusion to be false while the premises are true.The very fact that inductive arguments are not deductive means that the sort of justification they lend to their conclusions cannot be deductive either. On the other hand,there seems to be a problem in finding some non-deductive kind of justification.For instance,it might be natural to argue that it is justifiable to employ induction because experience tells us that inductive arguments are,by and large, successful.We inevitably employ inductive arguments every day.Indeed all our actions depend for their success on the world continuing to behave as it has done.Every morsel we eat and every step we take reflect our reliance on food continuing to nourish and the ground continuing to remain firm.So it would seem that our very existence is testimony to a high degree of reliability among inductive arguments.But,to argue that past success in inductive reasoning and inductive behaviour is a reason for being confident that future inductive reasoning and behaviour will be successful,is itself to argue in an inductive manner.It is to use induction to justify induction.In that case the argument is circular
arguments do not, may suggest that induction is inferior to deduction. This would be misleading. Consider the deductive inference: all colitis sufferers are anaemic, therefore this colitis sufferer is anaemic. You can see that the conclusion “this colitis sufferer is anaemic” says nothing more than is already stated in the premise “all colitis sufferers are anaemic”. To say all colitis sufferers are anaemic is to say, among other things that this colitis sufferer is anaemic. So logical certainty goes along with saying nothing new. But in saying something about the future we will be adding to what we have already said about the past. That is what makes induction interesting and valuable. There would be no point in wanting induction to give us logical certainty, for this would be tantamount to wanting it to tell us nothing new about the future but only to restate what we already knew about the past. Hume’s problem There is nonetheless a well-known difficulty associated with induction. The problem is this: Can these inductive arguments give us knowledge? This dilemma was first identified by the great Scottish historian and philosopher David Hume (1711–76) whose views on causation and induction, as we shall see, have had a very strong influence on the development of the philosophy of science. Although the dilemma is framed in terms of Humean induction, it is in fact perfectly general and raises problems for any form of induction. (Hume himself posed the problem in terms of causation.) It proceeds as follows. If an argument is going to give us knowledge of the conclusion, then it must justify our belief in the conclusion. If, because we have observed all colitis patients hitherto to have anaemia, we argue that all colitis sufferers are anaemic, then those observations along with the form of the inductive argument should justify the belief that all colitis sufferers are anaemic. How does this justification come about? On the one hand, it cannot be the sort of justification that a deductive argument gives to its conclusion. If it were, the conclusion would be a logically necessary consequence of the premises. But, as we have seen, this is not the case; in inductive arguments it is logically possible for the conclusion to be false while the premises are true. The very fact that inductive arguments are not deductive means that the sort of justification they lend to their conclusions cannot be deductive either. On the other hand, there seems to be a problem in finding some non-deductive kind of justification. For instance, it might be natural to argue that it is justifiable to employ induction because experience tells us that inductive arguments are, by and large, successful. We inevitably employ inductive arguments every day. Indeed all our actions depend for their success on the world continuing to behave as it has done. Every morsel we eat and every step we take reflect our reliance on food continuing to nourish and the ground continuing to remain firm. So it would seem that our very existence is testimony to a high degree of reliability among inductive arguments. But, to argue that past success in inductive reasoning and inductive behaviour is a reason for being confident that future inductive reasoning and behaviour will be successful, is itself to argue in an inductive manner. It is to use induction to justify induction. In that case the argument is circular. Philosophy of science 10
Introduction:the nature of science 11 Hume's point can be put another way.How could we improve our inductive argument so that its justification would not be in question?One way to try would be to add an additional premise that turns the argument into a deductive one.For instance,along with the premise "all observed cases of colitis have been accompanied by anaemia"we may add the premise "unobserved cases resemble observed cases"(let us call this the uniformity premise).From the two premises we may deduce that unobserved colitis sufferers are anaemic too.We are now justified in believing our conclusion,but only if we are justified in believing the uniformity premise,that unobserved cases resemble observed ones.The uniformity premise is itself clearly a generalization.There is evidence for it-cases where something has lain unobserved and has at length been found to resemble things already observed.The lump of ice I took out of the freezer an hour ago had never hitherto been observed to melt at room temperature(16C in this case).but it did melt,just as all previous ice cubes left at such a temperature have been observed to do.Experience tells in favour of the uniformity premise.But such experience can only justify the uniformity premise if we can expect cases we have not experienced to resemble the ones we have experienced.That expectation is just the same as the uniformity premise.So we cannot argue for the uniformity premise without assuming the uniformity premise.But without the uniformity premise-or something like it-we cannot justify the conclusions of our inductive arguments.Induction cannot be made to work without also being made to be circular.(Some readers may wonder whether the uniformity premise is in any case true.While our expectations are often fulfilled they are frequently thwarted too.This is a point to which we shall return in Chapter 5.) If the premises of an inductive argument do not entail their conclusions,they can give them a high degree of probability.Might not Hume's problem be solved by probabilistic reasoning?I shall examine this claim in detail in Chapter 6.But we can already see why this cannot be a solution.Consider the argument "all observed colitis suffers are anaemic, therefore it is likely that all colitis sufferers are anaemic".This argument is not deductively valid,so we cannot justify it in the way we justify deductive arguments.We could justify it by saying that arguments of this kind have been successful in many cases and are likely to be successful in this case too.But to use this argument is to use an argument of the form we are trying to justify.We could also run the argument with a modified uniformity premise:a high proportion of unobserved cases resemble observed cases.But again,we could only justify the modified uniformity premise with an inductive argument.Either way the justification is circular. Goodman's problem There is another problem associated with induction,and this one is much more recent in origin.It was discovered by the American philosopher Nelson Goodman.Hume's problem concerns the question:Can an inductive argument give us knowledge (or justification)?If the argument employed is right,it would seem that those arguments we would normally rely upon and expect to give us knowledge of the future or of things we have not experienced,cannot really give us such knowledge. This is a disturbing conclusion,to find that our most plausible inductive arguments do not give us knowledge.Even if there is a solution,there is the further problem of
Hume’s point can be put another way. How could we improve our inductive argument so that its justification would not be in question? One way to try would be to add an additional premise that turns the argument into a deductive one. For instance, along with the premise “all observed cases of colitis have been accompanied by anaemia” we may add the premise “unobserved cases resemble observed cases” (let us call this the uniformity premise). From the two premises we may deduce that unobserved colitis sufferers are anaemic too. We are now justified in believing our conclusion, but only if we are justified in believing the uniformity premise, that unobserved cases resemble observed ones. The uniformity premise is itself clearly a generalization. There is evidence for it—cases where something has lain unobserved and has at length been found to resemble things already observed. The lump of ice I took out of the freezer an hour ago had never hitherto been observed to melt at room temperature (16°C in this case), but it did melt, just as all previous ice cubes left at such a temperature have been observed to do. Experience tells in favour of the uniformity premise. But such experience can only justify the uniformity premise if we can expect cases we have not experienced to resemble the ones we have experienced. That expectation is just the same as the uniformity premise. So we cannot argue for the uniformity premise without assuming the uniformity premise. But without the uniformity premise—or something like it—we cannot justify the conclusions of our inductive arguments. Induction cannot be made to work without also being made to be circular. (Some readers may wonder whether the uniformity premise is in any case true. While our expectations are often fulfilled they are frequently thwarted too. This is a point to which we shall return in Chapter 5.) If the premises of an inductive argument do not entail their conclusions, they can give them a high degree of probability. Might not Hume’s problem be solved by probabilistic reasoning? I shall examine this claim in detail in Chapter 6. But we can already see why this cannot be a solution. Consider the argument “all observed colitis suffers are anaemic, therefore it is likely that all colitis sufferers are anaemic”. This argument is not deductively valid, so we cannot justify it in the way we justify deductive arguments. We could justify it by saying that arguments of this kind have been successful in many cases and are likely to be successful in this case too. But to use this argument is to use an argument of the form we are trying to justify. We could also run the argument with a modified uniformity premise: a high proportion of unobserved cases resemble observed cases. But again, we could only justify the modified uniformity premise with an inductive argument. Either way the justification is circular. Goodman’s problem There is another problem associated with induction, and this one is much more recent in origin. It was discovered by the American philosopher Nelson Goodman. Hume’s problem concerns the question: Can an inductive argument give us knowledge (or justification)? If the argument employed is right, it would seem that those arguments we would normally rely upon and expect to give us knowledge of the future or of things we have not experienced, cannot really give us such knowledge. This is a disturbing conclusion, to find that our most plausible inductive arguments do not give us knowledge. Even if there is a solution, there is the further problem of Introduction: the nature of science 11
Philosophy of science 12 identifying those arguments that are even plausible as inductive arguments.A deductive argument is one the premises of which entail its conclusion.I have said little about inductive arguments in general,except that they do not entail their conclusions.However, I did ascribe some structure to what I called Humean inductions.One example was "all observed patients with colitis have anaemia,therefore all colitis sufferers have anaemia". Another example beloved of philosophers is "all observed emeralds are green,therefore all emeralds are green".In these cases we are generalizing from what we have observed of a certain kind of thing,to all of that kind,both observed and unobserved,past,present, and future.We can express this in a pattern or general format for Humean induction: all observed Fs are Gs; therefore all Fs are Gs. Goodman's problem,the "new riddle of induction"as it is also known,suggests that this format is too liberal.It allows in too many arguments that we would regard as wildly implausible.Worse,it seems to show that for every inductive argument we regard as plausible there are many implausible arguments based on precisely the same evidence. Let us see how this works. Goodman introduces a new term grue.Grue is defined thus: X is grue =either X is green and observed before midnight on 31 December 2000 or X is blue and not observed before midnight on 31 December 2000 Such an unintuitive and novel concept is perhaps not easy to grasp,so let us look at a few cases.A green gem that has already been dug up and examined is grue.A green gem that remains hidden under ground until 2001 is not grue,but a hidden blue gem first examined in 2001 is grue.Were the blue gem dug up in 1999 it would not be grue. All observed emeralds are both green and grue.To see that they are all grue,consider that since they are all observed (and I am writing in 1997)they are observed before midnight on 31 December 2000.They are also all green.So they satisfy the definition of grue (by satisfying the first of the two alternative clauses). Returning to the general format for an inductive argument,we should have an inductive argument of the form: all observed emeralds are grue; therefore all emeralds are grue. The conclusion is that all emeralds are grue.In particular,emeralds that are first dug up and examined in 2001 are grue.From the definition of grue,this means they will be blue. So the format of our inductive argument allows us to conclude that emeralds first examined in 2001 are blue.And it also allows us to conclude,as we already have,that these emeralds will be green. Predicates such as "grue"are called bent predicates,as their meanings seem to involve a twist or change;they are also called non-projectible,because our intuitions are that we cannot use them for successful prediction.Bent predicates enable us to create,for every
identifying those arguments that are even plausible as inductive arguments. A deductive argument is one the premises of which entail its conclusion. I have said little about inductive arguments in general, except that they do not entail their conclusions. However, I did ascribe some structure to what I called Humean inductions. One example was “all observed patients with colitis have anaemia, therefore all colitis sufferers have anaemia”. Another example beloved of philosophers is “all observed emeralds are green, therefore all emeralds are green”. In these cases we are generalizing from what we have observed of a certain kind of thing, to all of that kind, both observed and unobserved, past, present, and future. We can express this in a pattern or general format for Humean induction: all observed Fs are Gs; therefore all Fs are Gs. Goodman’s problem, the “new riddle of induction” as it is also known, suggests that this format is too liberal. It allows in too many arguments that we would regard as wildly implausible. Worse, it seems to show that for every inductive argument we regard as plausible there are many implausible arguments based on precisely the same evidence. Let us see how this works. Goodman introduces a new term grue. Grue is defined thus: X is grue = either X is green and observed before midnight on 31 December 2000 or X is blue and not observed before midnight on 31 December 2000 Such an unintuitive and novel concept is perhaps not easy to grasp, so let us look at a few cases. A green gem that has already been dug up and examined is grue. A green gem that remains hidden under ground until 2001 is not grue, but a hidden blue gem first examined in 2001 is grue. Were the blue gem dug up in 1999 it would not be grue. All observed emeralds are both green and grue. To see that they are all grue, consider that since they are all observed (and I am writing in 1997) they are observed before midnight on 31 December 2000. They are also all green. So they satisfy the definition of grue (by satisfying the first of the two alternative clauses). Returning to the general format for an inductive argument, we should have an inductive argument of the form: all observed emeralds are grue; therefore all emeralds are grue. The conclusion is that all emeralds are grue. In particular, emeralds that are first dug up and examined in 2001 are grue. From the definition of grue, this means they will be blue. So the format of our inductive argument allows us to conclude that emeralds first examined in 2001 are blue. And it also allows us to conclude, as we already have, that these emeralds will be green. Predicates such as “grue” are called bent predicates, as their meanings seem to involve a twist or change; they are also called non-projectible, because our intuitions are that we cannot use them for successful prediction. Bent predicates enable us to create, for every Philosophy of science 12
Introduction:the nature of science 13 credible inductive argument,as many outrageous inductions as we like.So not only can we argue that all emeralds are green and argue that they are grue,we can also argue that they are gred and that they are grellow.A standard induction would suggest that water at atmospheric pressure boils at 100C.But we can form a Goodmanesque induction the conclusion of which is that water breezes at 100C (where the definition of "breezes" implies that water observed for the first time in the next millennium will freeze at 100C).In addition there are arguments that children born in the next century will be immortal and that the Earth's rotation will reverse so that the Sun will rise in the west. The problem with this is not merely that these conclusions seem thoroughly implausible.Rather,it looks as if any claim about the future can be made to be a conclusion of an inductive argument from any premises about the past,as long as we use a strange enough grue-like predicate.And,if anything can be made out to be the conclusion of an inductive argument,then the very notion of inductive argument becomes trivial.So how do we characterize the reasoning processes of science if their central notion is trivial? Philosophers have proposed a wide range of answers and responses to Goodman's problem,and I will myself make a suggestion in Chapter 3.But one response that will not do is to dismiss the problem as contrived.First,this begs the question:Why is it contrived?What distinguishes pseudo-inductive arguments with contrived concepts like "grue"from genuine inductions using "green"?Secondly,we can think of cases that are less contrived.Take someone who knows about trees only that some are evergreens and some are deciduous.If they were to observe beech trees every day over one summer they would have evidence that seems to support both the hypothesis that beech trees are deciduous and the hypothesis that they are evergreens.Here neither option is contrived. But we know this only because we have experienced both.(The idea of deciduousness might appear contrived to someone who had lived all their lives in pine forests.) Goodman's problem appears to be at least as fundamental as Hume's,in the following sense.Even if we had a solution to Hume's problem,Goodman's problem would appear to remain.Hume argues that no conclusion to an inductive argument can ever amount to knowledge.Say we thought that there were an error in Hume's reasoning.and that knowledge from inductive arguments is not ruled out.It seems we would still be at a loss to say which arguments these are,as nothing said so far could distinguish between a genuine inductive one and a spurious one involving bent predicates. Representation and reason When discussing Judge Overton's opinion regarding creationism,I stressed that he was not trying to judge whether creationism,or evolution,or something else is true,let alone whether anyone could rightly claim to know which of these is true.Rather,he was,in part,trying to judge what it is that makes a claim to be scientific.Roughly speaking,there were two sorts of relevant answer.The first dealt with the subject matter of science:laws of nature,natural explanation,natural kinds (if any).The second concerned the sorts of attitude and approach a scientist has towards a scientific theory:proportioning belief to the strength of evidence,avoiding dogma,being open to the falsifiability of a theory
credible inductive argument, as many outrageous inductions as we like. So not only can we argue that all emeralds are green and argue that they are grue, we can also argue that they are gred and that they are grellow. A standard induction would suggest that water at atmospheric pressure boils at 100°C. But we can form a Goodmanesque induction the conclusion of which is that water breezes at 100°C (where the definition of “breezes” implies that water observed for the first time in the next millennium will freeze at 100°C). In addition there are arguments that children born in the next century will be immortal and that the Earth’s rotation will reverse so that the Sun will rise in the west. The problem with this is not merely that these conclusions seem thoroughly implausible. Rather, it looks as if any claim about the future can be made to be a conclusion of an inductive argument from any premises about the past, as long as we use a strange enough grue-like predicate. And, if anything can be made out to be the conclusion of an inductive argument, then the very notion of inductive argument becomes trivial. So how do we characterize the reasoning processes of science if their central notion is trivial? Philosophers have proposed a wide range of answers and responses to Goodman’s problem, and I will myself make a suggestion in Chapter 3. But one response that will not do is to dismiss the problem as contrived. First, this begs the question: Why is it contrived? What distinguishes pseudo-inductive arguments with contrived concepts like “grue” from genuine inductions using “green”? Secondly, we can think of cases that are less contrived. Take someone who knows about trees only that some are evergreens and some are deciduous. If they were to observe beech trees every day over one summer they would have evidence that seems to support both the hypothesis that beech trees are deciduous and the hypothesis that they are evergreens. Here neither option is contrived. But we know this only because we have experienced both. (The idea of deciduousness might appear contrived to someone who had lived all their lives in pine forests.) Goodman’s problem appears to be at least as fundamental as Hume’s, in the following sense. Even if we had a solution to Hume’s problem, Goodman’s problem would appear to remain. Hume argues that no conclusion to an inductive argument can ever amount to knowledge. Say we thought that there were an error in Hume’s reasoning, and that knowledge from inductive arguments is not ruled out. It seems we would still be at a loss to say which arguments these are, as nothing said so far could distinguish between a genuine inductive one and a spurious one involving bent predicates. Representation and reason When discussing Judge Overton’s opinion regarding creationism, I stressed that he was not trying to judge whether creationism, or evolution, or something else is true, let alone whether anyone could rightly claim to know which of these is true. Rather, he was, in part, trying to judge what it is that makes a claim to be scientific. Roughly speaking, there were two sorts of relevant answer. The first dealt with the subject matter of science: laws of nature, natural explanation, natural kinds (if any). The second concerned the sorts of attitude and approach a scientist has towards a scientific theory: proportioning belief to the strength of evidence, avoiding dogma, being open to the falsifiability of a theory. Introduction: the nature of science 13