Conventionalism in |
Brian Ellis
La Trobe University, Bundoora, Australia
The conventionalist program of analysis and rational reconstruction of science was seriously undermined by the movements toward epistemological holism and scientific realism in the 1960s, and very little work is now being done in this tradition. The holists argued that the distinctions required for the program could not be made satisfactorily; the realists claimed that the accepted laws and theories of science should be interpreted, more or less as they are, as literally true statements about the world and, hence, that rational reconstruction is unnecessary. However, I am convinced that something important was lost in this process, and I want to argue for a new program, which preserves what is good in the old one but does not depend on the usual conventionalist distinctions and is compatible with a sophisticated form of scientific realism.
Let me indulge in some personal history. When I wrote Basic concepts of measurement (1966) (hereafter BCM), I was still very much influenced by the conventionalist writings of some of the early positivists, particularly Mach, Poincare, and Reichenbach. I found their works on classical mechanics, geometry, heat theory, space and time, and many other subjects exciting. They brought out various more or less hidden assumptions of the theories they investigated and classified them as empirical facts or conventions. Like the early positivists, I thought this was an important thing to do, because it indicated where we had degrees of freedom, if they should be needed, in reconstructing or improving our theories and where they were fixed by the empirical data. However, I thought it was always important to know just what would be involved in changing any of these conventions, for, like Poincaré, I did not believe they were arbitrary. Therefore, as I saw it, it was an integral part of the conventionalist program to discover what reasons, if any, there may be for adopting some conventions rather than others. I thought this was one of the more valuable things a philosopher of science could do.
Like many other philosophers at the time, I accepted a version of the analytic/synthetic distinction. Analytic propositions, like “A bachelor is an unmarried man,” I should have said are true simply in virtue of the meanings of words. As such, they are conventional but trivial and uninteresting. Of much greater interest, I thought, were the hidden conventions of scientific theories and practices, which could not be discovered just by linguistic analysis. These conventions, I supposed, reflected decisions, often unconscious, to accept certain propositions, although they were neither semantically nor empirically warranted, because it was useful (for the theory or practice) to do so. These propositions were not analytic, as I understood analyticity, because they were not true simply in virtue of the meanings of words. Nor were they straightforwardly empirical, because they could not be inductively supported and could be refuted (or made no longer useful to accept) only by replacing the theory or changing the practice to which they belonged.
The analytic/synthetic distinction was strongly criticised by W. V. O. Quine and others in the 1950s, but philosophers of science were slow to see this criticism as casting doubt on their own distinction between empirical and conventional propositions. Adolf GrÜnbaum (1968, p. 4) for example, argued that “… whatever the merits of the repudiation of the analytic-synthetic dichotomy and the antithesis between theoretical and observation terms, it is grievously incorrect and obfuscating to deny as well the distinction between factual and conventional ingredients of sophisticated space-time theories in physics.” He thought, as most of us did, that the analytic/synthetic distinction was only concerned with the meanings of words and could safely be abandoned. The conventions of interest to philosophers of science were not regarded as analytic propositions. They were not just trivial semantic conventions but conventions that had a substantial role in structuring our theories. Nevertheless the epistemological holism advocated by Quine gained ground in the 1960s through the writings of I. Lakatos, T. S. Kuhn, and P. K. Feyerabend who attacked the foundationalist epistemologies of the empiricists, and in doing so, they seriously undermined the empirical/conventional distinction that the conventionalists required for their program.
At the same time, J. J. C. Smart, D. M. Armstrong, and others were developing strong arguments for scientific realism, which was anticonventionalist for a different reason. The holists did not think the distinction could be made satisfactorily; the scientific realists wanted to interpret the laws and theories of science as literally true generalised descriptions of reality. For example, they wanted a much more literal interpretation of statements of the results of measurements and, hence, of quantitative laws, than I was prepared to give. Thus, if one quantity is proportional to another or to the square root of another, in certain circumstances, then their view was that these relationships must hold between the categorical bases of the quantities concerned, independently of how we might measure these quantities. Consequently, if we measure the quantities correctly in these circumstances, our measurements should reflect the underlying relationships of proportionality. If they do not, then we are not using the correct scales. On the view I took in BCM, there are no true or correct scales for the measurement of quantities—only more or less useful ones, resulting in more or less simple mathematical expressions for the laws of nature.
Since that time, I have also come to think there is no very useful distinction to be drawn between what is conventional and what is not.1 At the extremes, there are trivial semantic conventions, and there are empirical facts, and the distinction between them is easy enough to make. However, most of the interesting propositions of science are not clearly either facts or conventions. If we say these propositions are true by definition, and so conventional, then we are forced to say that other propositions, which are just as plausibly conventional, are empirical or factual.2 So the distinction often seems to be forced and arbitrary. Moreover, the distinction, when made in this way, has no epistemological force. A proposition, arbitrarily declared to be true by definition, is as much open to revision in the light of experience as any proposition that must consequentially be declared to be empirical or factual.
I am also much more of a scientific realist than I was when I wrote BCM. I now hold that a satisfactory ontology for science must include a categorical basis for the quantitative relationships we are able to observe. However, the question of what sort of categorical basis for quantities we need to assume is an important one for scientific realists to face up to, and very few have done so.3 In BCM, I identified quantities with objective linear orders, saying that a quantity exists, if and only if such an order exists. Thus, I claimed that quantities are essentially relations and denied that they are properties (which come in degrees or have magnitudes) that an object considered in isolation might possess. Our knowledge of quantities, I thought, was basically just knowledge of quantitative relationships, and I supposed that these relationships, or some of them at least, had primary ontological status (i.e., were not reducible to or dependent upon any intrinsic properties of the objects they related). Some such account as this seemed to be required anyway for distance and time-interval. I held that a similar account was required for other quantities.
Whereas I still have some sympathy with my earlier position on this, because it provides a nicely unified theory of quantities, I no longer think that the account, which is appropriate for distance and time-interval, can be generalised in this way. I now think we need an ontology that includes both primitive quantitative properties (upon which some quantitative relationships depend) and some primitive quantitative relationships, such as the spatiotemporal ones.
Between them, these two movements—epistemological holism and scientific realism—virtually demolished the conventionalist program of analysis. I think this is a pity. I do so, despite the fact that I agree with both the holists and the scientific realists on most of the important issues. I agree with the holists, for example, that it is useless to try to classify all of the propositions of science as empirical or conventional as the early positivists hoped to do, and I agree with Byerly and Lazara that my neo-operationist account of the meaning of law statements will not do. Nevertheless it remains of interest to know why we accept the propositions we do and what would be involved in rejecting any that we could reject. For these are just the sorts of case studies in the theory of knowledge that should be the foundation for any general epistemology of science.
In what follows, I shall:
1. Argue for a modified program of analysis and rational reconstruction of science that retains what is valuable in the original conventionalist program, without commitment to any particular epistemology or ontology.
2. Try to say what kind of theory of quantities a scientific realist should be committed to. I shall focus particularly on the theory of dimensions and the expression of quantitative laws, for this is where the conventionalist theory of measurement has been most fruitful, and the new program of analysis and rational reconstruction I wish to advocate can be best illustrated. It also raises all of the important questions concerning the nature of quantities I wish to discuss.
THE CONVENTIONALIST PROGRAM
The aims of the conventionalist program of analysis were as follows:
1. To provide a clear theoretical framework for discussion of the area to be studied by explicating the concepts we normally work with, introducing new concepts and distinctions where necessary.
2. To discover and distinguish between the empirical presuppositions and unexamined conventions that underlie our theories and practices in the area.
3. To find out the extent to which the choices we have effectively made in adopting these conventions depend on or are constrained by empirical facts, theories, considerations of convenience, formal simplicity, and so on, and to what extent their adoption appears to be arbitrary.
4. To reconstruct rationally the theory or practice that is the subject of this analysis.
I am convinced that these aims, or something like them, are still worth pursuing, although the distinctions they require seem muddier now than they once did.
UNITS, DIMENSIONS, AND SCALE SYSTEMS
The value of conventionalist analysis may be illustrated by the theory of units and dimensions developed in BCM. For it explained clearly, as had not, I think, been done before, the scope and limitations of dimensional analysis, and how and why it was successful to the extent it was. It exposed the nonsense that people used to go with about dimensionless quantities, and it showed how, by changing our conventions concerning the expression of quantitative laws, the power of dimensional analysis can be increased. The theory, therefore, has demonstrated its practical utility.
The intuitive idea of a unit of a quantity q is that it is the qness of a given, or well-defined, object that serves as a standard for the measurement of q, (on a scale that assigns the number 1 to this object). However, in BCM, I argued that the conception of a quantity that is implicit in this way of thinking about units is unacceptable. If a is greater in q than b, this is not because a has more qness than b, I should have said. For qness is not like a substance of which one may have more or less; rather, like position or velocity, it is purely relational, and if a has a certain degree of qness, this is only because it occupies a certain position in the order of q, which it does by virtue of the q-relationships it bears to other things (e.g., to b). For these (and other) reasons, I took the view that unit names should never be regarded as the names of any specific amounts or intensities of the quantities they measure (which the standards used might be supposed to possess intrinsically). Rather unit names should always be regarded as the names of scales.
On the account that was once standard and is still widely accepted, quantities have dimensions that are either simple or complex. A simple dimension is that of one of the basic physical quantities in terms of which all other physical quantities are ultimately definable. A complex dimension, or dimensional formula, is a product of powers of the simple or basic dimensions that reflects the way in which the quantity that has this dimension has been defined in terms of the basic quantities.4
However, I think this account of dimensions is quite unsatisfactory, as those who have tried to work with it have found. They did not know how to pick out the basic quantities, even in mechanics, and they were still more confused when it came to assigning dimensions to the quantities involved in heat theory and electrodynamics.5 They did not know how to deal with so-called dimensionless quantities, like angle, or mechanical advantage, which were alike in being dimensionless but obviously very unlike conceptually. Measures of such quantities were said to be “pure numbers.” Moreover, it was hard to explain, on the standard theory, why dimensional formulae should always be products of powers, or how and why dimensional analysis could be so useful. Finally the standard account obscured some important conventions concerning the construction of scale systems and the expression of quantitative laws.
According to the theory developed in BCM, dimensions should properly be identified with classes of similar scales (where two scales are said to be similar, iff measurements on these scales are uniformly related by similarity transformations, that is, transformations of the form y = mx, where m is a constant). The common dimension of length, for example, is the class of scales similar to the metre scale. It includes the foot and fathom scales, but it does not include the various diagonal scales described in the chapter of BCM on fundamental measurement or any other members of the power group that do not belong to the similarity class. The ordinary dimension of force is the class of scales similar to the newton scale. It includes the dyne and pound weight scales, but it does not include any scales that are not related to these by similarity transformations.
The so-called basic quantities are just those measured on independent scales that are used to the define various systems of scales of measurement (e.g., the MKS and FPS systems). The systems so defined may be said to be centered on these scales. However, apart from the fact that these quantities normally have a fairly central role in physical theory, there is nothing very special about them. We could use more or less highly centered scale systems, if we wished, or choose to center them on scales for the measurement of different quantities. A scale system is a set of interrelated dependent and independent scales. The dependent scales are either: (a) scales that we have made dependent on other scales by definition (e.g., our fundamental scales for the measurement of area and volume), or (b) derivative scales for the measurement of various quantities defined by the values of system-dependent constants in quantitative laws. A system-dependent constant is one that is characteristic of a system, or a system at a time, but that may vary from system to system or with time. Elasticity, density, and refractivity are quantities typically measured on derivative scales. In general, the complex unit names given to dependent scales simply indicate how they depend on the independent scales of the system. The fact that we center our scale systems as we do on some independent scales for the measurement of a few basic quantities reflects an important convention in the practice of measurement. It is also conventional that we center them on scales for the measurement of these particular quantities.
A quantitative law relates the idealised results of measurements of different quantities on various scales. Now there are many ways in which such a law might be expressed, depending on the classes of scales for which we wish it to be valid. For example, we could express any quantitative law so that it is valid for scales linearly related to the usual scales for the measurement of some or all of the quantities involved. The gas law, for example, could be expressed as: pV = R(aT + b), where a and b are scale dependent constants. (For the Celsius Scale, a = 1 and b = 273. For the Fahrenheit Scale, a = 5/9 and b = 255.)
In practice, however, we use a much more conservative form of expression. We always express quantitative laws either: (a) with respect to particular scales for the measurement of some or all of the quantities involved, so that the law is valid for just these scales; or (b) with respect to classes of similar scales for the measurement of some or all of the quantities involved, so that the law is valid for all scales in these classes. Thus, the laws of mechanics are normally expressed with respect to classes of similar scales for the measurement of all of the quantities involved and with reference to a scale system that is centered on independent scales for mass, length, and time-interval.
On the other hand, laws involving the so-called “dimensionless” quantities, like angle, and quantities measured independently on associative scales, like temperature, are always expressed with respect to particular scales (e.g., the radian or the Absolute scale). The so-called dimensionless quantities are not, however, dimensionless; they are just quantities measured on scales that are defined in ways which make them independent of the scale system. However, there is no good reason why we should not, and very good reason why we should, express laws involving angle with respect to the class of scales similar to the radian scale R and adopt the MKSR, or a similar scale system, for mechanical measurement. For the power of dimensional analysis would demonstrably be increased, if we were to adopt such a system.6 Likewise, the convention that we should always express our temperature laws with respect to a particular temperature scale has no rational basis. On the contrary, there are very good reasons for expressing laws involving temperature with respect to the class of scales similar to the Absolute scale.
The theory of dimensions developed in BCM was surely an advance on previous theories of dimensions. Yet the theory was arrived at by following the original conventionalist program of analysis and rational reconstruction as it applied to the practice of expressing quantitative laws. Therefore, whatever the merits of the theoretical underpinning of this program, there must be something to be said in its favor. It would be regrettable, if the arguments against conventionalism should prevent this sort of thing being done in future.
HOLISTIC OBJECTIONS TO CONVENTIONALISM
Holism in epistemology is the view that our knowledge consists of a theoretically and conceptually integrated system of items of knowledge. These items are assumed to have no identity independent of the system to which they belong, because of the pervasive influence of theoretical interpretations on both language and belief. In addition, epistemological holists typically hold the following theses, or variants of them:
1. There are no basic items of knowledge or belief that could serve as a foundation for knowledge.
2. There is no theory-neutral observation language; all observation reports are theory-laden.
3. An observationally acquired belief that is incompatible with our theoretical expectations may cause us to modify our belief system, but any belief can be defended, if we are prepared to make drastic enough revisions elsewhere in the system.
4. Normally many different theories are involved in testing a given hypothesis, and if the hypothesis is refuted, it is only because we prefer to abandon it rather than any of these other theories.
Epistemological holism is not strictly incompatible with an empirical/ conventional distinction. Indeed the Duhemian thesis, (4), specifically allows that some hypotheses may be maintained indefinitely as conventions, whatever the empirical evidence may be. However, it does suggest that the body of science may contain a range of more or less provisionally accepted propositions—some that might be given up quite readily and others (like the propositions of arithmetic perhaps) that we should probably retain in any circumstances. It also suggests that propositions accepted as conventions may be more or less arbitrary, depending on how they relate to other beliefs in the system. A convention that has been adopted for good theoretical reasons is not an arbitrary convention. Nor is it arbitrary, if it could not be abandoned without seriously complicating our theories. Conventions may, thus, be constrained by theoretical considerations. However, our theories in turn are constrained by empirical evidence. Therefore, the acceptability or otherwise of a convention may ultimately depend on such evidence. We should not, therefore, expect to be able to draw a clear distinction between what is empirical and what is conventional. On the contrary, epistemological holism suggests that the propositions of science should be interrelated in a wide variety of ways and that it is, or should be, one of the tasks of epistemology to map the interconnections between them.
REALIST OBJECTIONS TO CONVENTIONALISM
According to the dominant strain of scientific realism, it is most rational to believe that the world is literally more or less how our best scientific theories say it is. However, this runs counter to the main tradition of conventionalism, which regards the ontological claims of current science with a great deal of scepticism. Conventionalists would argue that what is needed is a critical approach to the claims of science—one that demands empirical justifications for these claims— if science is to be rid of the relics of past metaphysics. Thus, realists tend to regard conventionalists as sceptics, and conventionalists typically think of realists as naive. Both attitudes are well justified. What is needed for the new program I wish to advocate is a sophisticated form of scientific realism—one that takes the ontological claims of science seriously but is not above questioning the ontological status of entities postulated in scientific theories.
The main argument for scientific realism is an argument from the best explanation. Scientific realists make the reasonable assumption that the best explanations we have of why things behave as they do are the accepted scientific ones. These explanations normally purport to refer to theoretical entities of some kinds. Therefore, they argue, it is at least the case that the world behaves as if things of these kinds existed. Yet the best explanation of why this should be so is that things like these really do exist. Therefore, it is most reasonable to believe in the existence of the sorts of theoretical entities postulated in the best scientific theories.
However, the main argument for scientific realism does not apply to such theoretical entities as points in space or space-time that are not postulated as having causal roles. For it cannot be said of these entities that the world behaves as if they existed. Moreover, it clearly does not apply to the theoretical entities occurring in constructive explanations of the kind discussed by Nancy Cartwright (1983). For such entities are idealised objects to which real things are at best only approximations, and there is no reason whatever to believe in them—again because they are not postulated as causes. As I see it, the main argument for scientific realism is an argument from known effects to postulated causes in the best causal process explanation. As such, it is a good argument, but it is not a good argument for the existence of theoretical entities occurring in other kinds of theories or explanations.
On the other hand, the main argument for scientific realism is prima facie a good argument for the existence of many kinds of entities besides particles, fields, and other things that have mass or energy. It seems, for example, to be a good argument for the existence of the properties of these things, the spatiotemporal relationships between them, the fundamental forces (conceived as basic kinds of causal interactions) to which they are subject and the events to which they give rise.7
For it is certainly true to say that the world behaves (is) as if these properties, relationships, forces, and events existed (occurred), and surely the best explanation for this is that they really do exist (occur).
Now the most striking thing about the fundamental properties of nature is that they are nearly all quantitative. That is, they come in degrees. Therefore, any adequate ontology for science must recognise the fundamental existence of quantitative universals (i.e., universals like mass, charge, spin, colour, flavour, strangeness, and so on) that may be variously instantiated. I do not know how to develop such a theory, but somehow we must try to explain what two things that differ in mass, for example, have in common, as well as what differentiates them, and the same sort of account will be needed for all of the other fundamental quantities.
In BCM, I construed all quantities on the model of spatiotemporal relationships, which I did not think and still do not think could be explained as relationships between any intrinsic properties of the objects or events they relate. Thus, I would have denied that the charge on an electron is an intrinsic property of the particle, just as I would have denied that its position is intrinsic to it. Rather I would have drawn the explicit analogy with position in space and said that the electron has the charge it has in virtue of its electromagnetic relationships to other things (cf. the electron has the position it has in virtue of how it is related spatially to other things).
I no longer think that this account is tenable. For the best explanation of the fixedness and stability of the properties of the fundamental particles is just that these properties are intrinsic to them. The fact that all electrons have the same charge, mass, spin, and so on is best explained by supposing that these properties are all intrinsic to electrons. It is not well explained by the hypothesis that they all happen to bear the same electromagnetic, dynamic, and so on relationships to other things. For why should they do so, if these properties are not intrinsically but extrinsically determined? The case is different with spatiotemporal relationships. There is no such stability in the positions of things that would be explained by the supposition that their positions are intrinsic. The main argument for scientific realism, thus, comes out decisively in favor of the existence of some intrinsic quantitative properties.
I do not know of any satisfactory theory of quantitative properties and relationships. My own theory that quantitative properties always supervene on quantitative relationships, in the sort of way that spatiotemporal position supervenes on spatiotemporal relationships, will not do for the reasons given. On the other hand, D. M. Armstrong’s theory, that quantitative relationships always supervene on quantitative properties, does not account adequately for the continuum of spatiotemporal relationships. For we should at least need to have a non-denumerable infinity of points, each with their own, intrinsic locations in space-time, to ground such a continuum of relationships.
What is needed is a general, ontologically sound theory of quantitative properties and relationships that we can use to say what sorts of quantities exist in nature, what their properties are, and how they may be related to each other in lawlike or accidental ways. However, even when we have such a theory, there is still the problem of what degrees of freedom we have in metrising these relationships and what grounds there may be for choosing one metrical system rather than another. To answer these questions, we will have to engage in a new program of analysis and rational reconstruction, not so very different from the kind undertaken in BCM.
THE NEW PROGRAM OF ANALYSIS AND RATIONAL RECONSTRUCTION
The following issues are important ones that are independent of the global aims of conventionalist analysis:
1. There are many presuppositions of scientific theories and practices, which have no empirical justification, and so are conventional in the traditional sense, and it is not without interest to find out what they are, why they should be accepted (if indeed they should be), what other conventions might replace them, and at what cost or benefit.
2. There are many empirical presuppositions of science that have never in fact been tested, either because they have not been recognised as presuppositions or have been taken for granted, or because the technology does not exist to test them. These too should be stated, tested if possible, and the possibilities of other assumptions should be explored.
3. There are many propositions that are arguably either empirical or conventional, depending on your theory of empirical justification. For these, it probably does not matter much how you classify them. If you say they are conventions, then they will not be considered to be arbitrary conventions but ones adopted for good pragmatic reasons. If you say they are empirical, then it will be because you think that these very same reasons are empirically (as opposed to pragmatically) justifying. In either case, we should seek to be clear what the reasons are and whether they are sufficient to justify accepting these propositions rather than others.
As an example of the first kind of case, consider the conventions governing the construction and centering of scale systems. First, it is conventional that we should choose to center our scale systems for mechanical measurement around scales for mass, length, and time-interval. I suppose the practice derives from the seventeenth century mechanistic world view, according to which the universe consists of matter in motion. Obviously this implies that mass, length, and time-interval are in some sense primary quantities (cf. also the seventeenth century distinction between primary and secondary qualities). However, this would not be a good rationale for centering our scale systems for mechanical measurement on just these quantities, even if the seventeenth century world view were still tenable. For what is important in making this choice is what information it enables us to include (about the forms of the most fundamental laws of mechanics) in the complex unit names for indirectly measured quantities and constants. Systems centered only on scales of mass, length, and time-interval are demonstrably not optimal as systems for mechanical measurement.
Second, it is a matter of convention—one that has no rational justification— that we should express quantitative laws with respect to classes of similar scales for the measurement of some quantities and with respect to particular scales for the measurement of others. For example, mechanical laws are normally expressed dimensionally (i.e., with respect to classes of similar scales for all of the quantities involved), but laws involving angle, temperature, and electrical resistance are usually not expressed dimensionally, but with specific reference to the Radian, Absolute, or Ohm scales for these quantities. Rationally, however, these laws should also be expressed dimensionally, for then we should be able to use dimensional analysis more widely to discover the forms of derivative laws involving these quantities.
The most famous example of an empirical presupposition that was not recognised as such, and was consequently unquestioned by the scientific community, is probably that of the absoluteness of time. Before 1905, it had been almost universally assumed that times and time-intervals are the same for all frames of reference, so that time could be represented by a single variable. In his 1905 paper, Einstein abandoned this assumption, which, until then, had not been generally recognised even to be an assumption, and took the radical step of allowing that the times of events, and the time-intervals between them, might be different for different observers. Indeed he allowed that events that are simultaneous in one system might not be simultaneous in another. It may be doubted whether this was an empirical presupposition of classical physics; however, whatever its standing, it was unquestioned and, as it turned out, unwarranted.
The program of analysis and rational reconstruction I now wish to advocate would try to avoid disputes about the status of those propositions whose epistemic justification is in question, because such disputes detract from the program’s main purposes. The question whether there is a clear or useful distinction to be drawn between what is empirical and what is conventional in science is no doubt an important one. However, the process of analysis and reconstruction I propose may be carried out whatever view one takes on this, and the results obtained will not be affected. If a proposition is rejected, it does not matter, for our purposes, whether the reasons for rejecting it are judged to be empirical, pragmatic, or a complex of both kinds of reasons. What matters is whether they are good reasons.
The main purposes of the new program of analysis and rational reconstruction are as follow:
1. To explicate the basic concepts of the various theories and practices of science.
2. To specify adequate boundary conditions for these theories and practices. In the case of a theory, a clear statement of the aims of the theory (e.g., to construct a theory of such and such a kind, compatible with such and such facts to explain the following data) is required. In the case of a practice, such as measurement or taxonomy, we need to know what its purposes are and how it seeks to achieve them.
3. To determine what further assumptions or choices must minimally be made to justify our current theories and practices, given these explications and specifications.
4. To consider what reasons there may be for making these assumptions or choices rather than others.
5. To work out in detail the consequences of making other assumptions or choices, where the reasons for making the usual ones do not appear to be strong.
6. To advocate changes to our theories or practices, where it can be shown that other assumptions or choices are preferable to the usual ones.
The new program retains what was important in the old conventionalist one, but strips it of its verificationist and empiricist assumptions. It is not suggested that it is always possible to draw a clear distinction between empirical and conventional elements in scientific theories. The purposes of the analyses required have nothing to do with the empiricist objective of trying to specify the empirical contents of the laws and theories of science. The new program is indeed compatible with epistemological holism and, therefore, with the view that this empiricist objective cannot even in principle be achieved.
The program is also compatible with a sophisticated scientific realism that does not always take the ontological claims of science at face value. A naive scientific realism that does do so is untenable anyway and is not warranted by the main argument for realism about theoretical entities. For this argument has force, only if the theoretical entities concerned are among postulated causes of the events explained by the theory. Even then we should not automatically assume that the entities referred to exist. Newton, for example, distinguished absolute space and time from relative spaces and times, and referred his laws of motion to this absolute framework. There can be no doubt that his explanations of the phenomena of motion were the best then available. However, it was not irrational for philosophers then, or since, to challenge Newton’s ontology. If the ontological claims of science are justified, then let us see the arguments for them rather than appeal to authority.
REFERENCES
Brown, G. B. (1941). A new treatment of the theory of dimensions. Proc. Amer. Phys. Soc, 53, 418–431.
Byerly, H. C., & Lazara, V. A. (1973). Realist foundations of measurement. Phil, of Science, 40, 10–28.
Cartwright, N. (1983). How the laws of physics lie. Oxford: Oxford University Press.
Duncanson, W. E. (1941). The dimensions of physical quantities. Proc. Amer. Phys. Soc, 53, 432–448.
Ellis, B. (1966). Basic concepts of measurement. Cambridge: Cambridge University Press.
Ellis, B. (1971). On conventionality and simultaneity—a reply. Austral. J. Phil., 49, 177–203.
GrÜnbaum, A. (1968). Geometry and chronometry in philosophical perspective. Minneapolis: University of Minnesota Press.
Pettit, P., & Sylvan, R. (forthcoming).
_________________
1The clearest statement of my rejection of the distinction is to be found in Ellis (1971, pp. 177–203).
2I use these terms here and throughout this essay as though they were interchangeable, although I do not think they are. I do so, because the empiricists whose views I am trying to recapture made no distinction between facts and empirical facts.
3The most notable exceptions are Henry C. Byerly and Vincent A. Lazara in their excellent paper, “Realist Foundations of Measurement,” (1973), pp. 10–28. More recently, there have been excellent papers on the topic published by C. Swoyer, and by J. Bigelow and R. Pargetter.
4This is the kind of view taken by G. Burniston Brown, for example, in “A new treatment of the theory of dimensions,” (1941) pp. 418–431, and by W. E. Duncanson, “The dimensions of physical quantities,” (1941), pp. 432–448.
5For example, Bumiston Brown believed that two basic dimensions (of length and time) were enough for the whole of physical theory, but Duncanson thought we needed a dimension of charge as well. However, neither system is useful. The reductions in the number of basic dimensions required are achieved at the cost of the information contained in the dimensional formulae for other quantities.
6See BCM, pp. 145–151 for a demonstration of this.
7This range of possibilities is discussed fully in a paper being written concurrently with this entitled “The ontology of scientific realism.” The paper is to appear in a volume of essays edited by Philip Pettit and Richard Sylvan, and dedicated to J. J. C. Smart on the occasion of his retirement from the Chair of Philosophy at the Australian National University.