Russell, Experience, and the Roots of Science

Written by AC Grayling.


Empiricism is the family of theories which in one or another way locate the source or at very least the test of contingent knowledge in experience – specifically, in sensory experience. More circumstantially: it is the family of theories which variously require experiential grounds for concepts to have content or applicability, or for expressions in a given language to have sense. In these versions of a formulation, due allowance is made for the thought that the content of perceptual states, suitably construed, are to be considered the occasion or basis for certain kinds of fundamental judgments from which, together with other premises, our less fundamental judgments about the world (or things other than the content of those states of sensitivity themselves) can be inferred.

In a qualified sense of this broadly characterised position, Russell was an empiricist, and his epistemology remained in that qualified sense empiricist throughout its development. But he was also critical of certain forms of empiricism, and the focus of his own concerns were such that his aims in formulating epistemological views, and his evolving attempts to realise these aims in detail, are not straightforwardly traditional. The chief reason for this is that his overarching concern was the question of how science is related to subjective experience, beginning (in the work done in 1911-14) with attempts to show how the fundamental concepts of physics can be derived from experience, and ending (in 1948) by shifting attention to the question of the non-empirical features of knowledge-acquisition required for bridging the gap between experience and science.

In these aims for epistemology Russell was remarkably consistent throughout the period 1911- 48, which is to say, from the time he finished work on the first edition of PM until his last major philosophical book, HK. His concern was not the traditional epistemological one of showing that knowledge is justified by experience, where this task is typically specified by a response to sceptical arguments. Russell was thoroughly Lockean in his attitude to the theory of knowledge, in the sense that he did not think scepticism a serious option, and therefore did not waste time attempting to rebut it. Rather, he conceived epistemology's proper task as one of displaying how one gets from sense experience to science. For Russell this was an explanatory, not a justificatory, task.

In the cluster of texts addressing the question of the experience-science relation in the immediate post-PM period, Russell describes his aim as showing how physics is 'verified' by observation and experiment – by which he meant: having its predictions confirmed by these means. Given that all that can be directly observed are the data of sense, he saw the question as one of explaining the correlation of the contents of the physical world with the data of sensory experience by which they are alone verifiable . He did not put the point by saying that claims about the content of the physical world are verified (still less justified) by sensory experience; and this is neither an accidental nor a merely historically-conditioned trick of formulation. It is a feature of robust realism not to construe the point of epistemology as being the justification of knowledge-claims, but as being an explication of the relation between what the claims are about and the nature of experience. 'Justifying science by grounding it in experience' and 'showing how physics succeeds in being an empirical science, based on observation and experiment' are two different aims, and Russell's was the latter.

In PP, which gives the outlines of Russell's early view in popular form, the project begins by adopting the Cartesian air of a justificatory, scepticism-rebutting enterprise. The same is true of the discussion in IMT and Russell's replies in Schilpp. But that was because Russell saw the principal task of showing how experience and science relate as the obverse of the coin whose reverse is the more familiar form of discussion in which experience is invoked as the ground of knowledge. Because Russell assumed throughout that science is (or at least is on the way to discovering) the truth about the world (and his considered views consistently respected this assumption), he did not see epistemology's task to be the defence of science against doubt, but instead to be the demonstration of how finite human subjectivity acquires knowledge of the objective reality which science describes. In showing this, it also shows that the degree of certainty possible in contingent knowledge is less than absolute. In this sense, Russell was happy to concede something to scepticism without being much troubled by it; after all – so in effect he thought – what else is to be expected from contingent empirical knowledge.

In the earlier phases of his endeavour Russell saw the task of technical philosophy (philosophy conceived as logic; in fact, though, this aspect of Russell's endeavour is more accurately described as metaphysics) as principally being one of showing how the fundamental concepts of science (as he then took them to be) – space, time, causality and matter Ð can be constructed, and in his view this was a more important and more interesting matter than the epistemological question of how one relatively insignificant fragment of reality – humanity – manages more or less successfully to represent the rest of reality to itself. It is easy to overlook the fact that these two of Russell's tasks – the logical construction of the then-conceived fundamental scientific concepts, and the question of how finite subjective experience connects with scientific knowledge – are different, although of course they impinge upon one another at most points. But Russell's attention came rapidly to focus almost exclusively on the epistemological task, to which the larger part of his strictly philosophical writings after 1911 were addressed.

What changed over time in Russell's thought after 1911 was not his epistemological aim, but the strategies he successively adopted to try to achieve it. Perhaps because science itself dramatically altered the question of which concepts are fundamental to it (space and time had become space-time in Einstein's theories, and matter had vanished in the wake both of them and quantum theory), Russell ceased to look for a logical construction of these specific concepts. Indeed, he abandoned the logical constructivist programme long before the likes of Carnap and Goodman attempted them, and before Wisdom had shown that getting the world out of sense-data without residue is impossible.

The continuities and developments in Russell's relation-of-sense-to-science project are well displayed as the similarities and contrasts between his description of the project's aims, and of the methods to be employed in carrying it out, in the 1911-14 writings and HK in 1948. Commentators generally take at face value Russell's own claim, in MPD, that in AMi (1921) he abandoned not just the nomenclature of the sense-datum theory but what it was trying to achieve; and this is taken among other things to mark a more expressly 'neutral monist' turn as the metaphysical basis of his epistemological efforts until, in his very late work, another and final shift of perspective occurs, this time away from efforts to carry out the original project and towards the task of identifying the non-empirical supplements which, by that stage, he saw as the chief interest in discussing the bridge over the experience-science gap. But in fact it can be shown that despite the asseverations of MPD and the apparent elimination of the subject in AMi (courtesy of Russell's by then further developed conception of the 'neutral monist' stance), the underlying theme of specifying the connections between experience and science remained. Of course, from the period of AMi onward Russell changed the terms of the relation at issue dramatically; acquaintance vanished, and was replaced (to begin with) by 'noticing' (experiential salience) and successor conceptions. Acquaintance and the subject seemed to go so intimately together that their departure appeared jointly necessary; but it is no surprise to find the epistemic subject still in view in HK, having been merely in disguise in the interim.

The purpose in what follows is accordingly to illustrate, by way of an account of the development of Russell's project, the remarkable consistency of aim it displays. I do this by tracing the project's history, chiefly to establish an accurate characterisation of it, but also to provide a corrective to the impression that in epistemology Russell merely offered a sequence of ad hoc moves in response to a problem which has since been understood, but even then was already beginning to be recognised, as misconceived, viz. the endeavour to erect a justificatory theory of knowledge on the flawed Cartesian grounds of deriving certainty from the private data of experience. But to repeat: Russell's task was, interestingly and significantly, different from that; he did not see epistemology as a justificatory enterprise aimed at refuting scepticism, but as a descriptive enterprise aimed at explaining the fact (which he did not question) that finite subjects attain scientific knowledge. He was thus a naturalist long before Quine or anyone else, despite rightly insisting, as later naturalists did not, that one cannot premise science in epistemology ; and he was far more consistent in his aims and principles than most (agreeing with Charles Broad ) have allowed.

Certain corollaries attend the picture I offer. One is that Hylton misdescribes Russell's turn to epistemological themes after PM as involving 'considerable concessions to psychologism'. Whatever else the label means, 'psychologism' is at least the view that the objects of acquaintance and judgment (to use period Russellian terms for the purpose) cannot themselves be described independently of features attaching to them as a result of the psychological conditions of their apprehension. This is never Russell's claim, and indeed anything like it was expressly disavowed in his pre-PoM flight from idealism. Post-PM Russell was realist to excess, rather than psychologistic, in allowing a wider range of objective targets of acquaintance than a traditional empiricist would allow, embracing as it did both physical particulars and abstract entities of various kinds. So much is familiar. And this is not to deny that Russell's interests lay in connecting the content of psychological states (mental states of the subject-relatum in acquaintance and judgment) with the independent objects such states brought into the subject's ken; for, after all, it was the 'transition from sense to science' as he still called it at the end of his philosophical life (MPD 153) that was his focus, and this requires addressing the question of what and how much the psychological states of epistemic subjects can be said to give them of objective scientific truth.

A corollary of the consistency thesis which I here argue on Russell's behalf is that the celebrated derailment of Russell's project in TK, ascribed to Wittgenstein as a result of some (characteristically hyperbolic) remarks by Russell in a letter to Lady Ottoline Morrell, might not be quite what it seems; for in a footnote added to the text of 'On Knowledge by Acquaintance and Knowledge by Description' when this 1911 essay was reprinted in ML in 1917, Russell remarks of his multiple relation theory of judgment, 'I have been persuaded by Mr Wittgenstein that this theory is somewhat unduly simple, but the modification which I believe it to require does not affect [its fundamentals].' The same point occurs more fully in LLA where Russell discusses the difficulties faced by the theory, involving subordinate 'verbs'. He subsequently, somewhat without fanfare, abandoned the theory; but it is clear from the fact that he continued to the end with the larger project of clarifying the experience-science connection that he found his multiple relation theory of judgment to be inessential to it; and therefore the fact that Russell dismembered TK and left some parts of it unused is not the same as his abandoning the project in whose working out TK was a chapter.


A good way to begin is to observe the images Russell employs early and late in preparing readers for the epistemological task as he conceived it. In the Preface to HK he observes that the terms 'belief', 'truth', 'knowledge' and 'perception' all have imprecise common uses which will require progressive clarification as the enquiry proceeds. 'Our increase of knowledge, assuming that we are successful, is like that of a traveller approaching a mountain through a haze: at first only certain large features are discernible, and even they have indistinct boundaries, but gradually more detail becomes visible and edges become sharper.' Compare this to what Russell says in TK of the ambiguities of the words 'experience', 'mind', 'knowledge' and 'perception': 'The meanings of common words are vague, fluctuating and ambiguous, like the shadow thrown by a flickering street-lamp on a windy night; yet in the nucleus of this uncertain patch of meaning, we may find some precise concept for which philosophy requires a name' - which, Russell concludes, should best be the common expressions themselves, made suitably definite. Imagery aside, part of the method of both early and late epistemology is thus characterised as the same: clarification of concepts, on one familiar view the central task of analysis characteristic of 'analytic philosophy'. But Russell also took the view that analysis is only the propaedeutical part of the story; more important (so he early believed and hoped) was the constructive task of showing how complexes of various kinds – and not least, knowledge of complexes – can be constructed out of simples – early on, the simples with which we are acquainted. The constructive task is the one which ended in failure, and the changes in Russell's epistemology are a direct function of the difficulties met with in the course of the project, which he increasingly saw as insurmountable. The hope had been to couple analysis and synthesis, the first activity preparing the way for the second, reflecting Russell's early ambition, formed on a walk one day in Berlin in the 1890s, to link abstract and scientific knowledge into a grand synthesis.

The synthetic task failed, but one thing which did not change was the aim subserved by the method developed to carry it out. In TK Russell plunges straight into the task of analysing acquaintance, which he calls 'the simplest and most pervading aspect of experience', a dyadic relation (an important point, for cognate polyadic relations of higher order constitute something significantly different, namely, judgments) between a 'mental subject' and what turn out to be the catholically-conceived objects of its attitudes. This was to fulfil a promise implicit in the outline of a programme given in March 1911 in three lectures: the Aristotelian Society address 'Knowledge by Acquaintance and Knowledge by Description', and two lectures delivered in Paris, 'Le Realisme Analytique' and 'L'importance philosophique de la logistique'. In the first of these latter he reasserts his commitment to realism both in epistemology and as regards universals, and outlines the technique of analysis of complex into simples to which he there first applies the name 'logical atomism'. In that and the companion lecture he launches the work characteristic of the 1911-14 period, worked out in most detail in a series of papers - 'On Matter' (1912), 'The Relation of Sense-Data to Physics' and 'On Scientific Method in Philosophy' (1914), and 'The Ultimate Constituents of Matter' (1915; the three latter are reprinted in ML) - whose chief precipitate constitutes OKEW (1914). Notoriously, the project was first planned to result in TK; but the difficulties over the theory of judgment obliged Russell to dismantle the task into what he doubtless hoped would be more manageable components.

The project is sketched in a letter from Russell to Ottoline Morrell in October 1912. 'The sort of thing that interests me now is this: some of our knowledge comes from sense, some comes otherwise; what comes otherwise is called "a priori". Most actual knowledge is a mixture of both. The analysis of a piece of actual knowledge into pure sense and pure a priori is often very difficult, but almost always very important.' Russell had chosen both parts of the task: to trace the transition from sense to science, and to isolate the a priori elements of the latter and to axiomatise them, as a preparation for defining the central concepts (space, time, causality and matter itself). Arguably, the epistemological task came to seem pressing to Russell for the two reasons that whereas, at the outset, the business of defining the fundamental concepts of physics appeared to be a straightforward parallel to defining the fundamental concepts of arithmetic, it quickly transpired that the relation of sense to science was not easy to carry out, and moreover that it was a necessary preliminary to completing the task of logically constructing the concepts of physics from whatever primitive concepts could be discovered in the then fundamental areas of physics, electrodynamics and classical mechanics, together with the relations between them. The reason for the latter is that the empirical content of the primitives requires that they themselves be constructible from sensory experience, as required by the principle that everything we know must be anchored at last in acquaintance.

Russell accordingly deferred the attempt to construct science's central concepts to deal with the epistemological questions first. It is instructive to see how these, in their own right, came to seem to him problematic, given that his first sketch of them (in PP) was an optimistic one, in that it canvassed the traditional questions about the relation of experience to knowledge with a robust acceptance of the fallibility of such knowledge, and the presence in it of assumptions or principles themselves neither independently testable nor matters of logic alone.


In PP Russell introduced the label 'sense-data' to designate what is immediately known in sensation: particular instances in perceptual awareness of colours, sounds, tastes, smells and textures, each class of data corresponding to one of the five sensory modalities. Not only must sense-data be distinguished from acts of sensing them, they must also be distinguished from objects in space outside us with which we suppose them associated. Russell's primary question therefore was: what is the relation of sense-data to these objects?

Russell was not, as noted, concerned to address scepticism. His tack was to say that although sceptical arguments are strictly speaking irrefutable, there is nevertheless 'not the slightest reason' to suppose them true (PP p 17). Instead he assembles persuasive considerations in support of the view that having sense-data provides access to reasonable knowledge of things in space. First, we can take it that our immediate sensory experiences have a 'primitive certainty'. We recognise that when we register sense-data which we naturally regard as associated with, say, a table, we have not said everything there is to be said about the table. We think, for example that the table continues to exist when we are not perceiving it, and that the same table is publicly available to more than one perceiver at a time. This makes it clear that a table is something over and above the sense-data that appear to any given subject of experience. But if there were no table existing independently of us in space we should have to formulate a complicated hypothesis about there being as many different seeming-tables as there are perceivers, and explain why nevertheless all the perceivers talk as if they were perceiving the same object.

But note that on the sceptical view, as Russell points out, we ought not even to think that there are other perceivers either, for if we cannot refute scepticism about objects, we are as badly placed to refute scepticism about other minds.

Russell short-circuits the difficulty by accepting a version of the argument to the best explanation. It is simpler and more powerful, he argues, to adopt the hypothesis that, first, there are physical objects existing independently of our sensory experience, and, secondly, that they cause our perceptions and therefore 'correspond' to them in a reliable way. Following Hume, Russell regards belief in this hypothesis as 'instinctive'.

To this, he argues, we can add another kind of knowledge, namely, a priori knowledge of the truths of logic and mathematics. Such knowledge is independent of experience, and depends only on the self-evidence of the truths known. When perceptual knowledge and a priori knowledge are conjoined they enable us to acquire general knowledge of the world beyond immediate experience, for the first kind of knowledge gives us empirical data and the second permits us to draw inferences from it.

These two kinds of knowledge can each be further divided into subkinds, described by Russell as immediate and derivative knowledge respectively. He gives the name 'acquaintance' to immediate knowledge of things. The objects of acquaintance include particulars, that is, individual sense-data (and perhaps ourselves), and universals. Derivative knowledge of things Russell calls 'knowledge by description', which is general knowledge of facts made possible by combination of and inference from what we are acquainted with.

Immediate knowledge of truths Russell calls 'intuitive knowledge', and he describes the truths so known as self-evident. These are propositions which are just 'luminously evident, and not capable of being deduced from anything more evident'. For example, we just see that '1 + 1 = 2' is true. Among the items of intuitive knowledge are reports of immediate experience; if I simply state what sense-data I am now aware of, I cannot (barring trivial slips of the tongue) be wrong.

Derivative knowledge of truths consists of whatever can be inferred from self-evident truths by self-evident principles of deduction.

Russell concedes that despite the appearance of rigour introduced by the availability of a priori knowledge, we have to accept that ordinary general knowledge is only as good as its foundation in the 'best explanation' justification and the instincts which render it plausible. Ordinary knowledge amounts at best therefore to 'more or less probable opinion'. But when we note that probable opinions form a coherent and mutually supportive system – the more coherent and stable the system, the greater the probability of the opinions forming it – we see why we are entitled to be confident in them.

An important feature of Russell's theory concerns space, and particularly the distinction between the all-embracing public space assumed by science, and the private spaces in which the sense-data of individual perceivers exist. Private space is built out of the various visual, tactual and other experiences which a perceiver co-ordinates into a framework with himself at the centre. But because we do not have acquaintance with the public space of science, its existence and nature is a matter of inference.


Thus Russell's first version of a theory of knowledge, and because its chief outlines are found in PP it is the one most familiarly associated with his name. But he was by no means content with the expression of it in PP, which after all was a popular book and did not essay a rigorous exposition of its theses. The technical papers, TK and OKEW which followed were his considered versions of these same questions, and mark an advance over this first sketch. One difference between the theories of PP and OKEW is that Russell had come to see that the experiencing subject's basis for knowledge – the sense-data that appear to him alone, and his intuitive knowledge of the laws of logic – is insufficient as a starting point. He accordingly placed greater weight on an experiencer's memories, and his grasp of spatial and temporal relations holding among the elements of occurrent experience. The subject is also empowered to compare data, for example as to differences of colour and shape. Ordinary common beliefs, and belief in the existence of other minds, are still excluded.

This appeal to an enriched conception of cognitive capacities required at the foundations of knowledge is almost invariably made by empiricist epistemologists – consider Locke and Ayer also – when the thin beams of sensory experience and inference are found, as they invariably are, to be insufficient to bear the weight of knowledge.

With this enriched basis of what he now called 'hard data' Russell reformulated the question to be answered thus: 'can the existence of anything other than our own hard data be inferred?' His approach was first to show how we can construct, as an hypothesis, a notion of space into which the facts of experience – both the subject's own and those he learns by others' testimony – can be placed. Then, to see whether we have reason for believing that the spatial world is real, Russell gives an argument for believing that other minds exist, because if one is indeed entitled to believe this, then one can rely on the testimony of others, which, jointly with one's own experience, will underwrite the view that there is a spatial (a real) world.

This strategy is ingenious. In 'The Relation of Sense-Data to Physics' Russell adds an equally ingenious way of thinking about the relation of sense-experience to its objects. In PP he had said that we infer the existence of physical things from sense-data; now he described them as functions of or 'constructions' out of sense-data. This employs the technique of logic in which a thing of one (more complex) kind can be shown to be analysable into things of another (simpler) kind. Russell was here relying on what he called the 'supreme maxim of scientific philosophising', namely the principle that 'wherever possible, logical constructions are to be substituted for inferred entities.' Concordantly with this principle, physical objects are to be analysed as constructions out of sense-data – but not out of actual or occurrent sense-data only, but out of possible sense-data too. For actual and possible sense-data Russell coined the term 'sensibilia' by which is meant 'appearances' or, in Russell's phrase, 'how things appear', irrespective of whether they constitute sense-data currently part of any perceiver's experience. This is intended to explain what it is for an object to exist when not being perceived.

An important aspect of this view, Russell now held, is that sensibilia are not private mental entities, but part of the actual subject-matter of physics. They are indeed 'the ultimate constituents of the physical world', because it is in terms of them that verification of common-sense and physics ultimately depends. This is important because we usually think that sense-data are functions of physical objects, that is, exist and have their nature because physical objects cause them; but verification is only possible if matters are the other way round, with physical objects as functions of sense-data. This theory 'constructs' physical objects out of sensibilia; the existence of these latter therefore verifies the existence of the former.


Such was the epistemology Russell developed in the period to 1914. Instead of developing this distinctive theory further, Russell abandoned it. In later work, particularly AMt and HK, he reverted to treating physical objects, and the space they occupy, as inferred from sense-experience. A number of considerations made him do this. One was his acceptance of the standard view offered by physics and physiology that perception is caused by the action of the environment on our sensory surfaces. 'Whoever accepts the causal theory of perception,' he wrote (AMt p 32), 'is compelled to conclude that percepts are in our heads, for they come at the end of a causal chain of physical events leading, spatially, from the object to the brain of the percipient'. In AMi he gave up talk of 'sense-data', and ceased to distinguish between the act of sensing and what is sensed. His reason for this relates to his acceptance – long in coming, for he had repeatedly resisted it in print – of James's 'neutral monism'.

Another reason for Russell's abandonment of the sensibilia theory was the sheer complexity and, as he came to see it, implausibility of the views he tried to formulate about private and public spaces, the relations between them, and the way sensibilia are supposed to occupy them. He makes passing mention of this cluster of problems in MPD, before there reporting, as his main reason for abandoning the attempt to construct 'matter out of experienced data alone,' that it 'is an impossible programme ... physical objects cannot be interpreted as structures composed of elements actually experienced' (MPD p 79). This last remark is not strictly consistent with Russell's stated view in the original texts that sensibilia are not, and do not have to be, actually sensed; MPD gives a much more phenomenalistic gloss to the theory than it originally possessed. But it touches upon a serious problem with the theory: which is that it is at least problematic to speak of an 'unsensed sense-datum' which does not even require – as its very name seems per contra to demand – an intrinsic connection to perception.

In these early endeavours Russell gave only passing attention to other important questions in epistemology which he later, by contrast, came to emphasise. They concern the kind of reasoning traditionally supposed to be the mainstay of science, namely, non-demonstrative inference. It was some years before Russell returned to consider these questions: the main discussion he gives is to be found in HK, but promissory notes are issued in AMt and IMT.


Acceptance of James's 'neutral monism' was an important turning point. Summarily stated, James's theory is that the world ultimately consists neither of mental stuff, as idealists hold, nor material stuff, as materialists hold, nor of both in problematic relation, as dualists hold, but of a neutral stuff from which the appearance of both mind and matter is formed. By Russell's own account, he was converted to this theory soon after finishing LLA. He had written about James's views in 1914, and rejected them; in LLA itself he was more sympathetic, though still undecided; but finally in a paper entitled 'On Propositions' (1919) he embraced the theory, and used it as a basis for AMi.

The question that came to seem key to Russell is whether consciousness is the essence of the mental, given that, in line with traditional views, consciousness is itself taken to be essentially intentional. In light of Russell's difficulties with the multiple relation theory of judgment it is pointful to remember its partial ancestry in Meinong's view that the intentional relation has at least the three elements of act, content and object. In accepting neutral monism Russell was abandoning the irreducible assumptions of any such view. First, he says, there is no such thing as the 'act'. The occurrence of the content of a thought is the occurrence of the thought, and there is neither empirical evidence nor theoretical need for an 'act' in addition. Russell's diagnosis of why anyone might think otherwise is that we say, 'I think so-and-so', which suggests that thinking is an act performed by a subject. But he rejects this, for reasons similar to those advanced by Hume, who held that the notion of the self is a fiction, and that we are empirically licensed to say no more, on occasions of specifying them, than that there are bundles of thoughts.

Secondly, Russell criticises the relation of content and object. Meinong and others had taken it that the relation is one of direct reference, but in Russell's view it is more complicated and derivative, consisting largely of beliefs about a variety of more and less indirect connections among contents, between contents and objects, and among objects. Add to this the fact that, in imagination and non-standard experiences like hallucination, one can have thoughts without objects, and one sees that the content-object relation involves many difficulties – not least, Russell says, in giving rise to the dispute between idealists who think that content is more significant than objects, and realists who think objects are more significant than content. (Russell's use of these labels, although standard, is misleading: we should for accuracy substitute the label 'anti-realist' for 'idealist' here; this is because whereas, at bottom, realism and anti-realism are indeed differing theses about the relation of contents to objects, and thus are epistemological theses, idealism is a metaphysical thesis about the nature of the world, namely, that it is ultimately mental in character. This point is frequently missed in philosophical debate, so Russell is in good company.) All these difficulties can be avoided, Russell claims, if we adopt a version of neutral monism.

../page 2