Russell, Experience, and the Roots of Science page2

Written by AC Grayling.

In James's view the single kind of 'primal stuff', as he called it, is 'pure experience'. Knowing is a relation into which different portions of primal stuff can enter; the relation itself is as much part of pure experience as its relata.

Russell could not go along with quite all of this. He thought that James's use of the phrase 'pure experience' showed a lingering influence of idealism, and rejected it; he preferred the use made by others of the term 'neutral-stuff', a nomenclatural move of importance because whatever the primal stuff is, it has to be able – when differently arranged – to give rise to what could not appropriately be called 'experience', for example stars and stones. But even with this modified view Russell only partially agreed. He thought that is right to reject the idea of consciousness as an entity, and that it is partly but not wholly right to consider both mind and matter as composed of neutral-stuff which in isolation is neither; especially in regard to sensations – an important point for Russell, with his overriding objective of marrying sense to physics. But he insisted that certain things belong only to the mental world (images and feelings) and others only to the physical world (everything which cannot be described as experience). What distinguishes them is the kind of causality that governs them; there are two different kinds of causal law, one applicable only to psychological phenomena, the other only to physical phenomena. Hume's law of association exemplifies the first kind, the law of gravity the second. Sensation obeys both kinds, and is therefore truly neutral.

Adopting this version of neutral monism obliged Russell to abandon some of his earlier views. One important change was abandonment of 'sense-data'. He did this because sense-data are objects of mental acts, which he now rejected; therefore, since there can be no question of a relation between non-existent acts and supposed objects of those acts, there can be no such objects either. And because there is no distinction between sensation and sense-data – that is, because we now understand that the sensation we have in seeing, for example, a colour-patch just is the colour-patch itself – we need only one term here, for which Russell adopts the name 'percept'.

Before accepting neutral monism Russell had objected to it on a number of grounds, one being that it could not properly account for belief. And as noted, even when he adopted the theory it was in a qualified form; mind and matter overlap on common ground, but each has irreducible aspects. Nevertheless what at last persuaded him was the fact, as it seemed to him, that psychology and physics had come very close: the new physics both of the atom and of relativistic space-time had effectively dematerialised matter, and psychology, especially in the form of behaviourism, had effectively materialised mind. From the internal viewpoint of introspection, mental reality is composed of sensations and images. From the external viewpoint of observation, material things are composed of sensations and sensibilia. A more or less unified theory therefore seems possible by treating the fundamental difference as one of arrangement: a mind is a construction of materials organised in one way, a brain more or less the same materials organised in another.

A striking feature of this view is, surprisingly, how idealist it is. Russell had, as noted, charged James with residual idealism. But here he is arguing something hardly distinguishable: that minds are composed of sensed percepts–viz. sensations and images–and matter is a logical fiction constructed of unsensed percepts. Now Russell had often insisted (using his earlier terminology) that sensibilia are 'physical' entities, in somewhat the sense in which, if one were talking about an item of sensory information in a nervous system, that datum would be present as impulses in a nerve or activity in a brain. But then nerves and brains, as objects of physical theory, are themselves to be understood as a constructions from sensibilia, not as traditionally-understood 'material substance', the concept of which physics has shown to be untenable. At the end of AMi (pp 305, 308) Russell accordingly says that 'an ultimate scientific account of what goes on in the world, if it were ascertainable, would resemble psychology rather than physics ... [because] psychology is nearer to what exists'. This explains Russell's notorious claim that 'brains consist of thoughts' and that when a physiologist looks at another person's brain, what he 'sees' is a portion of his own brain (Schilpp p 705).

For robuster versions of materialism this aspect of Russell's view is hard to accept. But it is not the only difficulty with his version of neutral monism. Not least among others is the fact that he failed in his main aim, which was to refute the view that consciousness is essential to the distinction between mental and physical phenomena. He had not of course attempted to analyse consciousness quite away; his aim was rather to reduce its importance to the mind-matter question. But images, feelings and sensations, which play so central a role in his theory, stubbornly remain conscious phenomena, whereas the sensibilia (by definition including unsensed sensa) which constitute the greater part of matter are not. Russell accepted this, but tried to specify a criterion of difference which did not trade on these facts, namely, the criterion of membership of different causal realms. But whereas that difference is open to question – and even if it exists might be too often hard to see – the consciousness difference is clear-cut. Relatedly, the intentionality which characterises consciousness cannot be left out of accounts of knowledge; memory and perception are inexplicable without it. Russell later acknowledged this point, and gave it as a reason in MPD for having to return to the question of perception and knowledge in later writings.

He also later came to abandon the idea – anyway deeply unsatisfactory from the point of view of a theory supposed to be both neutral and monist – that images and feelings are essentially mental, that is, not wholly reducible to neutral-stuff; for in a very late essay he says, 'An event is not rendered either mental or material by any intrinsic quality, but only by its causal relations. It is perfectly possible for an event to have both the causal relations characteristic of physics and those characteristic of psychology. In that case, the event is both mental and material at once'. This, for consistency, is what he should have argued in AMi itself, where only sensations have this character.

But this view in turn generates another problem, which is that it comes into unstable tension with a view to which Russell returned after AMi, namely, that the causes of percepts are inferred from the occurrence of the percepts themselves. As noted earlier, Russell wavered between treating physical things as logical constructions of sensibilia and as entities inferred as the causes of perception; he held this latter view in PP and returned to it after AMi. But on the face of it, one is going to need a delicate connection between one's metaphysics and one's epistemology in order to hold both that minds and things are of one stuff, and that things are the unknown external inferred causes of what happens in minds. So those parts of the legacy of AMi which remain in his later thinking raise considerable difficulties for his views there about matter.


One of the chief reasons for Russell's reversion to a realistic, inferential view about physical things was the difficulty inherent in the notion of unsensed sensa or, in the later terminology, percepts. As noted above, the idea had been to replace inferred entities with logically constructed ones. If physical things can be logically constructed out of sensibilia, then two desiderata have been realised simultaneously: the theory is empirically based, and inferred entities have been shaved away by Ockham's Razor. But it is obvious that the idea of unsensed sensa (or unperceived percepts) is, if not indeed contradictory, at least problematic. It makes sense – although, without a careful gloss, it is metaphysically questionable – to talk of the existence of possibilities of sensation; but to talk of the existence of possible sensations arguably does not (recall Russell's definition of sensibilia as entities having the 'same metaphysical and physical status as sense-data without necessarily being data to any mind'.) If the choice lay between inferred material particulars and non-actual perceptions existing unperceived, it would seem best to accept the former. This is just what Russell himself came to think. But he did not return to the cruder form of inferential realism held in PP; he had something more ingenious – though in the end no more successful –up his sleeve.

Another reason for Russell's reversion to realism was his recognition that the notion of causality is problematic for phenomenalism. Things in the world seem to affect one another causally in ways hard to explain on the mere basis of reports of sense-experience. Moreover, a causal theory of perception is a natural and powerful way of explaining how experience itself arises. In Russell's mature philosophy of science, contained in AMt and HK, he did not opt for a Lockean view which says that our percepts resemble their causal origins, on the ground that we cannot be directly acquainted with things, and therefore cannot expect to know their qualities and relations. Rather, he now argued, changes in the world and our perceptions are correlated, or co-vary, at least for orders of things in the world that our perceptual apparatus is competent to register (we do not, for example, perceive electrons swarming in the table, so there is no associated covariation of world and perception at that level). The correspondence between percepts and things is one of structure at the appropriate level: 'Whatever we infer from perceptions it is only structure that we can validly infer; and structure is what can be expressed by mathematical logic' (AMt 254). And this means that we have to be 'agnostic' about all but the physical world's mathematical properties, which is what physics describes (ibid 270).

Russell had come to think that the best candidate for what is metaphysically most basic in the world is the 'event'. Objects are constructed out of events in the following way: the world is a collection of events, most of which cluster together around a multitude of 'centres' thus constituting individual 'objects'. Each cluster radiates 'chains' of events, which interact with and react upon chains radiating from other centres–among which are perceivers. When a chain interacts with the events constituting the perceptual apparatus of a perceiver, the last link in the chain is a percept. Since everything is ultimately constituted of events, they are in effect the 'neutral-stuff' of which minds and material things are made. Minds are clusters of events connected by 'mental' relations, not least among them memory; otherwise there is no metaphysical difference between mind and matter. Finally, the interrelations of event-chains is what scientific causal laws describe.

This view enabled Russell to formulate the argument he had long been trying to state satisfactorily, namely, that percepts are parts of things. For on this view it is not the case that there are events which constitute things, and then in addition other events which are perceptions of those things; rather, there are just events constituting the object, some of which are percepts – these being the terminal events of the chains radiating from the object which interact with events constituting the perceiver.

This theory is inferential not in the earlier sense in which the causes of percepts, lying inaccessibly beyond a veil of perception, are guessed from the nature of the percepts themselves. Rather, the inference is from certain terminal events, viz. percepts – which are interactions between (using the term heuristically) 'mental' events and that level of structure in the rest of the event-world with which the 'mental' events are capable of interacting – to the clusters and chains of events constituting the world as a whole.

In AMt the core of the theory is the idea that knowledge of the world is purely structural. We know the qualities and relations as well as the structure of percepts, but we know only the structure of external events, not their qualities. This seems somewhat reminiscent of Locke's distinction between primary and secondary qualities, but it is not; Russell is saying that all we can infer from our percepts is the structure of the qualities and relations of things, not the qualities and relations themselves; and that this is the limit of knowledge.

This theory has a fatal flaw, which was quickly recognised by the mathematician M. H. A. Newman and set out in an article published soon after the appearance of AMt. It is that since our knowledge of the structure of events is not a mere result of our stipulating them, but is manifestly non-trivial, it follows that our inferential knowledge cannot be limited solely to questions of structure. This is because – to put the point by a rough analogy – a number of different worlds could be abstractly definable as having the same structure, and if they were, knowledge of their structure alone could not separate them and in particular could not individuate the 'real' one. If science genuinely consists of discoveries about the world through observation and experiment, the distinction between what we observe and what we infer cannot therefore be collapsed into a distinction between pure structure and qualities.

Russell accepted Newman's point: 'You make it entirely obvious that my statements to the effect that nothing is known about the physical world except its structure are either false or trivial, and I am somewhat ashamed not to have noticed it myself.'


As repeatedly noted, the common thread linking Russell's earlier and later views is the aim of securing the move from perception to the objects of physical theory. On his view, this move must either be inferential, in which it takes us from the incorrigible data of sense to something else, or it is analytic, that is, consists in a process of constructing physical entities out of percepts. On the later view just reported, the inference has a special advantage over more usual inferential theories, in that the inference is not from one kind of thing to another, but from one part of something to its other parts.

In his earlier views Russell had accorded primary reality to sense-data and built everything else out of them. On the later view, reality belongs to events as the ultimate entities, and an important change of emphasis was introduced: percepts remain immediate and as certain as anything can be, but they are not construed as having accurately to represent the physical world, which, in the picture offered by science as the most powerful way to understand it, is anyway very different from how it appears.

Crucially, however, there remains a familiar and major problem about whether inferences from perception to the world are secure. A large part of Russell's aim in HK was to state grounds for taking them to be so. Throughout his thinking about the relation of perception and science he was convinced, as his above-quoted remark in the October 1912 letter to Ottoline Morrell shows, that something has to be known independently of experience for scientific knowledge to be possible. Earlier, as noted, he thought that purely logical principles provide such knowledge. But he now saw that logic alone is insufficient; we must know something more substantial. His solution was to say that inference from perception to events is justified in the light of certain 'postulates' which nevertheless state contingent facts about the word. So stated, Russell's view immediately reminds one of Kant's thesis that possession of 'synthetic a priori knowledge' is a condition of the possibility of knowledge in general, a view which Russell robustly dismissed in the Preface to HK. The difference is explained by the tentative and probabilistic account that Russell, in this last major attempt to state a theory of knowledge, felt was all that could be hoped for.

Two features of Russell's approach in HK explain this result. One is that he now thought that knowledge should be understood in 'naturalistic' terms, that is, as a feature of our biological circumstances, taken together with the way the world is constituted. The other is that he had come to make a positive virtue of the fact (which he always otherwise accepted) that contingent knowledge is never certain, but at best merely credible to some degree. This second point enters into the detailed working out of the views in HK. The first makes its appearance whenever Russell needs to justify the justifications which HK attempts to provide for scientific knowledge.

When data have a certain credibility independently of their relations to other data, Russell describes them as having a degree of 'intrinsic' credibility. Propositions having some intrinsic credibility lend support to propositions inferred from them. The chief question then becomes: how do propositions with some measure of intrinsic credibility transfer that credibility to the hypotheses of science? Another way of framing the question is to ask how reports of observation and experiment can function as evidence. This is where Russell's postulates come in.

There are five postulates. The first, 'the postulate of quasi-permanence', is intended to replace the ordinary idea of a persisting thing: 'given any event A, it happens very frequently that, at any neighbouring time, there is at some neighbouring place an event very similar to A'. Thus the 'things' of common-sense are analysed into sequences of similar events. The ancestor of this idea is Hume's analysis of the 'identity' of things in terms of our propensity to take a sequence of resembling perceptions to be evidence for a single thing, as when you have perceptions of a rose bush every time you go into the garden, and therefore take it that there is a single persisting rose bush there even when no perceivers are present.

The second, 'the postulate of separable causal lines', states that 'it is frequently possible to form a series of events such that, from one or two members of the series, something can be inferred as to all the other members'. For example, we can keep track of a billiard ball throughout a game of billiards; common-sense thinks of the ball as a single thing changing its position, which according to this postulate is to be explained by treating the ball and its movements as a series of events from some of which you can infer information about the others.

The third is 'the postulate of spatio-temporal continuity', designed to deny 'action at a distance' by requiring that if there is a causal connection between two events that are not contiguous, there must be a chain of intermediate links between them. Many of our inferences to unobserved occurrences depend upon this postulate.

The fourth is 'the structural postulate', which states that 'when a number of structurally similar complexes are ranged about a centre in regions not widely separated, it is usually the case that all belong to causal lines having their origin in an event of the same structure at the centre'. This is intended to make sense of the idea that there exists a world of physical objects common to all perceivers. If six million people all listen to the Prime Minister's broadcast on the wireless, and upon comparing notes find that they heard remarkably similar things, they are entitled to the view that the reason is the common-sense one that they all heard the same man speaking over the airwaves.

The fifth and last is 'the postulate of analogy', which states that 'given two classes of events A and B, and given that, whenever both A and B can be observed, there is reason to believe that A causes B, then if, in a given case, A is observed, but there is no way of observing whether B occurs or not, it is probable that B occurs; and similarly if B is observed, but the presence or absence of A cannot be observed'. This postulate speaks for itself. (HK 506-12)

The point of the postulates is, Russell says, to justify the first steps towards science. They state what we have to know, in addition to observed facts, if scientific inferences are to be valid. It is not advanced science which is thus justified, but its more elementary parts, themselves based on common-sense experience.

But what is the sense of 'know' here? On Russell's view, the knowing involved in 'knowledge of the postulates' is a kind of 'animal knowing', which arises as habitual beliefs from the experience of interaction with the world and experience in general. It is far from being certain knowledge. 'Owing to the world being such as it is,' Russell says, 'certain occurrences are sometimes, in fact, evidence for certain others; and owing to animals being adapted to their environment, occurrences which are, in fact, evidence of others tend to arouse expectation of those others. By reflecting on this process and refining it, we arrive at the canons of inductive inference. These canons are valid if the world has certain characteristics which we all believe it to have' (HK 514-5). These are the common-sense facts that the postulates in effect embody, and it is in this sense that we 'know' them. They are implied in the inferences we make, and our inferences are by and large successful; so the postulates can be regarded as in a sense self-confirming.

Although Russell thinks of the postulates as something we know a priori, it is clear that their status is odd. They are in fact empirical in one sense, since they either record or are suggested by experience. What gives them their a priori status is that they are treated as known independently of empirical confirmation (except indirectly in practice), rather than as generalisations in need of such justification. In effect Russell selected some general contingent beliefs which are especially useful to have as premises in thinking about the world, and elevated them to the dignity of postulates. Their indirect justification, in turn, is that on the whole they, or the results of their application, work. Allied to the extremely modest ambition Russell has for epistemology in HK this might be enough. But it has no pretensions to be a theory of knowledge as traditionally conceived, nor a rigorous account of non-demonstrative reasoning.

These last remarks suggest why Russell's arguments in HK received little response, much to his disappointment. He recognised well enough that canons of evidence and scientific reasoning are worth investigating only if we can be confident that, if we got them right, they would reliably deliver science. But the most that Russell's argument establishes is that, so far, the general principles on which our empirical thinking relies have been largely successful. But this looks like exactly the kind of unbuttressed inductive inference Russell was anxious to caution against, citing the example of the chicken who, on being fed day after day, grew increasingly pleased with the world – until the day the butcher came.

In particular, we have no guarantee against the possibility that use of the postulates leads to falsehood, either occasionally or in some systematic way. Now this possibility is in effect allowed by Russell in asking very little of epistemology. The complaint must therefore be that the argument in HK is in fact an admission of failure, when taken in the light of the epistemological tradition. Descartes and his successors in modern philosophy raised questions about the nature of knowledge and how we get it precisely so that they could distinguish between some enterprises – alchemy, astrology, and magic, say – and others – chemistry, astronomy, and medicine, say – which differ not merely in the number of genuinely practical applications they offer, but in telling us something true about the world; and where, moreover, the latter fact explains the former, and opens the way to more of both by the same route. Moreover, our ancient prejudices and animal beliefs might be controverted in the process, as indeed happens: for the world depicted by science is remarkably different from the world of common-sense. But Russell in HK says the utility of applications and those same animal habits of belief are the only final justification we can hope for in epistemology. This is very much less than the project of epistemology traditionally aims to achieve, and it is much less than Russell himself hoped to achieve on first launching his epistemological project after PM.


Russell had charged Kant with a 'Ptolemaic counter-revolution' in the Preface to HK, but it is not clear that HK itself escapes a Ptolemaic tinge. The postulates are expressly not transcendentally necessary framework features in any sense comparable to Kant's categorial concepts, or to any other species of foundational principle. They are in effect rules of thumb, 'distilled' as Russell puts it, from the epistemological pragmatics of common sense, and justified – if that is the right thing to expect them to be – by their manifest utility in scientific enquiry and ordinary life.

Nevertheless, they prompt two thoughts. One is that a solid argument can be given in favour of strengthening postulates of the kind envisaged by Russell into structural conditions of inquiry. For what are in effect temperamental reasons it was not open to Russell to consider investigating, by means of transcendental arguments, what is required for the possibility of the kind of knowledge in which science consists. No doubt the precipitate of something like the postulates would result; and that is a suggestive thought. Such an argument would be in fact Russellian, because it would follow his example in his earlier epistemological work of seeking the logical distribution of the problem, so to speak, as when, in the 1911-14 work, he distinguished what was logically primitive from what was derived from it, and how both parts of this classification related to one another in the structure they formed.

It is of course no more than a coincidence, but a remarkable one, that at the time Russell was writing HK, Wittgenstein was coming to not dissimilar conclusions in On Certainty – as if they had been travelling different routes and arriving at near-points at the end of the journey. Wittgenstein's late interest in problems of scepticism and knowledge is rather striking in being straightforward workaday philosophy of just the kind he earlier dismissed as fly-in-the-bottle. His interest in epistemology therefore looks like acceptance that philosophical problems are real ones after all, amenable to investigation – and even solution. His contribution is to insist on the internal connection between the concepts of knowing and doubting, and equally to insist that epistemic justification is provided by the conceptual scheme within which talk of knowledge and doubt alone gets content. The similarities between the very late Russell and Wittgenstein lie in the thought that (to put the matter neutrally as between them) a given area of discourse requires that we accept certain things in order to be able to get along in it – the 'grammatical' propositions which key a discourse's sense, in Wittgenstein; the postulates required by inquiry, for Russell. Of course the parallel is not direct, but it is suggestive.