Science Versus Science: Russell’s Problem of the External World

Bertrand Russell believed that the results of science caused a major and deep philosophical problem for our knowledge of the external world. To him, science had revealed that our knowledge about external objects was inferred and not direct, and as that knowledge was inferred Russell believed it therefore required justification. However, the fact that science in general did not itself see Russell’s problem or, even worse, believed that it had solved it raises issues for that argument. What side, if any, does science support? Do those inferences require the sort of justification that Russell argues they do? Can the sort of justification that Russell is looking for work to solve Russell’s problem, if there even is one? Is Russell’s problem a scientific one, or a philosophical one, or both?

Answering these questions requires starting by examining how Russell uses science to identify his problem. He starts out attempting to sort his beliefs into the categories of those which he would find it impossible or at least unreasonable to doubt and those which we can quite reasonably doubt. The former are what he calls “hard data” and the latter are what he calls “soft data”. But since we have to be able to reasonably doubt, Russell needs to define, at least in general, what forms the basis for reasonable doubt as opposed to unreasonable doubt if he is not to succumb to radical skepticism. Taking our instinctive beliefs, we can only reasonably doubt if that doubt is introduced by another belief, as he says: “It is of course possible that all or any of our beliefs may be mistaken, and therefore all ought to be held with at least some slight element of doubt. But we cannot have reason to reject a belief except on the ground of some other belief” [Russell, Problems of Philosophy, pg 8]. Russell talks about being unable to refute a universal scepticism, but points out “… data can only be criticized by other data….” [Russell, On Our Knowledge of the External World, pg 74]. So to legitimately cast any data into doubt means bringing other data to bear against it. Otherwise, one merely entertains verbal but not legitimate doubt. So Russell will need to bring other data to the table to get us doubting that there are tables.

What data will this be? Well, it turns out that this is, at least in part, the data afforded by physics, biology and physiology: “In attributing our perceptions to a normal causal origin outside ourselves, we run a certain risk of error … there may be reflection or refraction on the way to the eye, there may be an unusual condition of the eye or optic nerve or brain. All these considerations give a certain very small probability that, on a certain occasion, there is not such an outside cause as we suppose.”[Russell, An Outline of Philosophy, pg 136]. Ultimately, when we take the results of science, we discover that we do not have a direct line to the object outside of ourselves. Our sensory data is mediated by all sorts of things, such as the transmission medium, the state of our receptors, and the state of our brain that processes the data transmitted. All of these can, in fact, introduce errors into the perception. Moreover, not only is it possible that these things can introduce errors into our perception of the external object that is ostensibly causing our sense perceptions, science tells us that these things actually do, in fact, introduce errors into our perceptions. Sometimes, we get things wrong, such as when we see a stick bend when we put it into water, or when we dream and mistake the dream for a real experience. So, not only does science tell us that the method by which we get perceptions from external objects could be wrong, it tells us that a not insignificant amount of the time they simply are wrong. And this is vitally important; if that method was not direct but could not introduce error, there would be no problem for Russell. That it can introduce error is what really cements, in Russell’s mind, that he has a serious problem here.

At this point, we have been talking about data in general, not what Russell will call “hard” or “soft” data. For Russell “… our investigation starts from what may be called “data”, by which I mean matters of common knowledge, vague complex, inexact, as common knowledge always is, but yet somehow commanding our assent as on the whole an in some interpretation pretty certainly true.”[Russell, On Our Knowledge of the External World, pg 72]. The results of physics are certainly data for Russell, but so are the sense perceptions of tables and things like that. Russell then moves on to argue that there are different grades of data: “… yet we may distinguish different grades of certainty in the different kinds of common knowledge … “ [ibid, pg 74]. At the end of it all, he distinguishes between two broad categories of data: hard data and soft data. “I mean by “hard”data those which resist the solvent influence of critical reflection, and by “soft” data those which, under the operation of this process, become to our minds more or less doubtful.” [ibid, pg 78]. By this, for Russell, hard data are those data which critical analysis does not impact; it commands our assent and as we examine it critically it still commands our assent in precisely the same way. Soft data, then, would be data that as we critically examine it we can see it either firming or softening up; we either start to assent to it more or less based on the results of our examinations. While this is a continuum, Russell argues that “The hardest of the hard data are of two sorts: the particular facts of sense, and the general truths of logic”[ibid, pg 78]. There are other things that can be called at least mostly hard data, but these are the two things most certain. For Russell, then, the problem in general is how to move from hard data to soft data, and for the external world in particular the problem is how to move from the hard data of particular facts of sense – which is what forms the foundation of our beliefs about the external world – to the soft data of existent tables.

The first problem we have to note is this: the results of science themselves are not, in fact, included in the list of hard data that Russell relies on. And for good reason, as most if not all experimental sciences rely heavily on precisely the sort of inference from sense data to external objects that Russell is now doubting. As seen above, in order to have something more than a verbal doubt we must have data to cause the doubt, and we must presume that the data must be credible and reliable. But if the data afforded by science is now soft data, is it still credible and reliable? Well, it is certainly open to doubt itself, and so one might credibly doubt it. If, then, the assent to the results of science causes us to wonder if we should assent to the results of science, and if the results of science are challenged such that they can be legitimately doubted, on what grounds do we maintain the doubt? This is especially important considering that Russell’s conclusion is not – at least initially – itself among what Russell calls data; he is claiming to derive it from other data. But surely it is not reasonable to derive a conclusion and argue that it should be added to our set of data if it would simply undercut the data that generated it. Can Russell argue convincingly that we should add this hard and soft data distinction and the need to justify the move from hard to soft data if adding the belief that there is a problem would make the data that generated the problem itself soft data?

Russell can argue here that the results of science really do force us to introduce the belief that there is a problem of justification here, even though that makes those results soft data. When we critically examine what science says about this, we can see that we are indeed forced to make inferences to justify our beliefs about the external world. We do not, in fact, have direct access to the objects in the external world. Thus, in this case, we are not trying to add a belief to our set of data based on data that it makes dubious, but we are forced to add a belief that our perceptions are mediated to our set of data, and when we do that we discover that including that belief means that the data of science is soft data, which means no more than it is data that critical examination both casts into doubt and hopefully will make firmer if we do critical analysis in the right way. We must add the belief that our access to the external world is mediated and once we do that we can see that doing that makes science soft data, and so there is no contradiction.

However, since we are allowing that the results of science count as data, we note that at the end of the day it still is the case that Russell generates the problem on the basis of what is now soft data. This would not seem to be a problem except for the fact that Russell then claims that the problem is how to move from hard data to soft data; no soft data is to be allowed. It seems reasonable to ask how it can be that a problem generated by soft data cannot be solved by soft data. If there was no soft data available to solve the problem or if it seemed clear that all of the possible soft data that we could have access to could not solve the problem, then the contention would be reasonable. However, that is not the case, because the scientific data includes soft data that allows us to justify our inferences in the cases that Russell cites. The physical laws of light allow us to determine when the intervening media are influencing our perceptions, often simply just from the perceptions themselves. We can note that in a certain situation according to the theory of light that the intervening media will, say, make things look bluer than normal, like when we wear blue eyeglasses. Under those conditions, then, the colour of the object will not be accurate, and we will have to take steps to remedy the situation to get a true view of the colour of the object. Science will also tell us exactly what steps to take to remedy the situation. All of this, then, is part of our data.

The charge here is that Russell is being quite selective about what data he is using. In science, the facts that can be used to justify the inferences – including the inference that our sense perceptions are, in fact, mediated and can be in error – are built into the complete theories themselves. So to take the results of, say, theories of light and vision that generate the problem while then dismissing the entire rest of the theory when we attempt to solve it seems invalid. How can Russell justify taking the theory of vision, say, as anything other than a complete theory, which means including the data that explains when vision is to be relied upon and when it is not? It seems that if Russell is going to use the scientific theory of vision to generate his problem, he must accept all of it. And thus he will not be able to then retreat to arguing that only the “hard data” version of vision is allowed to solve the problem.

We can see undercurrents of this in the one of the criticisms of Russell by John Dewey. Dewey argues this about vision: “That color is visual in the sense of being an object of vision is certainly admitted in the common sense world, but this is the world we have supposedly left. That color is visual, is a proposition about color and it is a proposition which color itself does not utter.” [Dewey, Existence of the World as a Problem, pg 359]. And later: “But by description all that is required for the existence of color is certain physiological conditions” [ibid]. Russell’s reply is this: “We do not need to experiment by shutting the eyes and stopping the ears in order to find out whether the sense-datum of the moment is “visual” or “auditory:” we know this by its intrinsic quality.” [Russell, Review: Experimental Logic, pg 248]. Dewey, then, accuses Russell of going beyond hard data to get questions of the sense data being visual off the ground, and Russell ripostes with the argument that it is not necessary to experiment with ones sensory organs in order to distinguish visual sense data from auditory sense data in some manner. Russell’s reply, of course, is quite correct; it does seem quite possible to distinguish between visual and auditory experiences intrinsically, even if only as a basic classification. Dewey’s focus, then, on defining things like “visual” from the third-person view seems to allow for a quick and easy rebuttal of his charges. The problem is that at this point we can see that Russell has changed definitions on us. The response is assuming that one can distinguish between visual and auditory data at the level of basic sense data, from the sense perceptions themselves. At that level, Russell is absolutely right that you can tell the difference without appealing to eyes and ears. However, as we have seen, while it may indeed be true that one can distinguish “visual” and “auditory” purely from sense data, Russell is not, in fact, using that definition of “visual” in his argument. Russell, as we have seen, is using the scientific, third-person theory of vision as the basis for his problem. And that definition includes things like light. And eyes. And brains. And all sorts of external objects. There is a complete theory there, and a complete theory that includes a lot more than simple sense data. At the level of sense data, however, one cannot introduce this problem of inference; all we have are sense data, not the objects inferred or the inferences themselves. Thus, we can see that Russell can indeed be accused of using the full scientific theory to generate his problem and then immediately retreating away from it to answer objections like Dewey’s.

Russell can reply that the issue is not with the results of science; they are all data and so are all things we should give assent to. The problem, however, rests on discovering that what we were and were always doing was inferring the facts about the world from our internal sense perceptions. Thus our knowledge of tables and chairs and even the laws of physics is generated by inference and not direct access or apperception. Since we have to make inferences, we must justify those inferences. Russell’s demand, then, is for a justification of the inferences we are making.

However, this demand can be turned back on him. Clearly it is an inference that our sense perceptions being mediated mean that we are always doing inferences that need justification. It may seem like an exceptionally safe inference, and one that we are not prone to doubt, but then again our inferences about tables and vision seem equally safe. Thus, the contention that we need to justify the inferences to tables and cannot use soft data to justify those inferences itself is inferred and thus open to reasonable challenge. Moreover, it is challenged by physics itself, which claims to have, in fact, provided justified inferences for the things that Russell claims need to be justified. Thus, Russell needs to justify this inference just as much as the inferences about tables and the scientific data of vision. If Russell claims that this contention is justified by an appeal to physics itself, then he is appealing to soft data to justify that contention. This should, then, allow soft data to be used to justify our inferences about tables as well. If Russell denies that soft data can be used to justify these sorts of inferences, then it seems that he cannot use the data of physics to justify his inference either.

Russell can argue that he is allowed to use anything that is considered data to justify his inferences that generate the problem, and that the data of science is clearly data to him. Thus, he can indeed still use it without caring whether the data is soft or hard. Once we generate the problem, the distinction between hard and soft data becomes relevant and thus the problem can be described as how to make inferences from hard data to soft data. This move, however, is still problematic because if he simply says that data in general can always be used, then we can use the data of science to solve the issue as well. After all, data is data, right? Thus, Russell needs to justify his inference that there is a problem of inferring from hard to soft data for our knowledge of the external world by some data that cannot also be used to solve it. That data is not, in fact, going to be scientific data as long as science says that they know not only that our access to the external world is mediated but that we can still get access to the real objects with justified inferences as long as we stick to the full and complete data afforded by science. So if it is not scientific data that Russell can use to justify his inference, what data is it?

Examined this way, it becomes clear that Russell – despite his claims – is at least not using only scientific data to justify his inference. He is starting from a prior set of assumptions that are what leads him to his conclusion. It is obvious that one of them is that things that are inferred require justification. However, there is one other presumption that he uses to justify his contention, which is about what is the hardest of the hard data: the particulars of our sense data. Russell thinks that these are, in fact, indubitable, and as indubitable as they can be. “Even, therefore, when we assume the truth of physics, what we know most indubitably through perception is not the movements of matter, but certain events in ourselves … when Dr. Watson watches rats in mazes, what he knows, apart from difficult inferences, are certain events in himself.”[Russell, An Outline of Philosophy, pg 140]. What we cannot doubt and what we need not infer are, in fact, the particulars of our sense data. From there, we can get the claim that the particulars of our sense data are hard data and from there the argument that we need to justify the inferences that produce data about the external world on those non-inferred and hard data of the particulars of sense data.

Note that this is not quite a presumption. Russell engages in a lot of reasoning about our derivative knowledge and how it relates to how trustworthy it seems when we subject it to critical examination. If we are deriving our beliefs psychologically – ie they are based on other beliefs – but if we do not have a logical argument justifying them, then those become quite doubtful when we apply critical reasoning for them. We either provide an argument for them – making them psychologically and logically derivative – or we feel that we should provide an argument for them and so doubt them. We definitely derive our beliefs about the external world from other beliefs, using inference, and it does seem that we do not have a sufficient logical argument to justify those beliefs. Thus, there is a problem. We definitely need something to justify it, and Russell then argues that we can do this by grounding it in the things that do not require justification. Thus, we can do this – and only do this, under Russell’s view – by justifying inference from hard data to soft data. The particulars of sense data are among the hardest of the hard data and are directly related to experiences of the external world, and so that is where we will have to start in justifying our inferences to the external world.

Note, however, that there is still an issue here, since Russell is arguing that we have to base our inferences on the particulars of sense data because the particulars of sense data will provide a firm foundation for further inference. However, this is precisely what Russell’s opponents will deny, and it is precisely what the science that produces the theory of vision does not do. None of these accept that the particulars of sense data are an indubitable ground for knowledge. They reject that the particulars of sense data are such a ground because of their inherent privacy. They are accessible only to the individual and cannot be accessed at all publically. Even expressions of private data publically do not give access to this data. It is personal and completely subjective, and the demand is that the data be objectively accessible. If the data is only privately accessible, then it cannot be checked or tested and cannot be shared among individuals; we have no idea if we are having the same particulars of sense data as someone else. While I may see red patch, someone else may see yellow patch, even though both of us may report red patch. How, then, can this private data ground knowledge in any way? Perhaps Russell, in his office, may claim to come to know based on this, but he cannot justify that in any way based on his private data to anyone else, since they do not have access to his private data. He would have to appeal to the private data of the other person to justify his knowledge, but Russell again has no access to that private data. If it goes awry and the other person puts themselves into the same situation as Russell and claims to see “yellow patch”, or if that person claims to see “red patch” but does not accept the knowledge claims based on that private data, what can Russell say? What can Russell do? Private data, then, may well be more doubtful than public inference if one wants to argue for one’s knowledge beyond one’s own personal epistemic commitments.

Dewey, again, makes this argument clear by denying that the particulars of sense data are known at all. “It is often stated, for example, that primitive sense objects when they are stripped of all inferential material cannot possibly be false – but with the implication that they, therefore, must be true. Well, I meant to go this statement one better – to state that they are neither true nor false – that is, that the distinction of true-or-false is as irrelevant and inapplicable as to any other existence … This position when taken leaves over the question of sense perception as knowledge, as capable of truth or falsity.”[Dewey, Sense Perception as Knowledge, pg 50]. Unlike Russell, who uses the particulars of sense data as being some kind of known truths that can then be the basis for all knowledge, Dewey denies that they are knowledge at all. They exist, certainly, but they cannot be true or false, and so cannot be known. Now, a lot of this is driven by Dewey’s specific views on inference and validation. However, it is clear that for Dewey a big part of this is the fact that the sense data is private and not publically or empirically accessible. No one, as already stated, can know what someone else’s private data is. You cannot test your private data in a public fashion that can be agreed upon by multiple persons or even multiple perspectives. For this data is not only private, it is also immediate. Even changing positions or ways of observing will change this data, requiring some kind of link to stitch it all together into a coherent whole, a link that will itself require justification. And these links, again, cannot be tested against anyone else’s perceptions either. What use, then, in calling it knowledge? To whom?

It is clear that Dewey and Watson share this sort of objection to Russell, as evidenced by Russell’s challenge to behaviourism in “Physics and Perception”. And they seek to replace Russell’s reliance on private data with a different way to justify inferences. While there is far more to either of their theories than will be related here, the essential and hopefully mostly neutral description is this: Beliefs about the external world are tested in the external world by a combination of actions and expectations. First, a theory or hypothesis is formed. Second, the consequences of this theory or hypothesis are worked out in terms of publically observable expectations, where the expectations generally work out to “If we do X in the world, what would we expect to see occur by this hypothesis?”. Then, everyone is obliged to go out into the external world and do X. If the expectations are met, the theory or hypothesis is justified in some small way and therefore so is the inference. If the expectations are not met, then the justification for the inference is undercut, and so it must go back to the theorizing or hypothesizing or even all the way back to the belief formation stage for correction, which will result in a new hypothesis or theory and a new set of expectations and actions. The advantage of this is that all of this is public and accessible to everyone. The logical formation of expectations is based on logic – which is hard data for Russell – and the actions that prove those expectations can be taken by anyone. Thus, if someone has any doubts about the inference they can be walked through the theory and hypothesis and the expectations formed by them, and then invited to take the actions to test the expectations themselves. At the end of all of that, either the doubter will have their doubts erased by seeing the expectations fulfilled, or will have specific reasons why the justification is insufficient and thus why the inference is not justified.

Thus, for Dewey and Watson and, it seems, for physics itself if given a choice between relying on the private particulars of sense data or this inference tested by expectations, they will always choose the latter. Russell, of course, wants us to choose the former. His argument against taking this sort of course is essentially that this attempt to get out to the public realm and away from private and subjective data is doomed to failure from the start: “ … our perceptions are therefore infected with subjectivity on purely physical grounds.” [Russell, An Outline of Philosophy, pg 139]. Science demonstrates that our sense data are, in fact, subjective and in some sense only in us. They depend on a bunch of intervening states that are more or less particular to us, and so are subjective by definition. However, regardless of what anyone says our primary access to anything about this public world is through this subjective sense data. What Watson sees when he observes the rat is something he “sees”, subjectively. Vision, while causally connected to things in the world according to physics, is also a subjective experience according to physics. Even the reports of others are filtered through subjective sense data that is, in fact, filtered through the same sorts of subjective mechanisms as our own direct experiences. Thus, to get to any sort of public world and get any sort of results that can be seen at the public level requires going through inference from the private to the public. If that first inference is not justified, what can be done at the public realm to justify it?

We can now see how the philosophical debate is starting to shape up by comparing the most obvious problem that each position has. As Russell points out, since all public testing and verification has to rely on our primary means of access to the public realm, and since that primary means requires that we justify the inferences we make using it to what is really out there in the public realm, if we cannot justify those inferences we are completely stuck. However, the counter to Russell is that public realm problems must be justified publically. If we retreat to private data we will never be able to return to the public realm with our justifications; they will remain private since no one else can get access to our private data, and we have no reason to think that their private data and ours are the same, and no way to determine that publically. Thus, we can see the choice this leaves for us: either we retreat to the private realm with the worry that we will never be able to escape it, or we remain in the public realm with the worry that the underpinnings of our access to that world are unjustified.

Note here that science seems to be supporting both sides, depending on which problem one finds more problematic. The facts of physics do seem to support Russell’s contention that at least for the most part the particulars of our sense data are subjective and private; the various conditions seem to indeed allow for a red sense datum to be produced based on the particular state of one individual, even if some of those conditions are more inter-subjective. On the other hand, the theory that supports that move has been verified by inferences tested by expectations, and so the results of the theory that show precisely when we should trust our experiences and when we should not, and how to correct for these error-inducing conditions shows how verified testing of expectations claims to solve the problem without having to retreat to private data. If there is to be a solution here, then, it seems that science itself will not provide it.

One common objection that Russell makes to opponents like Dewey is that they are, in fact, simply missing the deeper question. What they are doing is, in fact, fair for the general practice of science, and possibly even fair for the purposes of their specific philosophical interests, but there is a deeper question here that they are ignoring and that he is interested in even if they are not. Is that a fair characterization under this understanding of the debate? It seems that it may not be. While Dewey and others claim that if Russell accepted their view he would see that there is not a problem here, in general they mean that in the sense that if Russell accepted that you can justify inferences by testing expectations, his specific problem of how to get from hard data that requires some private data to the soft data of the things that really exist in the world would simply go away. They would have demonstrated precisely how to justify the inferences to tables that Russell wants without having to accept his contention that one must start from anything that looks like private data to do that. Or, at least, that one does not have to start by presuming that private data is certain and importantly foundational to that justification. However, Russell would be right to charge them with ignoring the question of whether one can justify any move from private data to public data, and thus ignoring whether they can do anything that looks like testing without settling how they justify the link between their private, subjective perceptions and the public world they wish to test against. However, they are not actually ignoring this; what they are actually doing philosophically is rejecting the use of private data at all.

Dewey is fairly explicit about this, but his arguments suffer from the fact that his rejection of private data is tightly tied to his philosophical and roughly pragmatic viewpoints. It is far too easy to dismiss Dewey’s objections to private data as simply being the result of his particular view of inquiry, which results in a bit of nitpicking over what it means to know and if private sense data can really be known. His view of inquiry focuses strongly on action and specifically acting on desires, and since sense data cannot in fact be inquired into by his notion of inquiry, Dewey must deny that it can be known. Since it cannot be inquired on, and since for Dewey only that which can be inquired on can be known, it seems that Dewey’s objections to private particulars of sense data may well be driven by nothing more than a denial that they meet Dewey’s specific standards of knowledge, even as Dewey is forced to admit that they are in some sense data. However, as has been seen, this problem extends beyond Dewey’s specific theoretical commitments. There is, in fact, good reason to doubt that private data can be used publically, because it is in fact private. Since no one can have Russell’s private data, if Russell’s subjective private data is a crucial component of his justifications his justifications will remain specific to Russell as well. All Russell could do is try to get others to have the same private particulars of sense data as he has, and if they do not or do not admit to it then there is no where to go. Ultimately, to get a public justification Russell needs to rely crucially on public data, but there is no indication that he can do that.

By the same token, Russell’s argument extends beyond his specific theoretical commitments as well. One need not accept that it is a proper methodology to take the data of common sense and carve it up into loose categories based on what happens when we critically examine them to understand his objection. One can reject that there is such a thing as hard and soft data and stay entirely at the level of data and still be concerned that any method of justifying our inferences to tables publically seems to rely on at least the assumption that our private data has the right sort of link to the public realm to justify those inferences. When Watson sees the rat, he is seeing a rat, in an experience that is indeed unique to him. Even when he asks others what they see about the rat, he is still having an experience that is unique to him, even if the descriptions he is receiving are indeed about a public event. If Watson cannot justify inferring from the private particulars of sense data to public objects and events, then nothing he does on the public level will be able to fill in that gap.

Can this be said to provide a neutral perspective between the two views? Probably not. Both of the original objections are indeed tightly tied into their specific theories of how to do philosophy; they include their proposed solutions in the description of the problem. As such, this outlining of the problem does not quite capture their specific interests. However, it does, it seems, raise an interesting problem for philosophy in general. If one leans towards Dewey’s view, then one will not see a problem. If one leans towards Russell’s view, then one will see a big problem and make it a life’s goal to solve it. Here, in the not-so-neutral perspective, one may see more than a problem, but instead see an actual debate, a debate that neither Russell nor his opponents were able to actually see from their perspectives, and thus a debate that none of them could participate in even as they claimed to be participating in it. Which should we trust: private particulars of sense data (and a few other things) or justification by verifying against expectations? Is there another option? Can we combine the two in some way? Is this simply just a problem that will never be solved and that no one should worry about? All of these are interesting questions, and questions that are indeed formed by the results of science, results that are themselves maddeningly fickle about what answers they support.

Ultimately, there does seem to be a real philosophical problem here. To start, as Russell does, by arguing that the problem is how to move from hard data to soft data presumes that the hard data is itself really hard data and that it can or should be used to justify the soft data. This is what his opponents and even science itself may deny, or at least see little reason to accept. After all, at least part of Russell’s hard data is private, and private data cannot provide public justifications. Additionally, when one recognizes that the initial data that suggests a need to infer from hard data to soft data is itself of the precise same sort of data that needs justification by inference, it becomes difficult to accept that Russell’s way of inferring out of the problem could really provide the right sort of justification for knowledge of the external world. However, Russell can reply that like it or not private data is what we have, as is evidenced by science. This is not a problem itself for science per se; science has been quite successful ignoring this issue. But this is a deep philosophical problem, one that philosophy ought to be able to solve. To simply move to public justifications without settling how the private and the public link up is to simply ignore the problem, not solve it. Science itself cannot settle this question – as it supports both positions – and so it is up to philosophy to settle this dispute. So we can answer the question of whether or not there is a real philosophical problem here with a resounding “Yes”, even if it may not be the problem that either Russell or his opponents thought it was.
Bibliography

Dewey, John. “Existence of the External World as a Problem”. Philosophical Review, v. XXIV, no. 142, July, 1915: 357 – 370.

Dewey, John. “Sense Perception as Knowledge” in Boydston, Jo Ann. John Dewey: The Middle Works, 1899 – 1924, Vol 8: 1915. Carbondale: Southern Illinois University Press, 1979.

Russell, Bertrand. An Outline of Philosophy. London: Routledge, 1993.

Russell, Bertrand. On Our Knowledge of the External World. London: Routledge, 1993.

Russell, Bertrand. Problems of Philosophy. Retrieved from http://www.gutenberg.org/ebooks/5827: 2004.

Russell, Bertrand. “Review: Experimental Logic” in Morganbesser, Sidney (ed). Dewey & His Critics. London: Journal of Philosophy Inc, 1977.