Skepticism About the Conceptual World
I will describe a startling new form of skepticism, to be set beside more familiar forms. It lurks beneath the surface of recent work on meaning and reference. Consider “water”: it has both a meaning (sense, connotation) and a reference (denotation, extension). Suppose its meaning is equivalent to “tasteless transparent liquid found in lakes and flowing from taps”. These are the properties a typical speaker associates with the word (its “stereotype”). Combining these words with “water” produces an analytic a priori truth. They provide an analysis of the concept we associate with “water”. Yet they are not epistemically necessary: it could turn out that water is none of these things, as it could turn out that water is not H2O. Perhaps we have all been under a giant illusion about these properties of water: by some quirk of our nervous system a yellowish lemony-tasting liquid has appeared to be transparent and tasteless, and what fills lakes and flows from taps is some other liquid than water. These possibilities are not beyond the powers of an evil demon to contrive. We cannot be certain that water has the properties we typically associate with it—mistakes are possible, illusions conceivable. Water might turn out to have none of the properties included in its stereotype, i.e. its meaning or connotation. Yet we would still be referring to water by “water”, whatever water is. The reason this is possible is that we fix the reference of “water” in a certain way, namely by pointing to a sample of a certain natural kind and saying, “Let ‘water’ designate whatever natural kind underlies these appearances”—whether those appearances are veridical or not. We might thus have successfully referred to a certain liquid and yet acquired quite false beliefs about its properties. Reference is independent of opinion: appearances can be inaccurate even as reference succeeds. If we adopt a causal theory of reference, we can say that the reference-establishing causal relations are logically independent of whatever beliefs we form about the extension of the term. What this means is that the sentence “Water is a tasteless transparent liquid found in lakes and flowing from taps” is both analytic and conceivably false. It could be false because water might actually have none of these properties and yet the meaning we assign to the term includes them: they are contained in the connotation but the denotation lacks them. Thus we have an analytic but false statement—it makes a false ascription of properties to thing we refer to.
Now consider skepticism. Hearing about the semantics of natural kind terms the skeptic will seize his chance: he will insist that all our natural kind terms are vulnerable to a skeptical doubt, namely that propositions formed from them are not knowable with certainty, even when analytic. We might be wrong that water is tasteless and transparent, that lightning is bright and precedes thunder, or that gold is shiny and malleable. These are characteristic skeptical claims, but the extra turn of the screw is that analytic truth does nothing to preserve them from skepticism. The semantics of the terms combines demonstrative reference with fallible descriptive stereotype, so that the reference could succeed while the stereotype is inaccurate. Sense (descriptive content) doesn’t determine reference, but it can generate analytic truths nonetheless. Since the descriptive content of sense is possibly erroneous, we can generate fallibly known analytic truths—we can’t be certain that these are truths. Water might turn out not to be what we suppose, however much what we suppose fixes its meaning (one aspect of it at least). In the extreme, water might be a bitter tasting dark-colored solid that has presented a totally misleading appearance to us all these years—so the skeptic will contend, and he is notoriously difficult to thwart. What we have been designating by the term is quite different from the way it appears (if one day the scales fell from our eyes, we would exclaim “So that’s what we’ve been drinking all this time!” while beholding a mud-like substance).
How far can the skeptic push it? Consider knowledge (the word, the concept, and the thing): we customarily suppose that the meaning of “know” includes belief, truth, and justification—that is its sense or connotation. It also refers to a specific mental state of a person. We take it that this reference instantiates the properties contained in the meaning of the term—that it involves belief, truth, and justification. That’s what we mean and it instantiates. But the skeptic wants to know what makes us so sure that the thing we refer to has the properties we ascribe to it: why assume that the nature of the mental state that “know” refers to actually includes the properties we ascribe to it? Why couldn’t it be like the case of water? Suppose we are confronted by a sample of a certain mental kind and we announce, “Let ‘know’ refer to the mental state before us”, while believing that the state in question is an instance of true justified belief—maybe that’s just the way the sample happens to strike us. But suppose that, contrary to our impression, none of this is true: the state in question is unjustified belief in a falsehood, or not even belief at all but disbelief (the sample is being insincere in its assertions). Then the skeptic maintains by citing the semantics of natural kind terms we can say that knowledge is not true justified belief—the state we are referring to is unjustified false belief! Now the question is what we can say to rule this possibility out in our case: might it not turn out that knowledge is not true justified belief at all but false unjustified belief? This is epistemically possible, the skeptic contends, given the way the term “know” was introduced and given the facts of the case. So we should admit that it might turn out that knowledge is not true justified belief, because the term designates something quite different from what we supposed—we had false beliefs about the extension of the term as it actually was at the moment of reference-fixing. But that implies that the analytic truth “Knowledge is true justified belief” might turn out to be false, simply because the state originally designated lacked the properties we thought it had. Our false ideas entered its meaning, but reality failed to confirm these ideas. The proposition might be analytic but false, and the skeptic wants to know what we can say to rule out this possibility. Of course, it is also epistemically possible that knowledge is true justified belief, but the skeptic is asking why we prefer that alternative to his skeptical possibility. We should be agnostic.
Or consider “bachelor” and suppose that the initial sample is quite other than what the introducers of the term think: they think they are confronted with a bunch of unmarried males but in fact they are confronted by a group of married females. These individuals are masquerading as unmarried males while being just the opposite. The fooled introducers then stipulate, “Let the word ‘bachelor’ designate the marital and gender status of thisgroup”. They fix reference to the property of being married and female while mistakenly believing that the group in question is male and unmarried. Then the sentence “Bachelors are unmarried males” is false for these speakers, despite their firm belief in its truth. It may indeed be analytic in their language, but it is still false. And now the skeptic asks how we can rule this out in our own case: couldn’t it turn out that bachelors are married females? Maybe our ancestors introduced the term in the way described and thereby fixed its reference to married females; their beliefs were false of these individuals, but so what? Thus we today refer with “bachelor” to the natural kind of married females, even though we think we refer to married males. Or suppose all the people we have ever met who called themselves “bachelor” and gave every appearance of being male and unmarried were really married women in disguise—wouldn’t that tie down the reference to that group, not the group we thought we were referring to? If this is the way reference works in general, then such skepticism would seem indicated. It might turn out that bachelors are married females! It is not epistemically necessary that bachelors are unmarried males, despite the analyticity of the corresponding sentence. The skeptic thus extends his doubts to knowledge of analytic truths.
Let us make explicit what is going on in this argument. On the one hand, we have the concept, an item in the mind; it contains various components, which fix the set of analytic truths with respect to that concept. On the other hand, we have the extension of the concept, an item in the world; it has a certain objective nature, which fixes its de re essence. We normally suppose there is a correspondence between these two levels: the components of the concept actually capture the objective nature of the thing designated. In the water case it is easy to see how this correspondence could be disrupted, because we can be wrong about the properties of the natural kinds we are referring to. The skeptic then seeks to extend this point to other concepts by adopting the same type of analysis: there is the concept we have of knowledge, and there is the fact of knowledge itself; but the former might not correspond to the latter—knowledge itself might not have the properties the concept ascribes to it. What guarantees that the objective thing has the properties we think it has? It might be like the case of water. Similarly, the concept we have of a triangle implies that triangles have three sides, but the skeptic conjures a scenario in which we introduce the term “triangle” in reference to things that are actually four-sided, thereby referring to such things with the word “triangle”. Then “Triangles are three-sided” would be analytic for us, given our beliefs, but actually false. And the skeptical question is how to demonstrate that our present use of “triangle” is not like this: maybe we actually refer to four-sided figures with the term “triangle”! Might we not one day discover that triangles have four sides, contrary to what we now believe? We might discover that we are brains in vats, and we might discover that we refer to quadrangles with “triangle”. That would be strange, to be sure, but not logically impossible.
How could we respond to the skeptic? Gap closing is the obvious manoeuver: don’t let the concept and the property diverge. Then there will no logical space between what we think and what is. Thus we might identify the property with the concept: for x to have P is just for x to have C (correctly) ascribed to it. But this gives rise to an idealism that destroys objectivity—as is typically the case with this kind of counter to skepticism. Clearly there was water before there was the concept of water, and similarly for knowledge, triangles, and bachelors. At the other extreme we could try going radically externalist and make the concept nothing more than the property: then what is in the mind will not be separate from what is in the object. The trouble is that this will entail that we can’t be misled about the nature of water, or mistaken about what knowledge involves, because our concept will simply bewhatever these things are objectively. A more realistic suggestion is that there is a kind of pre-established harmony between the concept and the property: the constituents of the concept necessarily correspond to the constituents of the fact (the nature of the property). But again, this fails to allow both for error and for incompleteness: our concept may misrepresent the property and it may fail to exhaust its nature. For example, there may be more to knowledge than we think, and our conception of knowledge may be inaccurate in some respects. This is precisely what the skeptic is capitalizing on by pointing out the epistemic possibilities: water may not be as we suppose it to be, no matter how central to our concept a certain feature is; and similarly for other natural kinds. His point is that analytic containment in the concept is no proof against the possibility of such errors about reality. The world contains various phenomena and we are trying to capture them in our concepts, but we may fail; so it might always turn out that things are not as we take them to be. Water might not be transparent, knowledge might not be true, triangles might not have three sides. Of course, if these things have those properties, then it is plausible that they have them necessarily; but the question is whether we know with certainty that they do, and the skeptic finds reason to doubt this. Metaphysical necessity does not imply epistemic necessity.
It might be said that the underlying semantics presupposed by the skeptic applies to semantically simple expressions like proper names and natural kind terms but that not all terms fit this mold. The former terms denote by mechanisms independent of their descriptive content, which forms a separate component of meaning; but terms like “knowledge” and “triangle” and “bachelor” don’t work like that—here the descriptive content is active in fixing reference. Thus the meaning of “know” must imply truth in knowledge itself because that is simply how the term determines its reference—“know” refers by definition to what is believed and true (etc.). It is semantically complex and works as a logical conjunction of conditions, unlike a proper name. This, however, is all very debatable and anyway faces an obvious retort: what about the simple elements that make up the meaning of the term? These will be subject to the same skeptical argument that we started with: maybe “believe” and “true” denote properties other than we suppose because they were introduced under conditions of fallibility. We announce “Let ‘believe’ denote that mental state” in front of a sample, convinced that we are referring to a state of internal assent, but in fact our sample is in a state of suspended assent or outright dissent. We don’t have infallible access to other minds! Same for triangles (so we can’t wriggle out of the problem by appealing to introspective authority): we suppose there are three-sided figures in front of us and we stipulate that “triangle” refers to that geometrical form, pointing at the sample; but in fact we are suffering from a visual illusion and four-sided figures constitute the sample. The skeptic is saying that we can always misrepresent the properties of the sample that forms our semantic anchor, which is why it may turn out that we have actually done so. Analytic containment in the concept is no protection against this possibility. Skepticism about the external world thus generalizes to skepticism about what we regard as definitional. That is to say, we can be wrong about the essence of things as well as about their accidental properties, even when that essence is supposedly contained in our concepts. Since complex concepts resolve into simpler ones, the skeptical problem can always recur at the basic level.
This skeptical problem deserves to be called a skeptical paradox because whether or not I know anything I surely know what it would be to have knowledge—I surely know that I cannot know what I disbelieve or what is false! Similarly, I may not know whether there are any triangles in nature but I surely know what a triangle is—I know it’s not a circle! But the skeptic is denying, startlingly, that I do have such knowledge; his claim is that it is epistemically possible that knowledge is not of truths and doesn’t require belief. We just don’t have that degree of apodictic insight into the nature of the things in question; we merely conjecture that this is the nature of what we refer to. We may be profoundly ignorant of the objective nature of the kinds of which we speak. Philosophers took it for granted that knowledge of analytic truths is free from skeptical doubt, but it turns out that they are swallowed up too. How far can this skepticism go? What about our knowledge of what “red” means, or “plus” or “and” or “ought”: can we conceive of scenarios in which people are radically mistaken about what these terms designate? Could it turn out that genocide falls into the extension of “good”? Could “red” turn out to designate blue? Could “and” mean disjunction? These would be paradoxical results indeed, so any skepticism that implied such things would deserve that label. 
The skepticism I have been expounding doesn’t apply to our knowledge of our concepts as such: we canknow with certainty what our concepts contain. We know with certainty that our concept of water includes being transparent and tasteless, and similarly for our concept of knowledge in relation to truth. The skeptic questions the move from this to our putative knowledge of the reference of our concepts—whether we know that water itself is transparent and tasteless, or that knowledge itself involves truth. What holds of the concept is not the same as what holds of the object it refers to. Thus I can be certain of analytic truths in so far as they concern what is true of my concepts, but I can’t (with certainty) infer from this anything about the essence of what I am referring to with these concepts. Hence (according to the skeptic) water might turn out not to be transparent and knowledge might turn out not to be true and triangles might turn out not to have three sides. 
 Note that the skepticism I am considering does not contend that there is no fact of the matter about what words mean, only that we cannot know such facts. We could be radically mistaken about what words actually do mean.
 I have said nothing here about skepticism concerning rule following, as expounded in Saul Kripke’s Wittgenstein on Rules and Private Language (1982), but that is certainly a useful comparison point for the skepticism presented here (they are not at all the same thing).