Editors Note: The following new entry by Carlotta Pavese replaces theformer entryon this topic by the previous author.
In introductory classes to epistemology, we are taught to distinguish between three different kinds of knowledge. The first kind isacquaintance knowledge: we know our mothers, our friends, our pets, etc., by being acquainted with them. The second kind is knowledge of facts, propositional knowledge, orknowledge-that: this is the sort of knowledge we acquire when we learn that, say, Ithaca is in New York State or that Turin is located in Italy. It is customary to add to the list athirdkind of knowledge that is supposed to be distinct both from acquaintance knowledge and from propositional knowledge. One possesses this knowledge when one can be truly described asknowing howto do something: play the piano, make a pie, walk, speak, create, build, and so on.
The distinction between knowledge-how and knowledge-that was brought to scrutiny in analytic philosophy by Ryle in his seminalThe Concept of Mind(1949), where he raised some of the now classical objections to the so-called intellectualist legend: the view that knowledge-how amounts to knowledge-that. Ryle instead advocated an anti-intellectualist view of knowledge-how according to which knowledge-how and knowledge-that are distinct kinds of knowledge, and manifestations of knowledge-how are not necessarily manifestations of knowledge-that. This anti-intellectualism has been the received view among philosophers for a long time. Even psychologists and neuroscientists have explicitly appealed to Ryles classical distinction when discussing their empirical findings (e.g., Cohen & Squire 1980; Anderson 1983). Nevertheless, in the last twenty years, a renewed interest by epistemologists in the nature of knowledge-how has brought new life to the debate, where new versions of intellectualism and anti-intellectualism have been developed and argued for. The debate is partly epistemological: is knowledge-how an altogether distinct kind of knowledge, different from knowledge-that? But it is also about a psychological question: what kind of psychological state is knowledge-how? The goal of this entry is to overview the debate between intellectualists and anti-intellectualists, while highlighting the implications of this debate for related questions concerning intelligence, cognition, language, and skills.
This entry starts by looking at some classical arguments against intellectualism about knowledge-how: the regress argument (section 1), the insufficiency argument (section 2), and the gradability argument (section 3). Then two motivating arguments for intellectualism are considered: the linguistic argument (section 4) and the action theory argument (section 5). Section 6 overviews the recent epistemological debate on whether knowledge-how and propositional knowledge have the same epistemic profile. Section 7 discusses the cognitive science argument against intellectualism. Section 8 surveys what forms anti-intellectualism about knowledge-how has taken in the recent literature. Section 9 looks at the relation between knowledge-how and skills. Section 10 discusses knowledge-how and other related topics.
1.3 A Revival of the Regress Argument
4.1 The Details of the Intellectualist Proposal
5. The Action Theory Argument and the Question of Joint Action
6. The Epistemology of Knowledge-How
6.3 Knowledge How, Defeasibility, and Testimony
7. The Argument from Cognitive Science
7.4 Knowledge-How in Preverbal Children and Nonhuman Animals
8. Varieties of Anti-Intellectualism
8.2 Ability Based Anti-Intellectualism
9.2 Intellectualism and Anti-Intellectualism About Skill
10. Knowledge-How and Other Related Topics
Ryles most famous objection to intellectualist accounts of skills and knowledge-how is that they lead to a vicious regress:
The consideration of propositions is itself an operation the execution of which can be more or less intelligent, less or more stupid. But if, for any operation to be intelligently executed, a prior theoretical operation had first to be performed and performed intelligently, it would be a logical impossibility for anyone ever to break into the circle. (1949: 19)
Intelligent cannot be defined in terms of intellectual or knowing how in terms of knowing that, (1949: 20)
on pain of a vicious regress (see also Ryle 1946: 22). Exactly how to reconstruct Ryles argument is a matter of controversy (Stanley & Williamson 2001; Stanley 2011b; Bengson & Moffett 2011a; Cath 2013; Fantl 2011; Kremer 2020). The next sections discuss different possible ways of understanding the regress challenge and possible responses on behalf of intellectualism.
The contemplation argument assumes forreductiothat for any action to be intelligently executed, a prior theoretical operation of contemplating has to be performed first :
Contemplation premise (CP):In order to employ ones knowledge thatp, one must contemplate the propositionp.
Assume in addition the following definition of intellectualism:
Strong intellectualism (SI): For an action , knowing how to consists in knowing some propositionp.
And assume further that in performing an action , one employs ones knowledge-how to :
Action premise (AP): For an action , if one s, then one employs ones knowledge-how to .
With these premises the regress goes as follows. Suppose that one performs an action :
By AP, one employs ones knowledge-how to .
By SI, one employs the knowledge that
one employs ones knowledge-how to contemplate
By CP, one ought to contemplate another proposition
The contemplation argument aims at showing the falsity of SI, by showing that its truth, together with the truth of AP and CP, triggers an infinite regress. If SI were true, then performing any action would require contemplating an infinite number of propositions of ever-increasing complexity. On the assumption that this cannot be done in a finite amount of time, the argument goes, accepting SI would lead to the clearly absurd conclusion that no agent could ever perform an action within a finite time (see Fantl 2011: 122).
The question is whether AP and CP are plausible premises. Following Ginet (1975), Stanley & Williamson (2001) argue that AP is plausible only if the relevant is anintentionalaction. To use one of Ryles (1949: 33) own examples, if a clumsy person inadvertently tumbles, it does not follow that in doing so, they employ their knowledge-how to tumble. By contrast, the clown employs their knowledge-how when they tumble on purpose. Nevertheless, if we restrict AP to intentional actions, then the regress can be stopped by observing that contemplating a proposition might happen non-intentionally. For example, when I employ my knowledge that there is a red light ahead by applying the brakes, I need not intentionally contemplate the proposition that there is a red light ahead. Correspondingly, if contemplating a proposition can be done non-intentionally, such contemplation is not the kind of action that requires us to know how to perform ittherefore, it does not trigger the restricted AP and the regress is blocked altogether. Some object that the contemplation in this example might be intentional butunconscious(as suggested by No 2005: 282). But it is unclear what reasons there are for thinking that every time one employs ones knowledge, one intentionally contemplates the relevant proposition (Cath 2013: 365366).
The Contemplation Argument also assumes CPi.e., that in order to employ propositional knowledge when acting, one ought to contemplate the relevant proposition. Against CP, Ginet (1975: 7) observes that one might manifest ones knowledge that one can get the door open by turning the knob and pushing it (as well as my knowledge that there is a door there) by performing that operation quite automatically as one leaves the room; and one may do this without formulating (in ones mind or out loud) that proposition or any other relevant proposition. Ginet concludes that Ryles original argument does not teach us that intellectualism about knowledge-how is false but only that knowledge can be acted upon and manifested without requiring any contemplation on the part of the agent. Indeed, some scholars think that this last weaker claim was the only goal of Ryles original argument (Rosefeldt 2004; Sax 2010).
However, CP is not needed in order to trigger a regress. Perhaps the argument can be salvaged by replacing contemplation with a weaker relation. Consider replacing CP with EP:
The Employment Premise (EP):If one employs knowledge thatp, one employs knowledge-how to employ ones knowledge thatp(and ones state of knowledge thatpis distinct from ones state of knowing how to employ ones knowledge thatp). (Cath 2013: 3678)
The regress is triggered as before. Suppose one s:
By AP, one employs ones knowledge-how to .
By SI, that amounts to employing ones knowledge that
By EP, one needs to employ ones knowledge-how to employ ones knowledge that
But employing ones knowledge-how is an action.
By AP, one employs ones knowledge-how about employing our knowledge-how to employ ones knowledge that
By SI, that amounts to employing ones knowledge of
Intellectualists might object to EP in ways similar to how CP was resistedi.e., that not every action requires for its performance the employment of ones knowledge-how: only intentional actions do, as the clown example suggests. According to this line of reply, employing ones propositional knowledge might be more like a reflex in response to stimuli, rather than an action. Further, this version of the regress challenge may be accused of assuming that knowledge-that is behaviorally inert and needs to be intentionally selected or employed in order to be manifested. Yet, intellectualists have independent reasons to resist this picture (Stalnaker 2012). On the other hand, if Ryleans insist that employments of knowledge-that are actions of sort, it seems there is no principled reason why employments of knowledge-how would not be subject to the same requirement. Therefore, it looks like any regress generated for the intellectualist is generated for Ryle as well (Stanley 2011b: 14, 26; though see Fantl 2011 for a possible difference between the regress generated for Ryle and the regress generated for intellectualism).
A variety of actionssay, remembering to check the cars blindspot when reversingcan be intelligent even though they are not intentional. Or one might manifest intelligence throughprocessese.g., by coming to understand a difficult proposition, without them even being actions. If one accepts that intelligent performances, whether intentional or not, are necessarily guided by knowledge-how, one might try to recast the regress argument by replacing AP with IPP (Weatherson 2017):
Intelligent performance premise (IPP):For a performance , if one s intelligently, one manifests ones knowledge-how to .
Now it seems plausible that ones manifestation of propositional knowledge can be intelligent in some cases but not in others. For example, one might manifest ones knowledge intelligently by bringing to bear one maxim that is appropriate instead of any other that is not to the particular situation which the agent faces. By IPP, if ones manifestation of knowledge-that in a particular situation is intelligent, it requires ones manifesting ones knowledge-how. If intellectualism is true, that would in turn require manifesting ones knowledge-that. If this manifesting of propositional knowledge is intelligent too, though unintentional, it requires knowledge-how. And so on. We get an infinite regress if one accepts that manifesting propositional knowledge can be an intelligent performance, also when it is not an intentional action. (For similar lines of argument, see also Fridland 2013, 2015; Lwenstein 2016: 27680; Small 2017: 623).
Intellectualists might respond by distinguishingtwosenses in which a performance can be intelligent and two corresponding senses of manifestation, only one of which gives rise to the regress. First, an intelligent action might manifest ones knowledge-how in the case that it isguidedby this knowledge-how. On this reading, the regress is triggered. But there is also anotherepistemicsense in which an intelligent action manifests knowledge-how as long as itprovides evidencefor that knowledge-how. For example, the rings on a tree provide evidence for the trees age (hence manifest its age in the epistemic sense) but the rings on a tree are not guided by its age. Crucially, the regress does not arise on the epistemic sense of manifestation. Checking the blindspot might be intelligent in this epistemic sense of manifesting providing evidence ofknowledge-how. Yet, this epistemic manifestation itself is not something that qualifies as intelligent or unintelligent.
A less discussed regress that can be found in Ryle (1946: 67) is an adaptation of Lewis Carrolls (1895) regress. Suppose a student understands the premises of an argument and also its conclusion but fails to see that the conclusion follows. In order to help him, the teacher teaches him another propositionPi.e., if the premises are true, then the conclusion is true. The student understands this and yet fails to see how from the premises and the additional premisePthe conclusion follows. A second hypothetical proposition is added to his store, the proposition that if the premises is true, the conclusion is true too. The student still fails to see. And so on. Ryle concludes:
Knowing a rule of inference is not possessing a bit of extra information but being able to perform an intelligent operation. Knowing a rule is knowing how. It is realized in performances which conform to the rule, not in theoretical citations of it. (1946: 7)
One might respond (cf. Stanley 2011b) to this regress challenge that the student does not really understand the premises of an argument bymodus ponens(p, ifpthenq), for that involves grasping the concept of a conditional, and on an inferentialist understanding (Boghossian 1996, 2003), that would dispose one to accept the conclusion of an inference by that rule. Inferentialism about meaning is, however, a controversial doctrine (for several criticisms, see Williamson 2011, 2012). Other replies might be available. Maybe the student does not represent the rule practically (see next section), or she is simply incapable of granting that the rule applies to this case, for that would explain her failure to be appropriately disposed to arrive at the conclusion, given the truth of the premises. (For yet other versions of the regress challenge, see No 2005: 2856 and Hetherington 2006).
The claim that knowledge-how is a kind of knowledge-that encounters an immediate incredulous stare: how could propositional knowledge be sufficient for knowing how to do something? Ryle (1946: 5) himself poses this challenge as a starting point for his argument:
Obviously there is no truth or set of truths of which we could say If only the stupid player had been informed of them, he would be a clever player, or When once he had been apprised of these truths he would play well.
Certainly, one might know all the propositions that are relevant to how to perform a task, and yet fail to know how to perform it: knowledge-that does not seem sufficient for knowledge-how (see also Ryle 1940: 389).
In order to assess this objection, it is helpful to start with a toy intellectualist theory, on which knowledge-how is a matter of knowing, for some way or method to perform a taskw, thatwis in fact a way to perform it. Insection 4, we will see in more detail a linguistic argument for identifying knowledge-how with this sort of propositional knowledge (Stanley & Williamson 2001; Snowdon 2004). How could, the insufficiency objection goes, one know how to perform a task just in virtue of knowing a proposition about a way to perform it? Consider the following counterexample to intellectualism:
Swimming: Suppose I look at a swimmers swimming, and my swimming instructor pointing to the swimmer says to me, That is a way in which you could swim too. I believe my instructor and we may suppose that what she said is in fact true. I may thereby come to know a true answer to the question How could I swim? However, in the relevant sense, I may not have come to know how to swim. If I took a swimming test, I might still fail it. If thrown in the swimming pool, I might still drown. I do not know how to swim in the relevant sense and yet I do know a true proposition about how to swim.
In response to this sort of counterexample, intellectualists often appeal to practical modes of presentation: knowing a proposition observationally or demonstratively is not the same as knowing it practically. Knowledge-how is, at least in part, a matter of representing propositions about tasks and ways of executing tasks in a distinctively practical fashion. For one to know how to swim, in the relevant sense, one must know of a way to swim represented under a distinctive practical mode of presentation, which is essentially different from the observational or demonstrative mode of presentation inSwimming. This kind of practically represented propositional knowledge is what (some) intellectualists call knowledge-how and is what is absent in the example above.
The notion of practical modes of presentation has received several criticisms (Schiffer 2002; Koethe 2002; No 2005; Fantl 2011; Glick 2015), on the ground that it seems excessively obscure or even question begging. Koethe (2002: 327) worries that practical modes of presentation smuggle in an antecedent notion of knowledge-how (though see Fantl 2008: 461 for a response). This widespread skepticism about practical modes of presentation has led some intellectualists to explore ways of responding to the insufficiency objection that do not appeal to practical modes of presentation. For example, Stanley (2011b: 126) considers answering the sufficiency challenge inSwimmingby appealing to the context-sensitivity of the ability modal could. According to Stanley, depending on how the context for the modal is restricted, That is how you could swim could mean either that that is how you can swimgiven your current physical stateor that that is how you could swimafter training. But coming to know that that is how I could swim after training is clearly not enough for me to come to know how to swim now. Instead, the argument goes, what one needs to know is the former proposition:that that is a way to swim given my current physical state.
Yet, it is unclear that even this response works. Consider a variant of the previous scenario, where Mary is a skilled swimmer who is one day affected by memory loss and so forgets how she is able to swim (Glick 2015). Nothing has changed in Marys physical state: she is still able to swim but she just has forgotten how she is able to swim. Suppose she is told, by looking at a recording of her swimming the day before, thatthatis how she can in fact swim given her current physical state. She might come to know how she is in fact able to swim (just like that!). Yet, she would still fail to know how to swim in the relevant sense and still drown if thrown into the pool.
So, practical modes of representation are hard to escape if intellectualism is to be defended against the sufficiency objection. To assuage concerns about the intelligibility of practical modes of presentation, Pavese (2015b) proposes we think of them along the lines of practical senses, which in turn can be modeled after computer programs. Programs determine an output, just like Fregean senses determine a referent; and they are practical in that they break down a task into the smallest parts that the system can execute (the primitive operations of the system as well as into primitive ways of combining those parts) so they ground the ability to perform a complex task in terms of the ability to perform all of its parts. On this view, if one represents a task practically, one represents all of its parts, and the combination of those parts, through instructions that one has the ability to execute. So representing practically a task entails that one has the ability to perform the corresponding task. (For a critical discussion of practical ways of thinking, see Mosdell 2019. Habgood-Coote 2018c argues that the classical generality problem for reliabilism (Feldman 1985; Conee & Feldman 1998) arises for intellectualism.)
The notion of distinctively practical concepts is motivated by work outside the debate on intellectualism about knowledge-how. Other scholars have discussed concepts that are practical in that they dissociate from semantic and observational concepts and play a central role in explaining behavior. Peacocke (1986: 4950) talks of action-based ways of thinking, Israel, Perry,and Tutiya (1993: 534) of executable ideas, and Pacherie (2000, 2006) of action concepts. Mylopoulous and Pacherie (2017) suggest that executable action concepts might be needed to overcome the interface problemthe problem of how cognitive representations (intentions) interact with motor representations (Butterfill & Sinigaglia 2014). Pavese (forthcoming-b) advances an empirical-functional case for practical concepts, arguing that they are needed to explain a distinctive sort of productive reasoning. Yet, other intellectualists argue we can dispense with practical modes of presentation altogether and instead appeal toways of knowingthat are distinctively practical or executive (Waights Hickman 2019; Cath 2020).
Levy (2017) argues that a form of intellectualism that only invokes practical ways of thinking and practical concepts might not be able to explain skillful motor behavior, for motor representations of the sort required for skilled action and posited by cognitive psychologists are non-conceptual. Along similar lines, Fridland (2014, 2017) argues motor control and motor representation cannot be countenanced by Stanley & Williamsons (2001) and Stanleys (2011b) forms of intellectualism. So, more promising forms of intellectualism might have to invoke, in addition to practical ways of thinking,non-conceptual practical representations(Pavese 2019; Krakauer 2020). Just like perceptual concepts are distinguished from non-conceptual perceptual representations, we might distinguish between practical conceptual representations and practical non-conceptual representations. Motor representations would fall under the latter heading. Nonconceptual motor representations also represent practically, as they break down a task in terms of the most basic operations that a system can perform.
Ryle (1949: 46) formulates the argument from gradability thus:
we never speak of a person having partial knowledge of a fact or truth it is proper and normal to speak of a person knowing in part how to do something. Learning how or improving in ability is not like learning that or acquiring information. Truths can be imparted, procedures can only be inculcated, and while inculcation is a gradual process, imparting is relatively sudden.
As Kremer (2020: 102) points out, here Ryle is making two distinguishable points: (i) ascriptions of knowledge-how are gradable, whereas ascriptions of know-that are not; (ii) the gradability of these ascriptions is explained by the fact that knowledge-how must come in degrees, because learning-how brings improvement in knowledge-how. There is no parallel phenomenon in learning-that, and so no need for degrees of knowledge-that. Others have followed Ryle in thinking that the gradability argument shows intellectualism wrong. For example, Bengson and Moffett (2011b) argue that because knowledge-how is gradable, knowledge-how is more similar to acquaintance knowledge, which also comes in degrees (see also Ryle 1949: 46; Wiggins 2012; Santorio 2016; Kremer 2020: 102).
Pavese (2017) distinguishes between two kinds of gradability of knowledge-how ascriptions: one might know how to do something in part or entirely (quantitative gradability) or one might know how to do something better than somebody else (qualitative gradability). Crucially, these two kinds of gradability are also present more generally in other knowledge-wh (knowledge-when, who, why, where) ascriptions, which do seem to reduce to propositional knowledge. For instance, one might know in partwhocame to the party (Lahiri 1991, 2000; Roberts 2009) or know a better answer to that question than somebody else (see also Stanley 2011b: 315). If parts of an answer are propositions, then knowing an answer might still amount to knowledge of all of its parts. Knowing in part an answer would then amount to knowing at least one of the propositions that is part of that answer. Similarly, knowing a better answer amounts to knowing a proposition that better answers the relevant question. If this is true of other knowledge-wh ascriptions, it is certainly plausible that it is true for knowledge-how. One might know how to in part by knowing only certain (propositional) parts of the answer to how does one ? and one might know a better answer to that question than someone else.
This response to the first part of the gradability objection inspires a further response to the second part concerning learning-how. Suppose that knowledge-how is a matter of knowing a practical answer, where a practical answer encompasses a practical representation for a task or a way to (section 2). As we have seen, practically representing requires possessing certain practical capacities and entails certain sorts of abilities. On this picture, one might gradually learn how to perform a task by gradually learning a practical answer to that question, for one requires time and practice to master a practical representation of how to perform the task. Thus, gradual learning may be compatible with the intellectualist picture, if it amounts to gradually coming to learn more parts of a practical answer.
Intellectualism has been motivated on the basis of a linguistic argument concerning knowledge-how ascriptions in English (Vendler 1972; Stanley & Williamson 2001; Snowdon 2004; Stanley 2011b, 2011c). Begin by noticing that (1) is remarkably similar to (2)(3) (finite knowledge wh ascriptions as they embed a complement with a finitival verb) and to (4)(5) (infinitive knowledge-wh ascriptions as they embed a complement with an infinitival verb):
Mary knows where her piano is located in the house.
Mary knows what to do in case of an emergency.
Mary knows where to find an Italian newspaper.
According to the standard syntactic analysis, (2)(5) have an interrogative as complementwhere is her piano located in the house?, who can play the piano?, what to do in case of an emergency? are all interrogatives. Having said this, in broad outline, the linguistic argument for intellectualism has three steps. The first step is to follow the syntactic cues from (1)(5) and identify the logical form of Sknows how to with that of Sknows + interrogativeQ(= how to ). Call this premiseLogical Form. The second step is to accept the orthodox semantics of knowledge-wh ascriptions, according to which in Sknows + interrogativeQ,Qdenotes a question (C. Baker 1968) and according to which Sknows +Q is true just in caseSknows a proposition answering to the question expressed byQ. Call this premiseSemantics for Knowledge-Wh(cf., among many others, Hamblin 1958, 1973; Hintikka 1976; Karttunen 1977; Heim 1994; Groenendijk & Stokhof 1982, 1997; and Higginbotham 1996). Finally, the third step is to extend this semantics to knowledge-how ascriptions, such that knowing how to requires knowing a proposition that answers the question how can one ?
Next section (4.1) looks in some more detail to the intellectualist analysis of the truth conditions for knowledge-how ascriptions. The section after next (4.2) discusses several objections to the linguistic argument.
The linguistic argument concludes that Intellectualism is true:
Intellectualism about knowing howSknows how to just in caseSknows a proposition answering the question how to .
But what is the proposition that one knows by knowing how to ?
First, note that the subject of the infinitival construction (How to ), or PRO, can either be interpretedde se(de sePRO) or generically (generic PRO). According to the first interpretation, that an agent knows how to perform a ski stunt requires their knowing how to perform a ski stuntthemselves. According to the latter interpretation, it requires knowing howone(as a generic agent or any other agent) would perform a ski stunt. When it comes to ascriptions of knowledge-how, we care aboutde se, and not generic, readings of knowing how. If an agent knows how to in the relevant sense, they know how to themselves.
Secondly, infinitival interrogatives such as how to and what to are ambiguous between a deontic reading (how to =how one should;what to =what one should do) and an ability reading (how to =how one could;what to =what one could do). The deontic reading does not seem relevant when we ascribe knowledge-how. Hence the relevant reading must be an ability reading. Joining these two disambiguations, the truth conditions of knowledge-how ascriptions are (cf. Schroeder 2012):
(Truth conditions) Sknows how to is true just in caseSknows a proposition answering the question How could they themselves ?
Now, what counts as an answer to the question? Linguists distinguish between different kinds of answers that one might give to a question. Anexhaustive answerto How couldS? would specify all the