**4. Grounding Knowledge**

By looking to cognitive psychology and cognitive neuroscience we can gain a lot of insight regarding the cognitive underpinnings of knowledge (see, e.g., [38–45]). In particular, we will argue that various influential theories focusing on di fferent levels of analysis provide a multi-level coherent picture. From a first high-level perspective, dual process theory o ffers elucidating input for our discussion. According to this influential theory,<sup>5</sup> cognitive processes consist in Type 1 processes (System 1) and Type 2 processes (System 2). Type 1 processes are fast, reflexive, and non-conscious, whereas Type 2 processes are slow, reflective, and conscious. These di fferent processes are plausibly mapped to di fferent forms of knowledge. In short, knowledge how can plausibly be tentatively mapped to reflexive Type 1 processes and knowledge that can plausibly be tentatively mapped to reflective Type 2 processes (see, e.g., [46,47], see also [48]):

Motor behavior, moreover, can take place without the experience of intentional states or reasoning, not only in the case of true reflexes but also, importantly, in the case of expert performance, such as that of an athlete "in the zone" or a chess master. [49] (p. 34)

On a lower level of analysis, the human memory systems follow the same pattern (see, e.g., [20,43–45]. Tulving's account states that long-term memory consists in three parts working together—procedural, semantic, and episodic memory—where " ... procedural memory entails semantic memory as a specialized subcategory, and ... semantic memory, in turn, entails episodic memory as a specialized subcategory." [43] (pp. 2–3). Type 1 processes can plausibly be mapped to procedural memory involving bottom-up action schemas, abilities, and skills that are possible to learn through trial and error. Type 2 processes can plausibly be mapped to episodic memory involving propositional, factual remembrance, and the experienced first-person point of view. As an intertwined intermediate, semantic memory involves conceptual content and categorization. Using Cohen and Squire's ([20], see also [22]) canonical conceptualization, this division can be classified as non-declarative (or procedural) versus declarative memory:

Declarative memory includes what can be declared or brought to mind as a proposition or an image. ... Non-declarative memory refers to a heterogeneous collection of abilities: motor skills, perceptual skills, and cognitive skills (these abilities and perhaps others are examples of procedural memory); as well as simple classical conditioning, adaptation/eve/effects, pdming, and other instances where experience alters performance independently of providing a basis for the conscious recollection of past events. [22] (p. 171)

On ye<sup>t</sup> a lower level of analysis, procedural memory involves, for example, the basal ganglia, neocortex, cerebellum, striatum, and the premotor- and primary motor cortex (see, e.g., [50]). Semantic memory involves, for example, associative pathways, the prefrontal cortex, the lateral-, ventral- and medial temporal cortex, basal ganglia, and hippocampus (see, e.g., [50]). Episodic memory involves, for example, attentional pathways, the prefrontal, ventral fronto-temporal, medial temporal, retrosplenial, and posterior cingulate cortices, the parahippocampal, angular, middle temporal, the fusiform, and inferior temporal gyrus, as well as the left posterior insula and the hippocampus (see, e.g., [50]).

To use motor output as an example, thoughts, intentions, and knowledge that involve Type 2 processes, episodic memory, the frontal lobe and cortex, and varying layers of smaller and larger homotypical cells. Associations and conceptual knowledge involve semantic memory, the parietal lobe and association cortex, and varying layers of smaller and larger homotypical cells. Whereas knowledge how involves Type 1 processes, procedural memory, the primary motor cortex, cerebellum, and large

<sup>5</sup> We are aware of critical voices (see, e.g., [38,42], see also the discussion below).

agranular cells (see, e.g., [50]). Investigations on multiple levels of analysis, and from di fferent scientific perspectives, thus support a division of knowledge how and knowledge that.

Now, this understanding of memory and knowledge has recently been criticized by De Brigard [51]—although his focus is specifically on the non-declarative (procedural)/declarative view of the memory systems, which he calls the standard model of memory (SMM). We consider many of his points interesting and meriting further investigation that might ultimately lead to a new interpretation of the memory systems. However, note that the SMM account di ffers in certain details from Tulving's account which we have presented above. We therefore believe that De Brigard's critique loses a lot of its bite when this is taken into consideration. Regarding the intellectualist position specifically, we do consider De Brigard's arguments to be underwhelming since he presents a very dogmatic interpretation of what he dubs the "empirical argument." On his interpretation, this line of reasoning demands that "*P2*. Knowledge how is equivalent to (or, at least, exclusively depends on) procedural [non-declarative] memory, whereas knowledge that is equivalent to (or, at least, exclusively depends on) declarative memory." [51] (p. 728). We believe this premise to be a misleading interpretation and would change the phrasings "equivalence" and "exclusive dependence" to something less strong and domineering. For the anti-intellectualist argument, formulations in the line of that "knowledge how is primarily based in (and primarily depends on) procedural memory" would su ffice. This is enough to counter the intellectualist position. Importantly, we also question De Brigard's [51] (p. 728) third premise "*P3*. But the scientific evidence captured by the SMM demonstrates that procedural (non-declarative) and declarative memory are two entirely dissociable and independent systems." As we have briefly described above, Tulving's account of memory posits that long-term memory involves procedural, semantic, and episodic memory, and that the three systems work in parallel. However, importantly, procedural memory is evolutionarily prior to semantic and episodic memory which means that if any reduction is to be made, it ought to go the complete opposite way from how intellectualists see things (see, e.g., [52,53]).

Stanley [3] argues that because episodic and semantic memory are separate forms of declarative memory and knowledge, not all declarative knowledge is propositional (only semantic memory and knowledge is). This is a reasonable, albeit not undisputed, interpretation. However, it does not follow that procedural memory and knowledge should be viewed as being propositional. By gaining propositional knowledge it is possible to gain in skill, but only after using this propositional knowledge to actually perform actions repeatedly. Stanley's (see, e.g., [3], pp. 155–159) characterization of this process vastly downplays the reflexive and unconscious processes involved in acquiring procedural knowledge and vastly exaggerates the role of propositional knowledge in skillful actions. Stanley argues for his interpretation, where (explicit) procedural knowledge should be viewed as propositional, but does not present empirical evidence supporting this interpretation. He rather bases it on his discussions of how people ascribe knowledge how to others. However, empirical evidence instead indicates the non-propositional nature of procedural memory and knowledge, where "[d]eclarative or explicit knowledge is available to consciousness, and can readily be expressed verbally, whereas procedural or implicit knowledge is typically unconscious and can only be expressed verbally with some di fficulty and sometimes not at all" ([23], p. 65; see, e.g., [20,23–25], see also, e.g., [6,16,17,26]). Devitt [16] sums up the situation as follows:

Despite disagreement or uncertainty on many other issues, psychologists speak with one voice on this one. Even production-system theories, which seem to posit representations of processing rules and hence seem to have the most "intellectualized" picture of procedural knowledge, distinguish this knowledge sharply from declarative knowledge. Psychology presents a picture of procedural knowledge as constituted somehow or other by embodied, probably unrepresented rules that are inaccessible to consciousness. It is thus quite di fferent from declarative knowledge which consists of representations that are available to consciousness. ... In sum, a nonpropositional view of knowledge how is not just

philosophical prejudice or even just folk theory: it seems to be entrenched in psychology and cognitive ethology. [16] (pp. 213–215)

In other words, procedural (non-declarative) processes do not involve brain regions and processes linked to language in the same way as declarative memory and knowledge do. Stanley's semantic reading and idiosyncratic stipulation of what procedural and propositional knowledge involves thus runs against empirical evidence and scientific theorizing, and mischaracterizes the number of non-conscious processes involved in procedural knowledge how (see, e.g., [6] (Section 4.2), [15] (p. 287), [17] (Section 2)).

Devitt [16] argues for claims that are similar to those we have presented, pointing out that it is problematic to base one's view of knowledge solely on linguistic arguments<sup>6</sup> concerning ascriptions of knowledge. Devitt [16] (p. 206) further argues that knowledge how does not reduce to knowledge that, but does temper his claim to state that " ... perhaps one kind of knowledge how is knowledge that. But I still want to maintain that another kind is not. There is a common kind of knowledge how that a person can have simply on the basis of having the ability to perform an activity." It would then be wrong to claim that all knowledge how reduces to knowledge that.

We are sympathetic to Devitt's view, but maybe he gives too much slack to the intellectualist position? After all, all knowledge how involves vast amounts of non-conscious processes (see, e.g., [24,55]:

There is now considerable evidence to sugges<sup>t</sup> that the performance of reward-related actions in both rats and humans reflects the interaction of two quite different learning processes, one controlling the acquisition of goal-directed actions, and the other the acquisition of habits. This evidence suggests that, in the goal-directed case, action selection is governed by an association between the response 'representation' and the 'representation' of the outcome engendered by those actions, whereas in the case of habit learning, action selection is controlled through learned stimulus–response (S–R) associations without any associative link to the outcome of those actions. As such, actions under goal-directed control are performed with regard to their consequences, whereas those under habitual control are more reflexive in nature, by virtue of their control by antecedent stimuli rather than their consequences. [56] (p. 49)

The intellectualist thus ignores empirical evidence and scientific theorizing, and is thus, at its core, a problematic methodology (see, e.g., [57]).

It is clear that knowledge how for Stanley is substantially tied to language use. Even though it is not a requirement that the agen<sup>t</sup> in question is able to express in words their knowledge how, it is required that they, in some sense, are both a language-user and have a concept of self. This is so because having knowledge how for Stanley amounts to knowing the answer to a question, and moreover, to apply the acquired knowledge to oneself (realizing that I can Φ). In sum, it seems that Stanley must hold that the concepts of language and self are necessarily involved in having knowledge how. This, by comparison, is not a demand on our view since knowledge how on our account basically rests on procedural memory (see Section 4). The claim that knowing how to do something is conceptual is, or so we argue, potentially problematic or even false. We do believe that the bulk of (cognitive) scientific input strongly indicates that animals are capable of conceptualizing their environment, while not being capable of language, and that Stanley's position is deeply problematic. As we point out, regardless of how one draws the line concerning declarative memory (semantic and/or episodic LTM), it does not

<sup>6</sup> Stanley and Williamson ([5], p. 441) write that their account follows "from basic facts about the syntax and semantics of ascriptions of knowledge how". It springs from a theory of embedded questions: according to which knowing how is treated as on a par with knowing whether, knowing when and so on. We do not offer an extensive analysis of Stanley's linguistic arguments here. Our view, in short, sides with Glick's [54] view, saying that using mere linguistic premises, about knowledge how ascriptions, to support substantive conclusions about the nature of knowledge how just cannot do the job it is supposed to do.

follow that non-declarative memory (procedural LTM), which is most reasonably tied to knowledge how, or so we argue in the paper, is conceptional, and even less so; propositional. In sum, we hold the view that verbalization is not necessary for conceptualization. Moreover, we hold that there are plausible arguments to be had for both positions, but that we want to leave the question open to further debate. Importantly, however, we argue that it is up to Stanley to prove that non-propositional knowledge how is reasonable to tie to propositional knowledge that—which we argue that he does not succeed with. In other words, it strikes us that the burden of proof is on Stanley, where we consider his account to be idiosyncratic and not in line with a naturalistic interpretation of relevant scientific input.
