Ellen Fridland (King’s College London)
(Please do not cite without author’s consent)
One prominent feature of skill is the way in which it is acquired. Skills are learned through practice and not through deliberation, reflection, or memorization alone. Philosophers have appreciated this fact about skill for ages[1] but despite a general recognition, no philosophical account of the exact connection between skill and practice has ever been forwarded. To make clear what I mean, compare the following questions: one might think that when we ask about the connection between practice and skill, we are asking a question about final causes. As such, when we ask, “what is learned through practice?” the obvious answer would be: “skill”! However, we might take the question to be of a different sort, to be a question about formal causes and, thus, to be something like the following: “how does practice change our behaviors such that they go from being awkward, unskilled actions to elegant, skilled performances?” If that’s the question we are asking then what we want to know is how practice changes or impacts our behaviors. That is, we don’t want to know what practice is for but we want to know what practice does.[2] It is this latter question that I will explore in this paper. Importantly, once the answer to that question becomes clear, we will be in a position to see why skilled performances, though automatic in many ways, cannot be thought of as mindless, brute or unintelligent. Rather, skilled actions are cognitive and minded almost all the way down.
This paper will proceed in four sections: in the first section I will defend the tight connection between practice and skill and then go on to make precise how we ought to construe the concept of practice. In the second section, I will suggest that practice contributes to skill by structuring and automatizing the motor routines constitutive of skilled actions. I will cite how this fact about skilled action has misled many philosophers to conclude that skills are mindless or bodily. In the third section of the paper, I will challenge this common misconception about automaticity by appealing to empirical evidence of motor chunking. This evidence reveals that there are two opposing processes involved in the automaticity of skilled action: one process that is largely associative, which I will call “concatenation” and the second, a controlled cognitive process, that I will call “segmentation”. As a result of this evidence, we will be in a position to see clearly why skills are minded and intelligent not merely during their acquisition and not simply in virtue of their connection to intentional states, but, rather, in their very nature. I will end by reflecting on some theoretical reasons for why this is exactly what we should expect to be the case when it comes to skilled action.
1. Practice and Skill
It’s no secret that when we set out to learn a new skill, the way in which we learn that skill is through practice. After all, we learn to dance the ballet by dancing ballet and we learn to play the piano by playing the piano. Of course, Aristotle made this point in the Nichomachean Ethics ages ago. Aristotle writes:
What we need to learn to do, we learn by doing; for example, we become builders by building, and lyre players by playing the lyre. So too we become just by doing just actions, temperate by doing temperate actions and courageous by courageous actions (NE 1033 a 32-b2).
This way of acquiring skills, of course, is to be contrasted with other methods of acquiring knowledge or expertise, such as through deliberation, memorization, or reflection. Notably, skills are special in that we learn them by performing them and not just by thinking about them. This is not to say that we don’t also think a lot about the skills that we are learning to perform. It is only to say that, in addition to thinking, actual doing is required for skill acquisition.
It is, of course, also a familiar fact about skills that people often continue to practice their skills long after they have been acquired. That is, we humans have the strange habit of practicing a skill long after it is probable that we will be able to perform it to a reasonable degree of success in a reasonably large number of circumstances.[3] Just think, professional tennis players like Serena Williams and Roger Federer still practice on court for hours a day almost every day. And tennis is not an exceptional sport in this regard. When it comes to skill, practice is involved not only in skill acquisition but in the ongoing refinement of skill as well.
The connection between practice and skill has been widely appreciated by both philosophers and psychologists. So, for instance, when distinguishing habit from skill, that is, when distinguishing unintelligent, automatic behaviors from intelligent, fluid actions, Gilbert Ryle (1949) focuses on learning. He writes:
[A] mountaineer walking over ice-covered rocks in a high wind in the dark does not move his limbs by blind habit; he thinks what he is doing, he is ready for emergencies, he economises in effort, he makes tests and experiments; in short he walks with some degree of skill and judgment. If he makes a mistake, he is inclined not to repeat it, and if he finds a new trick effective he is inclined to continue to use it and to improve on it. He is concomitantly walking and teaching himself how to walk in conditions of this sort. It is of the essence of merely habitual practices that one performance is a replica of its predecessors. It is of the essence of intelligent practices that one performance is modified by its predecessors. The agent is still learning. (p.42)
This passage highlights the requirement for attentive, intelligent, deliberate instantiations of behavior in the development of skilled action. What we should notice is that there are different ways of performing actions. And, importantly, simply executing or even repeatedly executing an action is not sufficient for performing it intelligently or for developing from it an intelligent capacity. Rather, an action has to be performed in a particular way in order for it to contribute to that action becoming or continuing to develop as a skill. Specifically, the action must be performed in such a way that the agent intentionally continues to learn while she is performing the action. This kind of engagement, I’ll claim, is central to the characterization of practice. Barbara Montero (forthcoming) has made a similar point. She writes,
As the psychologist K. Anders Ericsson (1993) has documented, those who excel in a wide variety of fields have not only engaged in ten years of practice (this is a reference to the ten year rule that was first forwarded as definitive of expertise in 1899 by Bryan and Hartner), but have engaged in ten years of deliberate practice—that is, practice that involves not only doing the actions over and over again, as might be true of our daily activities like buttoning a shirt or driving to the office, but also involves working on aspects that are difficult and, after practice, analyzing one’s own successes and failures. Beyond this, I would add that in order for someone to count as an expert under my stipulative definition, such individuals engage in ongoing practice (Montero, forthcoming).
As we can see, it is a fairly mainstream in sports psychology to hold that a very particular kind of intellectual engagement is involved in the acquisition and refinement of skilled actions. In fact, it seems that both philosophers and psychologists largely agree that the intention to improve is essential to practice. After all, even those philosophers who characterize the pinnacle of expertise as ultimately mindless or bodily still acknowledge that deliberate, intentional, intelligent, minded activity is involved in practicing and developing skills in the first place. [4] That is, even philosophers who hold that skill, once acquired, is no longer best described as cognitive or intelligent, still accept that conscious, deliberate thought and effort are centrally implicated in practice.
To further precesify the notion of practice, I’d like to add that practice often involves not only the deliberate attempt to improve one’s performance in general (for example, making a basket or landing a cartwheel or swimming faster), but also involves aiming to improve the technique by which one is able to perform the skilled action successfully. That is, during practice, skilled agents often attempt to improve not only their ends but also the means by which those ends are achieved. The point, then, is that practice often involves the means of a task becoming ends in themselves, at least temporarily. [5]
So, for example, a gymnast may practice her hand placement on the beam independently of the way in which that hand placement is related to successfully performing a back-handspring. That is, even though proper hand placement is essential for successfully performing a back-handspring on beam, during practice, the gymnast will practice proper hand placement as an independent task. I maintain that skills, very generally, have this kind of structure and that practicing a skill often involves the means, at least temporarily, becoming ends in themselves.[6]
In this section, I’ve claimed that the intimate connection between skill and practice has been widely appreciated by both philosophers and psychologists. And, further, I’ve claimed that practice is commonly understood to be an explicitly intellectual or intelligent phenomenon. I’ve also maintained that in skill learning the goal of practice is often not only to successfully perform a task, but also to improve or refine the way or means by which the task is performed. I take it that, until now, I have not said anything particularly controversial.
2. Practice and Automaticity
Another claim that I take to be relatively uncontroversial is that actions become automatic as a result of practice. That is, as a result of practice, motor routines go from being slow, error-prone, conscious, deliberate and difficult to control to becoming fast, accurate, and transparent—i.e., not requiring deliberate, conscious attention or effort.[7] This fact about skilled actions is often taken to entail that skills are bodily phenomena and not properly minded or cognitive ones. I take the conceptual move from automatic to mindless to be fundamentally flawed. I will present examples of a number of philosophers making this move below and present empirical evidence of why it is mistaken in section 3.
First, it is worth noting that the idea that skills become automatic as a result of practice is widely accepted by both philosophers and psychologists. Further, it is widely acknowledged that one of the main benefits of automaticity is that it alleviates an agent’s cognitive burden. That is, it is commonly accepted that automatic behaviors do not require (or require much less of) the limited resources of executive-level attention and higher-order thought (e.g., Bo and Seidler (2009); Ericsson et al. (1980). As Beilock and colleagues (2002) write:
Current theories of skill acquisition and automaticity suggest…that well learned skill execution is ‘automated’—controlled by procedural knowledge that requires little online attention & control and operates mainly outside working memory (p. 1211).
Moreover, because automatic motor routines can be executed without (or with relatively little) conscious effort or deliberate attention, the effort and attention that would be necessary for performing those actions can be dedicated to other aspects or features of a skill, performance, or situation. For example, compare the effort required when learning how to use chopsticks to the experience of eating confidently with them. Very generally, once a skill becomes automatic, one can focus one’s energy on higher-order aspects of a performance, such as strategy, aesthetic qualities, or the dinner conversation. Additionally, because the cognitive load of acting non-automatically has been reduced, automatic actions can be executed in parallel with other tasks. Think of driving a car and flipping through radio stations at the same time or walking and texting. In fact, a resistance to dual-task interference is one of the hallmarks of automaticity.[8] Accordingly, as skills become automatic, they can also become more complex.
The fact that skilled actions are in many respects automatic has led philosophers of many different stripes and with varying and often conflicting accounts to conclude, incorrectly, as I’ll argue below, that skills are mindless or bodily. That is, falling into the trap of dichotomous thinking, philosophers often equate being automatic with being unintelligent. As I’ve argued elsewhere:
It is not rare in philosophy and psychology to see theorists fall into dichotomous thinking about mental phenomena. On one side of the dichotomy there are processes that I will label “unintelligent.” These processes are thought to be unconscious, implicit, automatic, unintentional, involuntary, procedural, and non-cognitive. On the other side, there are “intelligent” processes that are conscious, explicit, controlled, intentional, voluntary, declarative, and cognitive. Often, if a process or behavior is characterized by one of the features from either of the above lists, the process or behavior is classified as falling under the category to which the feature belongs. For example, if a process is implicit this is usually considered sufficient for classifying it as “unintelligent” and for assuming that the remaining features that fall under the “unintelligent” grouping will apply to it as well. Accordingly, if a process or behavior is automatic, philosophers often consider it to be unintelligent (Fridland, forthcoming).
In fact, many philosophers have explicitly stated that skilled actions are controlled by the body and not by the mind. So, for instance, even though Fred Dretske’s (1998) view is much more subtle than the following quote suggests, he can still be found saying things like: “[I]n the case of all skilled actions, whether it be tying your shoelaces, playing a musical instrument, or dribbling a basketball – the mind goes elsewhere while the body performs.” And John Searle asserts that “[R]epeated practice enables the body to take over” (1983, p.150). Further, in explicating a very different kind of view, Sutton and colleagues, presenting Dreyfus’s account of skill, writes: “Only when one is involved, and gets a lot of practice, will the body take over and do the rest. There is then neither thinking nor awareness, neither attention nor choice: at this level of fluid performance, “an expert’s skill has become so much a part of him that he need be no more aware of it than he is of his own body” (2011, p. 89).
All of this is meant to indicate (not prove) that there seems to be some consensus among philosophers that automatic behaviors are somehow mindless or mechanical, bodily phenomena.[9] As such, skills, even very impressive skills, because they run automatically, are not considered to be properly classified as cognitive or intelligent events. The idea is that even though during practice, skills are the objects of cognitive control, once a skill is acquired and becomes automatic, it no longer remains cognitive or intelligent. So, whereas during practice, that is, when an agent was a novice, her actions were minded, once she becomes skilled, they become mindless.
In the following section, I will argue that the conceptual move from automatic to unintelligent is mistaken. I will do this by appealing to empirical evidence, which indicates that automatic motor routines are fundamentally minded, cognitive, and intelligent in both their nature and structure.
Before moving on, however, I’d like to direct the reader’s attention to some bizarre implications of the kind of view that most philosophers hold. First off, on the face of it, it should strike as backwards to think that the actions of an unskilled novice are more intelligent than the actions of an expert. After all, doesn’t just the opposite seem to be true? Surely, at the very least, a theory of skill should be committed to the fact that the more skilled an agent, the more controlled, appropriate, flexible, and manipulable her actions are.[10] That is, skills are qualified by features that seem to be fundamental to intelligence, generally, despite the fact that they are automatic.[11], [12] In line with this understanding of skill, does it not also seem that predicates such as “clever” or “brilliant” have more business being ascribed to experts than to the novices? [13] And if this is so, shouldn’t we take this to at least potentially tell us something about the cognitive or intelligent nature of skill?
Finally, even if one is unmoved by these general considerations, I expect that the empirical evidence concerning motor chunking that I will present below will illustrate how automatic motor routines are importantly cognitive in both their nature and structure.
3. Motor Chunking and the Intelligence of Automatic Motor Routines
In the previous section, I claimed that because skilled behaviors are automatic, many philosophers have characterized them as mindless. In this section, by focusing on empirical research elucidating the underlying processes of automaticity, I’m going to make clear why the standard way of thinking about the automatic motor routines constitutive of skilled action is largely wrong. Specifically, by considering the nature of motor chunking, a process that is widely believed to underlie the automation of motor routines, I will highlight how a central part of being automatic remains very much cognitive, controlled, and intelligent. I will close by providing some theoretical reasons for why this is very much as it should be. That is, I will claim that the automatic processes involved in skill exhibit a flexibility, manipulability, and responsiveness to our goals and reasons, which requires that they remain cognitive and controlled. Moreover, I’ll claim that this kind of flexibility and control cannot be achieved if automatic processes become large, fused, mindless motor programs.
To begin, motor chunking is the (unfortunately named) “process that facilitates movement production by combining motor elements into integrated units of behavior” (Wymbs et al. 2012, p. 936). These units or chunks are then stored and executed as unified wholes (Bo and Seidler, 2009; Kennerley et al., Sakai et al., 2003, 2004; Verwey and Eikelboom, 2003; Verwey et al., 2009; Verwey, 1994). The idea is that by combining individual elements into a single motor plan, we have the ability to store, access and execute integrated units in a way that substantially reduces our cognitive burden (Bo, Borza and Siedler, 2009; Ericsson et al., 1980). More precisely, we don’t have to think of and execute each element in a sequence or series individually but, rather, once a routine has become automatic, we can store, access and implement a cluster of elements together. So, for example, whereas we would need to control 6 individual elements in a novel sequence, when those 6 elements are chunked into, e.g., 2 units of 3, we only need to control, access, and execute 2 elements. The same principle should apply to tasks that have several parts but not several individual elements, like riding a bike or doing a cartwheel. That is, this principle should apply not only to sequential learning but to skill learning in general.
Behaviourally, a motor chunk is identified by the execution of a quick succession of action elements preceded by a longer pause (Kennerley et al., 2004; Verwey and Eikelboom, 2003). So, using a simple example, if one had a series of elements such as the following: “A B C A B C” a typical way of chunking those elements would be like this: [ABC] [ABC]. The elements within brackets exhibit a strong connection to one another as evidenced by their execution in quick succession and the brackets and space in between indicates a chunk or unit break as evidenced behaviorally by a pause before the quick instantiation of elements inside the set or unit.
It’s vital to see that the empirical evidence concerning motor chunking indicates this kind of unitization proceeds as a result of two different, likely hierarchical processes (Verwey, 2001; Hikosaka et al. 2002, 1999; Sakai, 2003; Wymbs et al., 2012). That is, motor representations are transformed in different ways at different processing levels. And, for my purposes, it is crucial to notice that the processes are not only different but likely opposing. As Wymbs et al. (2012) describes these processes they are:
- Concatenation: the formation of motor-motor associations between elements or sets of elements.
- Segmentation: parsing of multiple contiguous elements into shorter action sets.
As I discuss each process and its function in facilitating motor chunking, it will become clear why getting clear on the details of chunking challenges our notion of automatic motor routines as mindless or bodily.
3a. Concatenation:
The first kind of process responsible for motor chunking is what I’ll call, following Wymbs et al. (2012), concatenation. This is the process that I think naturally comes to mind when we think about automaticity. Concatenation is best thought of as an associative learning process where individual motor elements are combined into wholes (Verwey, 2001). So, individual elements of a sequence or task are fused or bound together into larger combined units. In brief, individual elements in a series develop stronger connections to the other elements in the sequence through the process of concatenation. Concatenation can be observed behaviorally in the faster execution of a sequence or series after practice.
In order to elucidate the nature of concatenation, I’ll appeal to two distinct kinds of evidence: the first is direct fMRI evidence found during motor sequence learning and the second is evidence from individuals with motor learning disorders. Wymbs et al. (2012), using a cued sequence production task analyzed sequence learning using fMRI. What they found was that during the concatenation of motor elements, the occurrence of which was assessed behaviorally, there was increased activation in the basal ganglia. This result should be unsurprising since the basal ganglia has long been implicated in motor learning, especially in procedural learning and in the production of implicit, automatic, and habitual behaviors (Albin et al., 1989).
This is why it is thought that individuals with Parkinson’s Disease (PD), a disease which involves deficits of dopamine uptake by the basal ganglia, have difficulty executing implicit, automatic actions (Trembley et al., 2010). And this is also why it is thought that individuals with basal ganglia damage, such as that caused by basal ganglia stroke, experience motor sequence learning deficits (Boyd et al., 2009). Both groups fail to automatize motor sequences in the same way as healthy controls. Specifically, for this group of individuals, the normal strengthening of connections between action elements that develops during practice in healthy controls does not seem to take hold and, thus, prevents normal motor chunking. As such, PD patients and those who have experienced basal ganglia stroke experience difficulties acquiring learning new automatic routines and, at times, in executing old ones.
Importantly, for my purposes, neither patients with PD nor those with basal ganglia stroke seem to have any deficits when it comes to explicit knowledge of motor sequences (Boyd et al., 2009). That is, these patients have no more trouble recognizing or reporting the sequences or repeated patterns of sequences involved in a motor task than healthy individuals. As such, we can be fairly certain that the motor learning deficits of PD and basal ganglia stroke patients are not the result of a personal-level or executive-level, cognitive failures to learn or recognize the relevant sequence elements or their pattern of connection.
I think that concatenation is the kind of process that most philosophers and sports psychologists have in mind when they think of the automatization or chunking that occurs when individuals practice motor skills. And since it is this process that they is central to their thinking, it is unsurprising that they go on to characterize automatic motor routines as bodily or non-cognitive.
However, if concatenation was the only process involved in the automatization of motor sequences then we’d expect to see the connection between all elements in a sequence increase to form a large single unit or motor plan. That is, if only concatenation were involved in the unitization of automatic motor routines then we should see an undifferentiated associative increase in the connection between all the motor elements in a motor sequence, at least until the relevant limit on working memory had been reached.[14] So, for example, if concatenation was alone responsible for motor learning, what we’d see if we started with a sequence like “A B C A B C” is a stronger connection between all the elements in the series, such that the entire succession of elements could be executed faster. We could represent such an increase in associative strength as follows: [ABCABC]. However, as I indicated above, in healthy individuals, motor chunking actually looks more like this: [ABC] [ABC].
As such, we can see that it is not the case that in healthy individuals, motor chunking is simply the result of a concatenation process. That is, it is not the case that motor chunking just manifests in a stronger, undifferentiated connection between all action elements equally. Rather, motor chunking manifests in a strong connection between a set of elements in an action sequence, and that set is separated from other sets of very well-integrated elements in a larger series. Importantly, the way that motor elements are arranged into groups often follows a logical pattern, when such a logical pattern exists in the sequence and can be gleaned by the learner. For example, repeating elements are grouped together and there is a break after the completion of the last element in a pattern that is repeated. This grouping, as I’ll discuss below, is not simply the result of the limitations of working memory but, rather, an efficient strategy for structuring the sequence.
All of this to say that concatenation is only one part of the chunking process, which underlies the automatization of motor routines. Concatenation is an associative process where (1) the basal ganglia is critically involved as indicated by direct fMRI evidence and (2) the basal ganglia is an important site of implicit, procedural learning, which has little to do with personal-level executive function, as we know based on evidence of motor learning disorders such as those exhibited in PD and after basal ganglia stroke.
3b. Parsing:
The second process underlying motor chunking is a process that I’ll call, also following Wymbs et al. (2012), parsing. This process is responsible for cutting motor sequences into shorter segments. That is, parsing or segmentation works by breaking down longer sequences into smaller sets or units. Importantly, parsing appears to work in the opposite direction of concatenation, which fuses together individual elements into larger wholes.
Crucial for the purposes of this paper, the process of parsing seems not to be an associative process but an executive or control one (Pammi et al., 2012; Kennerley et al., 2004 Verwey et al., 2011; Verwey, 2010). This characterization of parsing is supported by two forms of empirical evidence: direct fMRI evidence observed during motor chunking and considerations based on motor learning deficits that are related to ageing and general cognitive decline.
First off, in an fMRI study of motor chunking during a cued sequence reaction task, Wymbs et al. (2012) found increased activity in the dorso-lateral prefrontal cortex in sequences that were easier to segment. That is, in sequences that had more natural, logical joints to parse and thus were more likely to be parsed, Wymbs et al. observed that that the prefrontal cortex was more active. This evidence suggests that the segmentation of sequences or series into units or sets is the result of cognitive or control processes. We can conclude this because the prefrontal cortex is the area of the brain widely acknowledged to be related to executive function. Moreover, upon reflection, this finding seems to make sense since the choice of how to parse a sequence should involve considerations about which ways of parsing it would be the most effective. In particular, these choices should involve consideration of the logical structure of the sequence or series since this will determine how one could best group the sequence such that it can be recalled and executed most efficiently.
Another reason to think that parsing involves executive areas is because we see deficits in motor chunking that arise in elderly populations (Verwey et al., 2011; Verwey, 2010). These deficits are thought to be related to overall cognitive decline in ageing. Moreover, these deficits are also related to decreased explicit knowledge, specifically, a decrease in the ability to recognize and report relevant patterns or elements in a sequence (Bo, Borza and Seidler, 2009; Verwey et al., 2011, Verwey, 2010). This is not to say that elderly individuals are incapable of motor learning altogether. In fact, the massively redundant motor system makes it the case that almost no injury or decline results in a complete inability to learn motorically. So, just as patients with basal ganglia stroke are able to improve and learn motor sequences to some extent (Boyd, 2009), elderly patients are also capable of some motor learning (Verwey, 2010). But, upon examination, it seems that what the elderly patients do is to learn associatively—that is, they concatenate exclusively. And so, after practice, elderly individuals show some improvement in their ability to produce motor sequences faster. However, they do not parse or segment sequences in the way that would produce characteristic chunks like those in healthy controls. All of this to say that cognitive decline seems connected to a decrease in the process of segmentation, crucial for producing the motor chunks characteristic of automaticity.
We can visualize parsing in the following way: we have a sequence with, say 6 elements, and that sequence is broken down or segmented into two sets. So, if the sequence is “A B C A B C” parsing would be responsible for segmenting the sequence into [A-B-C] [A-B-C] where the repeating pattern ABC is broken into two sets, each of which repeats the three-element pattern once. The dashes in between the letters are meant to indicate that parsing is not responsible for the increased connection between elements within the set or unit. Rather, it seems that concatenation is the process responsible for fusing those elements together.
Further, what’s important to see is that parsing isn’t just a stage in concatenation. That is, parsed segments aren’t just waiting to be combined into larger motor units as practice continues. Rather, these units are robust. That is, even though their joints become harder to detect as a result of practice, with the introduction of a dual interference task, we see that the segments persist even after extensive practice of over 1000s of practice rounds (Verwey & Eikelboom, 2003). As such, we should conclude that motor routines retain their cognitive structure even after becoming automatic. That is, motor representations do not simply become associative processes once they are automated but, instead, continue to be structured in a controlled way. All of this to say that it is not only in practice that a clear cognitive element is present in skill. In the last section of this essay, I will discuss why this finding should be unsurprising and, in fact, it is exactly what we should expect to be the case if we consider the nature and requirements of skill carefully.
What’s more, we know that the parsing of motor chunks is not simply the result of capacity or load constraints. That is, it’s not that a motor sequence is parsed simply because working memory or the motor buffer can’t handle more elements in the sequence. And we know this because even sequences with as few as 4 elements tend to be parsed, in a robust way, into two sets of two elements. And it is universally accepted that 4 elements do not exceed the capacity limits of working memory (Miller, 1956; see Miyake and Shah, 1999 for an overview).
All of this to say that parsing is an essential part of motor chunking. Parsing is a control or cognitive process as we have seen indicated by both fMRI evidence and evidence from the abnormal motor learning of elderly patients due to cognitive decline. Importantly, parsing is involved in structuring the automatic motor routines according to the internal logical of the sequence and also according to the demands of the task. Further, parsing is not just a stage in concatenation. It is fundamental to the process of chunking.
As such, we must accept that the automatization of motor routines that occurs in skill learning is at least partly cognitive. That is, both in the learning stages of automatizing motor routines and the later stages of executing and sustaining them, cognitive control is involved. From this we can conclude that the nature and structure of automatic motor routines is not simply associative, mindless or bodily, but very much cognitive. This is not to say that these processes remain necessarily conscious or are accessible to introspection, but it is a mistake to think that consciousness is required for cognition or mentality.[15] So, what we see is that the automaticity of motor routines is not simply a brute, bodily phenomenon but importantly controlled by higher-order cognitive functions.
3c. Parsing and Concatenation
Together then, the way that we can visualize chunking is as follows: we begin with a sequence such as “ A B C A B C” and the sequence is concatenated such that the connections between individual elements become stronger. We might visualize that a series becomes strengthened through concatenation such that it looks like this: [ABCABC]. Simultaneously, the series is parsed into smaller sets. One natural way of parsing this series would be: [ABC] [ABC]. The elements inside the sets benefit from being fused together by the concatenation process and this explains why the chunked elements inside the set are executed in fast, quick succession. However, the sets retain robust joints, which explains the pauses before the quick execution of the elements in the set. It is likely the case that at a higher processing level the sets, treated as individual units, are also concatenated such that the pauses in between them become smaller and smaller (Sakai et al., 2003; Verwey, 1996)—but, importantly, the joints still remain structurally robust, as we can see through the addition of a dual-interference task. In the following section, I’ll claim that if we reflect on the nature of skill, it should become clear that the reason that sets retain robust joints is so that they can be controlled, intervened upon and manipulated as is necessary for skilled action.
So, to conclude, we see that automaticity proceeds by two simultaneous processes: on one level we have sequences that are broken down or parsed into smaller units and on another level those units are combined or concatenated into larger representations. And, as we saw above, the parsing likely involves executive or control functions and the concatenation is likely a result of lower level associative processes.
4. Parsing & Concatenation: Theoretical considerations
In this section, I want to turn our attention to some theoretical considerations for why it should be unsurprising that the automatization of motor routines through motor chunking involves both associative and control processes.
Let’s begin with a simple example of skill: bike riding.
The reason for concatenation, I take it, is easy to provide. It’s what everyone naturally assumes about automaticity. That is, for reasons of efficiency of storage and execution, it’s important for motor routines to be combined into larger units so that they require fewer cognitive resources to implement. Concatenation, it seems, is likely responsible for our capacity to fluidly, smoothly, and elegantly combine several tasks together. Concatenation allows us to move from one task to another without having to think explicitly about each consecutive task and its connection to other tasks. For example, concatenation would be responsible for our ability to pedal our feet in constant motion and lean into a curve and shift gears smoothly, and steer all at once. And maybe even to hold a conversation with a fellow rider or to navigate in an unfamiliar city, to boot.
The reason for parsing, however, I think is rather different. As opposed to reducing cognitive load, it seems to me that the parsing or segmentation of a sequence is required for the flexibility, manipulability, and capacity for easy adjustments and varied responses that is at the heart of skillful activity. For instance, to return to our example, the parsing of action parts is necessary to retain the ability, to, for example, slow down one’s pedal strokes to the appropriate degree, let’s say, when one notices that there is traffic on the road ahead or to lean into a curve, but maybe a little less than usual, because of another vehicle in the next lane. Parsing may also be responsible for one’s ability to slow down to take a look at a new store front or to focus on keeping one’s feet parallel to the ground, when trying to improve one’s cycling posture. The idea is that keeping action joints robust allows us to make the relevant adjustments and interventions in a performance based on our specific goals and strategies. That is, if all action elements were fused together into wholes without joints then one would lose the easy ability to intervene on the skill as it is unfolding in the various relevant and context-sensitive ways that are characteristic of skill. In short, is the existence of joints that I am claiming allows for the flexibility of skill.
We might take these considerations about what is required for skilled action even further. Rather than starting with an individuated set of distinct elements, as is almost always done in the lab, we might think that what often happens with skill learning in the wild is that we start with a more or less undifferentiated movement—one that, through trial and error, we have gotten to achieve task success in some context or other—and then we break down that movement, that is, we parse or segment that movement into smaller parts. Those parts then become elements that we are able to control, manipulate, and adjust in a more and more flexible and controlled manner. That is, through practice, we insert joints into our skilled motor routines.
And it seems that if we look at skill acquisition and skill refinement from a developmental perspective, this is exactly what happens. That is, we just so happen to stumble on a successful solution to a problem and then, by performing the action over and over and over again, we develop the capacity to control the relevant temporal and spatial parameters of the action in a more and more fine-grained way. This control is necessary for successfully instantiating the task in multiple different but similar contexts.
For example, we can think of something as simple as learning to use a fork. At first, we can see children struggle to get the fork to their mouths without dropping their food along the way. Even if the food stays on the fork, they might not be able to get it directly into their mouths but instead they bring the fork to their chin, cheek, or nose instead. But, as is well known, with time and practice, children turn into experts. They figure out how to balance different kinds of foods on their forks and to smoothly get the food directly into their mouths in ways that reflect a sensitivity to the different kinds of foods on the fork, the different size of bites, their different weights, etc. My suggestion is that we should conceptualize this process as follows: instead of learning each part of the fork to mouth task independently and then combining individual elements together, children seem to begin with one more or less undifferentiated movement and then come to break that large, clumsy movement down into a movement with smaller and smaller sections that they can then adjust and manipulate flexibly as needed. That is, they develop control by instituting action joints at the relevant points of an action.
If this is the case, then our conception of the chunking process would go something like this: we start with a more or less large, undifferentiated, and thus cumbersome, clumsy movement and then we break that movement into segments—into smaller parts that we can exert control over. The process of concatenation would continue to be required in order to ensure that those segments are connected in such a way as to retain the flow, fluidity and smoothness that is characteristic of skilled action and, further, that when particular manipulation isn’t required, the action can unfold in a straightforward way without intervention and without the need for attention or higher-order thought.
What’s important for my purposes is to see that when we frame things in this way, even after acquisition, automatic motor routines remain importantly cognitive. That is, we see that the cognition involved in skill is not simply relevant for the acquisition phase of skill, but is retained in the shape and structure of the automatic motor routines constitutive of skill. It stands to reason that if one continues to refine one’s skills, that is, to practice them with the deliberate aim of improving at a task, then one would anticipate that actions would continue to be broken down into more and more fine-grained parts that are themselves concatenated but which can be executed successfully in more and more variable circumstances. If this were the case, then we would be able to explain why continued practice yields continued and expanded control and flexibility and not simply faster and more accurate but ultimately inflexible movements.
Further, this way of framing things may also give us an explanation of why variability in practice not only produces more efficient learning,[16] but why it is that experts often attempt to reduce the automaticity of their motor routines. As John Sutton writes:
In many distinctive domains, elite practitioners specifically resist the kind of automation which Dreyfus ascribes to the highest levels of expertise, worrying that trusting the body alone to take over will lead to arrested development. Just as they challenge themselves constantly and deliberately in practice, they know that in performance they will be constantly opened up to new limits. As Rietveld argues, “every situation contains perturbing influences”, with new affective influences always potentially altering our evaluations of significance. So expert performers precisely “counteract automaticity”, because it limits their ability to make specific adjustments on the fly (2011, p. 95).
This would explain why practicing in a variety of circumstances and focusing on different aspects of one’s technique at different times is so important for developing expertise. That is, if we think of segmentation as that process by which we achieve control over our motor skills—and we think of variability as the way in which we promote parsing or segmentation, or resist concatenation—then we see why variability is so important for skill. One way of putting this would be as follows: variability in practice is required to keep our skills from becoming habits.
Taken together, all of this should help us to see that the mindless conception of automatic motor routines is simply misguided. Rather, we should hold that cognition is involved both during the practice stage of skill acquisition and, also, guides the representation and implementation of automatic motor routines, after a skill is acquired. That is, cognitive control dictates the structure of our motor representations and our ability to manipulate and control those representations during skill refinement and execution.
So, it isn’t simply that when it comes to skill, cognitive control is present at the level of strategy or high level planning, or at the level of, say, aesthetic expressiveness or even simply during the practice leading up to skill acquisition—but that intelligence is present in the actual representation and execution of automatic motor routines. That is, cognition has to be present in the automatic execution of skills if we are going to have control that is flexible, manipulable and responsive to our goals, reasons, and wills.[17]
5. In defence of control
I have offered an explanation of parsing in terms of the control of skilled action. Specifically, I’ve suggested that skilled action instantiations needs robust joints or segments to remain flexible and manipulable.[18] However, this story is at odds with a popular, competing explanation. Verwey (2001), though agreeing that motor chunking occurs at two processing levels, interprets these two levels differently from the way I do. According to Verwey (2001), “a motor processor rapidly executes the tightly coupled elements within each chunk, and the cognitive processor prepares each chunk for the motor processor. In this case, the pauses are due to planning at a supraordinate cognitive level.” (Wymbs et al., p.13). Specifically, Verwey claims that in a familiar, overlearned movement, “ the cognitive processor…selects a single representation –a motor chunk—for the entire sequence which is subsequently read and executed by a dedicated motor processor” (p.71). As such, instead of being involved in the control and manipulation of the action, as I have suggested, according to Verwey, cognition is required for planning, which is accomplished through the representation of a familiar movement as a cluster or whole.
However, the empirical evidence does not support Verwey’s explanation for two reasons: First off, if the cognitive aspect of skill was concerned with planning skilled behaviors by producing higher-order representations of action sequences then we would expect to see cognitive areas at work in concatenation. That is, if planning were responsible for combining action segments together, as concatenation seems to do, and if planning is a higher-order cognitive process then the fMRI evidence should show executive areas involved in concatenation of action elements. However, as I have reviewed above, this is not what we see. In fact, we see that concatenation seems largely to be an associative process. As such, it is unlikely that the cognition responsible for motor chunking is concerned with planning through the higher-order representation of actions sequences as wholes.
Moreover, if it were the case that cognitive control was required for planning or preparing the motor segments stored in the motor processor then one would expect pauses in between segments to max out the capacity of working memory. However, this also seems not to be the case, as we regularly see chunks with as few as 2 elements (Verwey, 1996; Sakai, 2003; Pammi et al. 2012). Since the higher-order representation of action elements could easily represent more than 2 elements at a time, it remains plausible that the pauses in between chunks are not primarily due to planning or preparation. I have suggested above that the breaks in between motor chunks are best explained in terms of control, flexibility, manipulability and the easy adjustment of skills in light of changing goals and reasons. I maintain that such an explanation is able to do justice to the empirical evidence regarding motor chunking.
6. Conclusion:
To review, in the previous sections, I have argued that practice and skill are intimately related such that practice is responsible for automatizing our skilled actions. I have also argued that the fact that skilled actions are automatic has led many philosophers to misconstrue skill as mindless or bodily. In order to remedy this misconception, I have appealed to empirical evidence of motor chunking. Importantly, this evidence reveals that there are two opposing processes involved in the automaticity of skilled action: one process that is largely associative and another that is cognitive and controlled. This evidence demonstrates that skills are minded and intelligent in both their nature and structure. I have ended by presenting some theoretical reasons for why these findings are exactly what we should expect, if we reflect seriously on the character of skilled actions.
Works Cited:
Albin, R., Young, A., Penney, J. (1989). The functional anatomy of basal ganglia disorders. Trends in Neuroscience, 12 (10), 366–375.
Aristotle, In Ross, W. D. (1981). Aristotle’s Metaphysics. Oxford: Clarendon Press.
Aristotle, In Ross, W. D., & Brown, L. (2009). The Nicomachean ethics. Oxford: Oxford University Press.
Aristotle, In Ross, W. D. (1936). Aristotle’s Physics. Oxford: Clarendon Press.
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review, 89(4), 369.
Beilock, S. L., Wierenga, S. A., & Carr, T. H. (2002). Expertise, attention, and memory in sensorimotor skill execution: Impact of novel task constraints on dual-task performance and episodic memory. The Quarterly Journal of Experimental Psychology, 55(4):1211-40.
Bo, J., & Seidler, R. D. (2009). Visuospatial working memory capacity predicts the organization of acquired explicit motor sequences. Journal of neurophysiology, 101(6), 3116-3125.
Bo, J., Borza, V., & Seidler, R. D. (2009). Age-related declines in visuospatial working memory correlate with deficits in explicit motor sequence learning. Journal of Neurophysiology, 102(5), 2744-2754.
Boyd, L. A., Edwards, J. D., Siengsukon, C. S., Vidoni, E. D., Wessel, B. D., & Linsdell, M. A. (2009). Motor sequence chunking is impaired by basal ganglia stroke. Neurobiology of Learning and Memory, 92(1), 35-44.
Bryan, W.L., & Harter, N. (1899). Studies on the telegraphic language: The acquisition of a hierarchy of habits. Psychological Review, 6, 345-375.
Call, J., Carpenter, M., & Tomasello, M. (2005). Copying results and copying actions in the process of social learning: Chimpanzees (Pan troglodytes) and human children (Homo sapiens). Animal Cognition, 8, 151–163.
Chalmers, D. (1995). Facing Up to the Problem of Consciousness. Journal of Consciousness Studies, 2 (3), 200-19.
Dreyfus, S. E., & Dreyfus, H. L. (1980). A five-stage model of the mental activities involved in directed skill acquisition. California: University of Berkeley Operations Research Center.
Dretske, F. (1998). Where is the mind when the body performs? Stanford Humanities Review, 6 (2). Available at: https://web.stanford.edu/group/SHR/6-2/html/dretske.html
Dreyfus, H., & Dreyfus, S. (1986). Mind over machine. New York: The Free Press.
Ericsson, K., Chase, W., & Faloon, S. (1980). Acquisition of a memory skill. Science, 208,1181–1182.
Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological review, 100(3), 363.
Fridland, E. (2012). Knowledge-How: Problems and Considerations. European Journal of Philosophy. doi:10.1111/ejop.12000.
Fridland (2013). Imitation, Skill Learning, and Conceptual Thought: An Embodied, Developmental Approach. In L. Swan (Ed.), The Origins of Mind: Springer Book Series in Biosemantics. Springer, 203-224.
Fridland (2014). Skill Learning and Conceptual Thought: Making Our Way Through the Wilderness. In B. Bashour and H. Muller (Eds.), Philosophical Naturalism and its Implications. Routledge, 77-100.
Fridland, E. & Moore, R. (2014). Imitation Reconsidered. Philosophical Psychology. DOI:10.1080/09515089.2014.942896
Fridland, E (2015). Learning the way to Intelligence: Reflections on Dennett and Appropriateness. In F. De Brigard and Carlos Munoz-Suarez (Eds.), Content and Consciousness Revisited. Springer, 143-161.
Fridland, E. (forthcoming). Automatically Minded. Synthese.
Hall, K. G., & Magill, R. A. (1995). Variability of practice and contextual interference in motor skill learning. Journal of motor behavior, 27(4), 299-309.
Hikosaka, O., Nakahara, H., Rand, M. K., Sakai, K., Lu, X., Nakamura, K., & Doya, K. (1999). Parallel neural networks for learning sequential procedures. Trends in neurosciences, 22(10), 464-471.
Hikosaka, O., Nakamura, K., Sakai, K., & Nakahara, H. (2002). Central mechanisms of motor skill learning. Current opinion in neurobiology, 12(2), 217-222.
Horner, V., & Whiten, A. (2005). Causal knowledge and imitation/emulation switching in chimpanzees (Pan troglodytes) and children (Homo sapiens). Animal Cognition, 8, 164–181.
Kennerley, S. W., Sakai, K., & Rushworth, M. F. S. (2004). Organization of action sequences and the role of the pre-SMA. Journal of Neurophysiology, 91(2), 978-993.
Lee, T. D., Magill, R. A., & Weeks, D. J. (1985). Influence of practice schedule on testing schema theory predictions in adults. Journal of motor behavior, 17(3), 283-299.
Levy, N. (2015). Neither fish nor fowl: Implicit attitudes as patchy endorsements. Noûs, 49, 800–823.
Logan, G. D. (1979). On the use of concurrent memory load to measure attention and automaticity. Journal of Experimental Psychology: Human Perception and Performance, 5, 189–207.
Logan, G. D. (1985). Skill and automaticity: Relations, implications, and future directions. Canadian Journal of Psychology/Revue canadienne de psychologie, 39(2), 367.
Mandelbaum, E. (2015). Attitude, inference, association: On the propositional structure of implicit bias. Noûs. doi:10.1111/nous.12089.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97.
Millikan, R. G. (2000). On clear and confused ideas: An essay about substance concepts. Cambridge: Cambridge University Press.
Miyake, A. & Shah, P. (1999). Models of Working Memory: Mechanisms of Active Maintenance and Executive Control. Cambridge: Cambridge University Press.
Montero, B. (2016). Thought in Action: Expertise and the Conscious Mind. Oxford: Oxford University Press.
Moxley, S. E. (1979). Schema: The variability of practice hypothesis. Journal of Motor Behavior, 11(1), 65-70.
Over, H., & Carpenter, M. (2012). Putting the social into social learning: Explaining both selectivity and fidelity in children’s copying behavior. Journal of Comparative Psychology, 126(2), 182–192.
Pammi, V. C., Miyapuram, K. P., Samejima, K., Bapi, R. S., & Doya, K. (2012). Changing the structure of complex visuo-motor sequences selectively activates the fronto-parietal network. Neuroimage, 59(2), 1180-1189.
Posner, M. I., & Snyder, C. R. R. (1975a). Attention and cognitive control. In R. L. Solso (Ed.), Information processing and cognition: The Loyola symposium. Erlbaum: 153–175.
Posner, M. I., & Snyder, C. R. R. (1975b). Facilitation and inhibition in the processing of signals. In P. M. A. Rabbit & S. Dornic (Eds.), Attention and performance V. Academic Press: 669–682.
Rosenthal, D.M. (1986). Two Concepts of Consciousness. Philosophical Studies, 49, 329-59.
Ryle (1949). The Concept of Mind. Chicago: University of Chicago Press.
Sakai, K., Kitaguchi, K., & Hikosaka, O. (2003). Chunking during human visuomotor sequence learning. Experimental Brain Research, 152(2), 229-242.
Schmidt, R. A. (1975). A schema theory of discrete motor skill learning. Psychological review, 82(4), 225.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of practice: Common principles in three paradigms suggest new concepts for training. Psychological science, 3(4), 207-217.
Searle, J. (1983). Intentionality. Cambridge: Cambridge University Press.
Shepherd, J. (2014). The contours of control. Philosophical Studies, 170 (3), 395-411.
Stanley, J. and Williamson, T. (2016). Skill. Nous. doi: 10.1111/nous.12144.
Stichter, M.(2007). Ethical Expertise: The Skill Model of Virtue. Ethical Theory and Moral Practice, 10 (2), 183-194.
Sutton, J., McIlwain, D., Christensen, W., & Geeves, A. (2011). Applying Intelligence to the Reflexes: embodied skills and habits between Dreyfus and Descartes. Journal of the British Society for Phenomenology, 42 (1), 78-103.
Tomasello, M. (1996). Do apes ape? In C. M. Heyes & B. G. Galef (Eds.), Social learning in animals: The roots of culture (pp. 319–346). San Diego, CA: Academic Press.
Tomasello, M. (1999). Emulation learning and cultural learning. Behavioral and Brain Sciences, 21 (5), 703–704.
Tremblay, P., Bedard, M., Levesque, M., Chebli, M., Parent, M., Courtemanche, R., Blanchet, P.J. (2009). Motor sequence learning in primate: role of the D2 receptor in movement chunking during consolidation. Behavior and Brain Research, 198, 231–239.
Verwey, W. B., Abrahamse, E. L., Ruitenberg, M. F., Jiménez, L., & de Kleine, E. (2011). Motor skill learning in the middle-aged: limited development of motor chunks and explicit sequence knowledge. Psychological research, 75(5), 406-422.
Verwey, W.B. (2010). Diminished motor skill development in elderly: indications for limited motor chunk use. Acta Psychologica, 134, 206–214.
Verwey W.B., Abrahamse, E.L., & Jimenez, L. (2009) Segmentation of short key sequences does not spontaneously transfer to other sequences. Human Movement Science, 28:348–361.
Verwey, W.B. & Eikelboom, T. (2003) Evidence for lasting sequence segmentation in the discrete sequence- production task. Journal of Motor Behaviour, 35:171–181.
Verwey, W.B. (2001). Concatenating familiar movement sequences: the versatile cognitive processor. Acta Psychologica, 106:69–95.
Verwey, WB. (1996). Buffer loading and chunking in sequential keypressing. Journal of Experimental Psychology Human Perception and Performance, 22,544–562.
Verwey, W. B. (1994). Evidence for the development of concurrent processing in a sequential keypressing task. Acta Psychologica, 85(3), 245-262.
Notes
[1] For instance, see Matt Stichter’s (2007) defense of Aristotle’s view of skill as “empiricist”, i.e., holding that skills are learned through practice puts Aristotle’s view in line with the accounts of Isocrates and the rhetoricians.
[2] See Aristotle Physics II 3 and Metaphysics V 2 for a discussion of the four causes.
[3] See Millikan (2000), section 4.3, for a discussion of the ways in which abilities can succeed and be improved.
[4] See, for instance, Anderson (1982) and Dreyfus & Dreyfus (1986).
[5] For a defense of this notion of practice and consideration of the significance for understanding human cognition see: Fridland (2013, 2014) and Fridland & Moore (2014).
[6] This is an important point to recognize if we want to tie skill learning to human cognition, generally, and to imitation, in particular. This is because imitation is often thought to have the functional, evolutionary role of allowing humans to learn skills. And what the studies on imitation and especially on overimitation show clearly is that the humans have a particular obsession not only with the goals of a task (that is, humans not only emulate) but with the fine-grained detailed strategy with which a task if performed (i.e., they imitate). For more on this, see, for instance: Call, Carpenter & Tomasello; (2005); Fridland and Moore (2014); Horner & Whiten (2005); Over and Carpenter (2012); Tomasello 1996, 1999.
[7] I don’t take this to be a comprehensive list of features of automaticity. And, in fact, I don’t think such a feature list is possible. I’ve argued so much in Fridland (forthcoming). Rather, I take it that I am pointing to several features of automaticity that are relevant to this discussion.
[8] See, for instance: Logan, 1979; Posner and Snyder, 1975a/1975b.
[9] See Fridland (forthcoming) for several examples of philosophers equating being automatic with being unintelligent.
[10] See Logan (1985) for more on skill, control and automaticity.
[11] See Fridland (2015) for more on the relationship between intelligence and learning, and more specifically, on the relationship between intelligence and flexibility, transferability, manipulability and appropriateness.
[12] See also Levy (2015) and Mandelbaum (2015) for more on the connection between flexibility and intelligence.
[13] See Fridland (2012) for an explicit argument for this view.
[14] There is some controversy over how many elements can be stored in working memory. Almost everyone agrees that at least 3-5 elements can be stored in working memory (e.g., Bo & Siedler, 2009; Pammi et al., 2011) but recent studies suggest that working memory can store as many as 7 -10 (Kennerley et al., 2004; Sakai, 2003) elements and maybe more (Ericsson et al., 1980).
[15] See, for instance, Rosenthal (1986) and Chalmers (1995) for more on mentality and cognitive function without consciousness.
[16] See, for instance: Hall & Magill, 1995; Lee, Magill, & Weeks, 1985; Moxley,1979; Schmidt, 1975; Schmidt & Bjork, 1992 for more or the effects of variability in practice on skill learning.
[17] To put it provocatively: if we are to have dispositions that are going to track knowledge, that is, dispositions that will allow us to know what to do, when and where to do it and with how much, then those dispositions themselves must be structured in an intelligent fashion. This is provocative relative to Stanley and Williamson’s (2015) view of skill.
[18] See Shepherd (2014) for a similar way of conceptualising control as connected to flexible execution.
You write that practice is necessary for skilled performances and use some examples from sports. I was wondering what you make of this recent meta-analysis suggesting that practicing only accounts for about 18% of the variance in sports performance, and just 1% among elite performers.
http://pps.sagepub.com/content/11/3/333.abstract
Hi Wesley,
Thanks for a great question.
A few thoughts:
First, a clarification of my view.
I hold that practice is required for skill acquisition. I maintain that one of the things that practice does is to automate action routines thereby making them more efficient, controlled, and fluid. I also hold that ongoing practice is necessary for skill refinement, as it is required to develop control at different levels, e.g., the strategic level, top-down attentional control, and motor control.
Now, the study you cite claims that deliberate practice alone does not account for variation in skill level. That seems totally plausible and totally compatible with my view. There are lots of things to consider in addition to the sheer number of hours that an agent practices, if one is concerned with skill attainment. At least one thing to consider is what kind of practice the agent engages in. Another is the inherent talent of the agent (how quickly they learn; their strength, etc.). It seems absurd to me that anyone would hold that simply counting the number of hours of practice will allow you to discover the skill level of a performer (though, it is totally conceivable and probably true that, depending on the activity, a certain number of hours of practice are necessary for various levels of skill attainment–and the study doesn’t rule that out).
But the kind of obviously false claim about the number of hours of practice alone determining skill level does have a hold on the thinking of lots of people. The authors are right about that to some extent. Just a few weeks ago, I was watching Simone Biles compete in the Olympics and the BBC commentator, discussing her almost unimaginable power and ability says, “yes, it’s all about practice. she’s in the gym every day doing this routine over and over again”. And I thought: “get out of town!” Practice has something to do with it, but her untouchable ability as a gymnast is clearly not reducible to the number of hours she spends in the gym repeating her routine.
Hi Ellen,
Thanks for a very interesting paper/talk. Can I just ask you to think out loud a bit about parsing? As a prompt, in one place you say:
‘the choice of how to parse a sequence should involve considerations about which ways of parsing it would be the most effective. In particular, these choices should involve consideration of the logical structure of the sequence or series since this will determine how one could best group the sequence such that it can be recalled and executed most efficiently.’
A half-baked question: how much do we know about the parsing process, and how we determine the logical structure of the sequence or series? And a half-baked thought: it might be that experienced success or particularities of one’s action-contexts contribute to particular parsings. If you took two agents and structured their environments/actions such that different ways of parsing the same action-type led to success, would you find the different parsings naturally learned by the agents? (A similar thought: to what extent are our parsings due to features of our environments and particular experiences of failure/success, and to what extent due to features of the action control system?)
How did I miss this? Super, super interesting and really, really important!
What seems to be the case is that when sequences have an obvious and simple structure, agents parse them more or less in the same way. So, if you have: ABCABC, most people will put the break in between the first ABC and the second ABC.
However, when the structure or pattern of the sequence isn’t so clear (in the lab, the tests are almost always of sequences), things seem to be pretty much as you suggest: they vary from agent to agent and I’d guess that the actual parsing has a lot to do with the affordances that are available to each agent, which will vary according to experience, attention, salience, etc. So, I think the answer is that parsing depends both on environment and agent and sometimes the environment determines the agent’s response and other times, when different responses are possible and there is no clear reason to prefer one over another, agent’s structure their responses in different ways.
HI Ellen,
Thanks for a really interesting paper – tonnes of stuff to think about here!
I just want to ask a couple of questions:
The first is about your idea that learning movements proceeds by taking undifferentiated actions, and then parsing them into chunks which are more amenable to control. There certainly do seem to be cases like this — the baby example is awesome — but, i wonder whether there are also cases of learning which proceed via concatenation putting different acts together into a bigger chunk? You might think that part of what’s hard about video games is the difficulty of stringing different keystrokes together to produce special moves that one can produce on demand (I’m thinking of games like street fighter here, where you have to get competent at activating a series of keystokes quickly in order to use special moves). Would it be right to think about learning the movement to activate a particular special move as involving concatenation rather than parsing?
I also wanted to ask about the connection between the Ryle quote in the first section and the material about the exercise of intelligence in chunking of motor routines in the second half of the paper. One thing I think comes out of this quote is the idea that skilled action (as opposed to habit) should be understood as involving concomitantly doing and teaching oneself to do (which I’d think of as working out how to act on the fly). You connect the Ryle idea to practice, and the idea of continuous improvement, which seems to suggest that you think that this kind of self-teaching goes on in practice. But, isn’t the Ryle idea that all skilled action involves this kind of on the fly self-teaching? One question is whether you’re amenable to this thought? Then there’s a follow up issue, about whether we should understand the intelligence in manipulating motor chunking that you describe in the second half of the paper as being continuous with self-teaching, or being a distinct kind of phenomenon?
Anyway thanks for the paper, really interesting stuff!
Hello Josh!
Thanks for the excellent questions.
Regarding learning either by building up a set of actions into a combined, fluid sequence or breaking apart a more or less undifferentiated movement into smaller parts, I think both happen and I think that both involve both concatenation and parsing. For the former, the building up kind, that’s the kind that is always tested in a lab. There’s concatenation as the elements in the sequence become more connected and parsing at their “logical/ practical” joints. My guess is that over time, with practice (not just repetition), you’d get both more parsing and to compensate for this you’d also need more concatenation of the parsed segments. For the breaking down of larger, less controlled movements, I think you have parsing and then concatenation of the parsed elements. So, you’d always need both aspects for automation but the order (and maybe the emphasis, conceptually) may differ.
I want to think about the Ryle question a bit more. And I’ll probably check in later to add/amend/take back this preliminary reply. I want to say that I do think of practice as a kind of self-teaching. And I definitely think that most skilled performances are also instances of this kind of self-teaching. Though, perhaps not all. I think that there may be instances of competition where one is too absorbed in action to, in the moment, be involved in self-teaching. So, in that sense, skills would require a kind of ongoing learning (or, as you put it, self-teaching) though not every instantiation of a skill would be an instance of ongoing learning.
As I said, though, let me think about this a little…especially the difference between this and your notion of working out how to act on the fly. Can I ask: on your account is working out how to act on the fly only characteristic of skill or of all intentional action? I’d think the latter and, in that sense, I think the notion might be insufficient to capture the ongoing learning that I’m trying to highlight as characteristic of practice and which is realized in the shape and structure of automatic motor routines.
On breaking vs building, that’s super interesting. I was initially tempted to think of concatenation and parsing as relatively independent processes, but your description here makes it a lot clearer that in actual learning processes, the two are going to be closely connected. I guess this makes sense, as presumably three’s going to be some trial and error in motor learning, and you’ll need concatenation to work your way back from (say) over-enthusiastic parsing.
On the role of self-teaching in skill, the position you’re sketching out here seems like an extremely plausible one. So, most skilled action involves self-teaching, but in certain kinds of demanding action, the self-teaching process shuts down in action (although, presumably one can still treat performance in competition as a learning process if e.g. you can watch video of the performance, or get feedback from a coach).
Awesome question about working out how to act on the fly! This is a super difficult point in the paper, and it’s not something that i’ve fully worked out. There’s definitely a tension between thinking that working out on the fly is some special process that only really skilled people engage in, and thinking that everyone’s doing it all the time, and it’s really banal.
I think that the view I kind of push towards in the paper is that working out how to act on the fly is involved in all intentional action. However, in conversations that i’ve had since i’ve changed my mind some, and i’d probably say that habitual actions (like typing in the lock code on a phone), and basic actions (like raising one’s finger) don’t require working on the fly, although they still involve knowledge-how. But this means that I still think that working out on the fly is involved in basically all other intentional actions, and is not restricted to a special set of ‘skilled’ actions. For example, I’d want to say that if I set out to walk from Kings Cross to Hoxton without first consulting a map, then I’d work out how to get there on the fly (I take it that this isn’t something I’m meant to be skilled at).
I’m not sure why you suggest thinking that most intentional actions involve working out how to act on the fly would stand in the way of using this notion to understand the ongoing learning that’s involved in practice. Surely if working out how to act on the fly is involved in all intentional actions, then we’d expect to find it in skilful intentional actions too?
Hey Josh!
Great to continue the discussion here. I’d put it this way: motor chunking involves two processes, one is concatenation and the other is parsing. Without either process, the automatic motor routines characteristic of skilled actions do not develop.
In terms of working it out on the fly and the connection to intentional action: You have it totally right. If WiOONF applied to all intentional actions (or most) then we’d expect it to be characteristic of skill too. what I meant was that the intelligence of skill wouldn’t be simply a matter of WiOONF.
Just to clarify: the intelligence of skill doesn’t reside in the manipulation of motor chunks but in the shape and structure of the automatic motor routines that result from practice–and those routines are formed according to the principles of motor chunking (parsing and concatenation).
Being a bit sloppy there: by ‘manipulation of motor chunks’ I was meaning to pick out the process of changing motor routines by parsing and concatenating them through practice. I wasn’t meaning to suggest that you think that intelligence is only found in the implementation of motor routines. (This is all new terminology to me, so I’m sure i’m messing this up)
Just checking that i’m understanding your view correctly. The following passage claims that intelligence is involved in the representation of of automatic motor routines. I’m tempted to understand this claim as equivalent to the claim that the process of reforming motor routines through parsing and concatenation is intelligent. Is that right? (If so, that was the thing I was trying to pick out above).
‘So, it isn’t simply that when it comes to skill, cognitive control is present at the level of strategy or high level planning, or at the level of, say, aesthetic expressiveness or even simply during the practice leading up to skill acquisition—but that intelligence is present in the actual representation and execution of automatic motor routines. That is, cognition has to be present in the automatic execution of skills if we are going to have control that is flexible, manipulable and responsive to our goals, reasons, and wills.’
Hi Josh, I wish I made this clearer in the paper. Thanks for pushing me to get clear here!
I’d put it this way: the automatisation of embodied skills proceeds via motor chunking. Motor chunking itself is composed of two opposing processes (parsing and concatenation). It seems that at least one of those processes, parsing, is an intelligent process. What I’d like to emphasize is that though motor chunking occurs during motor learning (during the automatisation of motor routines), the structure of the motor representations that develop through this process is robust. As such, automatic motor routines retain their cognitive structure even after the automatisation of a behavior is complete.
Hi Ellen,
Thanks for the great paper, and for your other work on automaticity and skill, which has been a source of inspiration. Sorry I’m writing so late in the session, but I wanted to ask a couple of questions. (Please don’t feel obliged to answer; I just wanted to share them just in case.)
[1] Executive control seems to have at least two roles in parsing:
(i) Executive control is involved in the initial production of parsing during motor skill acquisition.
(ii) Executive control is involved in the adjustment and refinement of parsed motor chunks during skilled performance.
You argue for both (i) and (ii), but I’m not sure about the extent to which (ii) applies. Would you agree that (ii) is a necessary component of skilled performance? Or can skilled performance sometimes take place without online executive control of its structure? I ask because someone may say that executive control is necessary only in novel situations, or when adjustments are necessary, or when the agent seeks to refine her skill… but that a habitual, routine skilled performance wouldn’t need the involvement of executive control.
[2] This takes me to a subsequent question: in part 4 you say that
About habits: are they different from skills in that habitual routines are unintelligent performances? This certainly seems to be in line with Ryle’s quote: “[i]t is of the essence of merely habitual practices that one performance is a replica of its predecessors”. But if we accept this view of habits, couldn’t we say that they are the true representatives of automaticity: pure associative concatenations without intelligent parsings? If so, doesn’t this conception of habit reinstate the dichotomy between intelligent cognition and unintelligent automaticity?
[3] Just looking at the same issue from a different angle: in a reply to Josh you claim that
Would you say that the lack of variability in practice would lead to a dissipation of the motor routine’s structure? This seems to underlie the skill-habit distinction. But it also seems to go against the claim in part 3b, that the structure of motor chunks is not susceptible to dissipation. If the structure of motor chunks cannot be dissipated, then perhaps even habits cannot become fully unintelligent? And if so, maybe skills and habits aren’t so different after all?
Thanks again for the paper!
Hi Juan,
Thanks for some great questions and for a great discussion on the other thread. I’ve been following along but haven’t gotten my thoughts together to comment.
Your questions are triangulating on related issues, so I’ll answer them together.
I think that skills and habits are different and I think that they are different exactly as you suggest because the structure of skills (the segmented joints) continue to be robust over time and practice. This doesn’t just happen by itself but takes effort and deliberate training. My guess is that over time (months? years?) actions that are repeated regularly but not practiced will become more and more concatenated and will lose their (parsed) joints. And, thus, they will become more difficult to control. There is some evidence of increased concatenation with training. The problem is that the empirical work doesn’t track deliberate practice vs. repetition very well and certainly not over sufficient time scales. But, yes, that’s the idea.
The motor representations constitutive of embodied skills will always be characterized by intelligence whether they are performed in novel situations or not–this is because the representations themselves have been structured by cognitive mechanisms Those joints then allow for control over the action that may or may not be relevant for each instantiation of the action. What I’d emphasize, though, is that even if the online adjustment and manipulation of motor elements is not part of a skilled action instantiation (say, the situation is familiar and there isn’t a lot of pressure) that doesn’t mean that the intelligence of the skill is merely in the past, since the execution of the motor representations requires the tokening of mental representations that retain their cognitive structure. It would be fair to say that motor representations are not intelligent if they become fused, associative processes after prolonged repetition. But this is not what I claim happens in skill.
You might think that automatic motor representations become rigid or lose their flexibility over time and thus are no longer intelligent. But my point is that when it comes to skill, this is exactly what skilled performers combat. So, though the intelligence of skilled automatic motor routines is a matter of diachronic processes, the processes are ongoing- it’s not as if there is a a practice phase, a motor chunking phase and then a repetition phase. Through practice, the parsing of action elements is continuously refined. Barbara Montero makes a lot of ongoing practice in her book on expertise and I think she has it exactly right. One doesn’t just gain expertise and get to keep it. It requires constant vigilance.
Thanks, Ellen! This session has been terribly interesting, and I’m sad the time’s running out, but I guess we can always keep exchanging ideas.
Your reply helps me get your view straight. I think your claim that
is a great answer. An ‘intellectualist’ about skilled action can offer this reply to objections that in familiar situations there seems to be no online executive control. I’ll keep thinking about this.
Now, I’m still wondering what you would say about what was perhaps the more metaphysical issue, namely that a strong distinction between skill and habit may reintroduce the dichotomy between intelligent higher-order processes and unintelligent automatic routines. After all, someone could argue that associative behavior is completely unintelligent unless it is accompanied by some online executive control, or it retains an intelligent structure that is, anyway, a residue of previous executive control. What we have, in the end—the argument would go—, is that all intelligence comes from non-automatic processes (i.e. either present or past exercises of top-down executive control), and that if we isolate the truly automatic aspects of performance (i.e. purely associative processes, mere concatenations without parsings) all we have is unintelligent behavior.
This worries me… I do want to say that there is intelligence in automaticity, but I suspect that if we concede a distinction between skill and habit, the point may be lost.
This is related to your question in the other thread, about the relationship between automaticity and control, which I will go on to try and say something intelligent about.
Thanks again for a terrific discussion!