Deliberate practice as a way of overcoming limitations of working memory

Draft outline of a theory of deliberate practice by Dominik Lukeš.

Deliberate practice background

Deliberate practice has gained a lot of attention recently as a way of developing and improving expert performance without reference to innate predispositions such as talent. Of course, certain innate as well as environmental predispositions will play a role, particularly when it comes to the differentiation of performance at the very peak. The height of basketball players is often given as example [ref]. However, the key lesson of deliberate practice is that the lack of talent or merely not being naturally "good at" something does not prevent someone from achieving a significant level of competence.

Deliberate practice has been the subject of many popular books, as well as academic treatments [ref] and was summarised by its originator Anders Ericsson in Peak [ref]. However, many of these books and countless articles and blogposts focus on the process. They emphasise things like getting feedback, working with coaches, breaking tasks into smaller components, working in short focused spurts. And that is all good advice. However, it ignores what I believe is the main and most revolutionary contribution of deliberate practice. Namely, redefining the goal of skill acquisition as the building of mental representations in order to overcome the universal limitations of working memory.

Mental representations and chunks

What are mental representations? Most simply mental representations are chunks that can be held in working memory and with which we can perform operations in order to achieve the task. Or in other words, articulate our intent. The idea goes back to GA Miller's seminal paper from 1956 "The Magical Number Seven Plus or Minus 2". In it., Miller talks about what appears to be fundamental limit of working (or in his words "immediate") memory in that it can only hold approximately 7 pieces of information. The precise number has been debated over the years [ref] but the fact that a small finite limit exists is beyond doubt. However, and this is where Miller's and Ericsson's insights merge, is that there does not seem to be any meaningful limit to the size of these pieces. They are not bits of information but rather chunks, or in Ericsson's terms "mental representations".

Both chunks and mental representations are terminologically laden and subject to confusion across various research traditions. I will use them interchangeably simply to mean pieces of mental or physical performance that can be held in working memory in such a way that they can be manipulated without need for further articulation. To give trivial examples. To express the plural of the word "dog", we need to add the suffix "s" which is furthermore pronounced /z/ due to the coarticulation with the preceding consonant. All of those components need to be successfully articulated. Yet, the only demand of saying 'dogs' on working memory is topronounce /dogz/. We could go through the same step by step analysis of an entire sentence such as "The dogs are barking." where not only do we need to produce this articulation for each individual word and also their interaction and that of the sentence as a whole. But all that is required of working memory is to deal with a unitary chunk. The sentence "the dogs are barking".

The same applies to primarily physical tasks such as driving. Shifting gears consists of a number of physical steps but the driver is only minimally competent when she can perform them as one. The process is then incorporated into the process of turning and slowing down which includes changing gears and its many subarticulations but also assessing the situation, turning the wheel, etc.

Mental representations and expertise

What sets apart experts from lay performers is the size of the chunks upon which they can confidently operate. For example a racecar driver, can perform all of the actions involved in shifting gears while turning under conditions of stress. All this at high speed while paying attention to her competitors, instructions from the navigator and while planning manoeuvres, several hundred or even thousand meters hence.

A sports commentator can produce a continuous flow of relevant and coherent speech for extended periods of time while paying attention to what goes on on the field, technical aspects of the broadcasting equipment and instructions from the production team [ref Goffman]. Yet, when these and other experts' working memory is tested, they are not found to exceed the limitations identified by Miller [ref].

It is not only the raw size of the mental representations but also their richness. For instance, mnemonic devices are often used to facilitate recall. What this means that one has to remember more things in order to learn something simpler. Memory champions often focus on creating rich associations to facilitate memory, thus increasing the richness of the chunks they hold in working memory without increasing the demands on memory capacity. [ref - Moonwalking with Einstein]

This also explains why there are limits to the transferrability of expertise between different domains or sometimes even different situations [ref]. The mental representations, in order to be useful for their primary task of reducing the working memory load, must be richly embedded in their environment which in turn also introduces new demands on working memory. For instance, public performances of music require additional practice compared to solitary playing. Equally, any foreign language learner will have experienced the notion of being tongue tied when faced with a real situation, even using language that they acquired perfectly in the classroom [ref Goin in Brown]. Practice therefore needs to be focused on the performance of the skill in real context. We can also use this phenomenon to explain why teaching to the test works and fails to produce actual skill without additional focus.

Deliberate practice as method

Deliberate practice can then be best understood as a method for increasing the size and richness of mental representations or chunks in a particular domain of expertise. However, the realization that this is the goal of practice, and then learning how to identify the next mental representations to develop are just as important as the techniques associated with the process of practice, if not more. Without chunks or mental representations as the clear goal, deliberate practice can devolve into meaningless repetition.

What then is the nature of deliberate practice? First, Ericsson [ref] differentiates between purposeful practice and deliberate practice. However, his use of the terms is inconsistent [ref] and for the sake of simplicity, I will use deliberate practice to include activities that are primarily characteristics of purposeful practice which Ericsson sees as being more effective than pure repetition but less effective than full deliberate practice. However, when trying to apply deliberate practice to areas outside its original scope, these fine distinction would be more detrimental than beneficial. It is important to keep in mind that such differentiation exists when exploring the concept further.

The key characteristic of deliberate practice is focused repetition of performing the component of a skill in question. The success depends on timely feedback and guidance from a coach. The feedback needs to lead to a revision of the performance as well as the reassessment of the desired building blocks of the skill. Deliberate practice is most likely to be successful if the learner follows an established programme of study supported by an experienced coach who exceeds the learner's level of competence.

The role of the coach is important in order to increase effectiveness of the practice in that the coach can provide feedback and suggest ways of correcting performance. This is essential at the peak levels of expert performance where any improvement is going to be small and require extreme focus and effort to achieve. Since the concept of deliberate practice came at least in part out of research of elite performers, the focus on coaching and training programmes is a big part of the literature. However, in order to make deliberate practice accessible in non-elite, less resources contexts, it is essential that alternatives to coaching be identified.

The literature suggests a number of approaches. Peer coaching and feedback has been shown to be particularly effective. However, using automated feedback instruments - for instance, music or language apps - can also serve to provide actionable feedback. The learner can also benefit from recording their performance (music, physical activities, foreign language) or keeping track of measurable aspects of performance where recording is not practicable (number of errors, time, routine maintenance). Finally, significant performance improvements can also be gained simply by following an established routine. Here, the famous example is the Couch to 5K running programmes that can take most beginners from no ability to run to being able to run 5 kilometres in approximately 8 weeks.

There is no one established description of the steps of deliberate practice but for the purposes of further discussion, we can summarise it as follows.

1. Identify performance target and the mental representations or chunks required for performance within the limits of working memory. At some levels, this may only be possible with the help of a coach or a teacher. The performance target also needs to be frequently revised as intermediate stages are reached and the learners ability to perceive their own performance changes.
2. Identify components of mental representations (levels of performance) to provide target for practice. This may include scaffolding, i.e. practicing skills required for the acquisition of the skill rather than the skill itself (for instance acquiring a metalanguage, ability to count time in music, or developing a practice routine). In skills, where performance is not immediately susceptible to measurement, identify measurable aspects of performance or a way of recording progress in order to receive feedback.
3. Practice component chunks in a focused way until their component actions stop requiring individual attention. This requires repetition with close focus on small details of performance. The level of focus required means that this level of practice cannot be sustained for long periods and can only be effective if performed regularly in relatively small chunks of 30-60 minutes. Practice should always be slightly above the current level of skill. Ideally, at the level where the learner can perceive the target performance. The actual practice techniques will vary by the target performance, the current skill level and interest of the individual. In most areas, there are likely to be many different practice techniques that can lead to successful acquisition of the required mental representations.
4. Measure progress regularly by reviewing when multistep-actions become single performance chunks. This step may require some creativity in areas not naturally susceptible to measurement. Some areas, such as sport and musical performance have established assessment metrics. In sport, aspects of performance such as speed, distance or weight are not only inherently measurable but also can form the target for improvement. However, in most areas any measurement will only be a proxy for progress. It is essential to avoid the trap of confusing the measure with the target performance. The only true measure of progress is the ability to perform operations on the acquired mental representations in working memory as single chunks. Where this is not susceptible to external observation or reflected in a measurement proxy, introspection will be required of the learner.
5. Review target mental representations and identify non-performance scaffolding chunks. One of the most common mistakes is to confuse the performance target with certain stages required to achieve it. For instance, acquiring metalanguage or other metacognitive apparatus is necessary to be able to effectively monitor progress or identify areas to focus on. However, that is rarely part of the performance itself. For instance, memorising vocabulary is an essential part of learning a language. However, recalling words in response to a prompt is a separate performance to speaking a language. Fluent speakers do not 'recall' words, they use them to achieve their communicative goals. It is therefore always essential to focus on the context of performance.

Identifying and monitoring mental representations

On this account of deliberate practice, identifying and monitoring mental representations is essential to learning. However, no systematic study has been done of mental representations across various areas. There is not even a coherent theoretical account of mental representations. In fact, even research on working memory building on Miller's initial insight has mostly stalled [ref].

The identification of the mental representations has mostly been left to practitioners. The target mental representations are often left implicit in the practice guides or didactic manuals. However, it would appear that any one performance level can be achieved via different progress routes and any skill can be broken down into different chunks for successful performance. There are obvious limits to the variability but it is not necessary to assume that all learners need to follow identical paths or even focus on exactly the same chunks as targets for practice.

This is most important when considering the starting point for any activity.

Transfer of mental representations across tasks

As hinted, mental representations or chunks only transfer in a limited way across domains. The transfer only happens if two different performances of a task share a similar articulation. The mental representation required for fluent spelling or basic grammatical structures such as agreement or basic morphology transfer across all writing tasks. However, going from writing down sentences to fluently composing a coherent paragraph of academic English is an entirely independent journey of acquiring mental representations. But even at that level, different genres or contexts will require different mental representations.

Academics find it notoriously difficult to write for the general public. That is because the chunks built up to write an academic papers are unsuitable for the press. And new chunks have to be developed. Both tasks are articulated through a certain level of syntactic complexity that makes the transfer easier but ultimately, the process is akin to learning a foreign language.

This same issue is present across all domains of expertise. Chess players cannot just pick up Go and succeed. Medical diagnosticians cannot easily diagnose difficulties with their computers. Even though in both cases, the abstract principles that emerge when the task is described as a feat of reason are almost identical. Yet, the mental representations required to actually perform the reasoning within the limits of working memory make the transfer almost impossible.

This could also explain why generalist managers so often struggle across significantly different domains. While in practice there are both successes and failures [ref], the successes can only be possible if the manager develops sufficient fluency about their new field. But equally, this explains why domain experts frequently make inadequate managers. While they can easily conceptualise the set of actions required for success, they never acquired mental representations to reason with sufficient fluency about the process through which they can be achieved.

The lesson is that the primary focus of any practice should be acquiring mental representations for the task at hand. And as we saw above, successfully identifying the task and the requisite mental representations is a key component of successful deliberate practice

Mental representations and perception

From the above, it is clear that mental representations also have an impact on perception. Experts can not only produce and operate with larger chunks, they can also perceive larger chunks. An expert can see at a glance a complex situation and identify the problem whereas a novice needs to decode their world step by step, in a manner much more similar to a computer algorithm. For instance, a master chess player will see a very complex board configuration and immediately see the outcome without having to consider individual configuration of pieces. Concert musicians frequently see entire sections of a complex score as simply one element whereas a novice would have to decode every individual note.

This is, of course, trivially true of reading and writing at the lower levels of articulation. Obviously, expert readers can operate at higher speed and focus on more aspects of text other than simply the decoding of sequences of letters into mental images of words or their aural representations. However, expert readers of academic literature can pay attention to many more aspects of the text, they can skim, scan and make judgements at a glance. An expert reader of academic literature will also see be able to make connections, understand contexts and make more informed decisions about which parts of text to pay special attention to. This is also relevant to the writing process since a more expert reader will be able to better edit their own writing. Although it needs to be pointed out that the task of editing and test revision is different from reading for content and the transfer is not complete. However, some level of expertise in reading a text in a specific genre will be required for editing.

The same applies even in more complex areas such as medical diagnostics, emergency response, strategy, teaching, management, however, it has not been studied much outside of medical diagnostic fields and surgery.

It is important to note the great controversy around the impact of mental representations on perception. However, the claim here is not that the expert is "literally" seeing or hearing something different than the non-expert. The expert merely requires less time and fewer steps to process the relevant information whereas a novice will require so much time and so many step that their working memory will be effectively swamped and the effect will be some species of "attentional blindness". [ref]

Mental representations and automaticity

This requires an important caveat. What I have been describing could lead us to assume that the goal of deliberate practice is automaticity. The ability to perform tasks without thinking about them. This might be reminiscent of the famous distinction between levels of competence based on meta-cognition or awareness: Unconscious Incompetence, Conscious Incompetence, Conscious Competence, Unconscious Competence. It might appear, that deliberate is a process of achieving unconscious competence. But nothing could be further from the truth.

It is true that some levels of articulation form mental representations that are beyond the scope of conscious notice [ref Talmy]. For instance, articulation of phonemes in fluent speech is an example of competence that is not only unconscious but cannot be made conscious by an expert speaker without explicit training. However, almost any effort through which we might want to acquire mental representations will not focus on articulations of this kind.

While it is not based on research or theory [ref], this framework makes intuitive sense and is frequently deployed in training. The assumption is that unconscious competence is the target of instruction. It bears noting that in the original conception of this metaphor, this was not the case. It was first used to describe missionary educators and those with conscious competence were given as the examples of more effective practitioners [ref]. But true expertise requires a much more nuanced perspective that can be guided by the focus on the limitations of working memory.

In order to be able to reason and act in a sufficiently fluent manner, the expert cannot consciously attend to all the component mental representations required to perform the action. And as noted, at lower levels of articulation, this is impossible. However, this does not mean that the chunks or mental representations have become opaque monads to them. They can still make parts of their articulated structure available for inspection while attending to their use. The structure of the chunk and the chunk itself create a larger chunk which can be operated on in working memory.

In short, experts do not perform actions automatically without thinking, they perform them deliberately while using larger and richer chunks. To an outside observer, it may appear effortless or almost extrasensory in terms of perception. And experts frequently describe themselves thus. But when the expert performance is analysed, the structure and use of mental representations become much clearer. In short, the expert needs to both have access to the structure of their mental representations and operate on them as if they were monadic units.

Practice, enjoyment and motivation

The final consideration for deliberate practice is that it is unlikely to be inherently enjoyable due to the level of focus required. This is made worse by the fact that the particular thing practiced may not be directly meaningful and only serve as a scaffold. For instance, repeating scales in different patterns does not feel like playing music, or memorising vocabulary does not seem like speaking a language.

Ericsson reports that successful practitioners learn to find enjoyment in the process, particularly with respect to the progress. Given the amount of effort required, maintaining motivation is essential. Again, there is no definitive prescription for this but the most common suggestion is to monitor progress and reassess goals in order to receive perceptible performance rewards. This can develop a positive feedback loop, since monitoring progress is essential to the process of deliberate practice.

However, there is also a danger of circularity. It is a common side effect of gamification where the goal of obtaining the performance reward supersedes that of learning. For instance, learners using gamified vocabulary apps may focus on achieving high scores and optimise the process for achieving those as opposed for building a repository of words they can use. The measurement of progress must include a reflection of how the change in skill contributes to the ultimate goal of building larger chunks.

It is important that learners find ways to make repeated focused practice not just bearable but also enjoyable. This can include various rewards and varying the practice routine. Practicing in different tempos, with and without background music, etc. can reduce some of the tedium of the repetition required. But ultimately, the main source of enjoyment needs to be from the fact that the practice leads to improvement on the journey to the final goal. Seeing the results of practice in actual performance will be the greatest motivator.

To paraphrase the virtuoso bassist Victor Wooten from an interview he gave on stage:

Don't think about how bad you feel now when it seems like you will never get it. Think about how great you'll feel next week, when you'll be able to do this.

The importance of this for writing was stated most clearly by Basboll [ref] who suggests that finding enjoyment in the act of writing should be one of the goals of the process of practice. In his words, "enjoyment is a trainable skill". This can be achieved by periodic reflection on the benefits of the "writing moment", the focused time when writing or writing practice is the main focus.

Variability in effects of repetition and aptitude

One of the commonly recognised realities of instruction is the variability in how individuals respond to the same instruction across different subjects. Some people seemingly develop mental representations in certain fields by mere repetition, sometimes without any outward effort.

First, we know that that some chunks and mental representations do develop simply by repetition in almost anyone engaging in them repeatedly. Any typist will improve their speed over time, an adult studying a foreign language will acquire new words, beginner drivers will improve over time in how they navigate traffic and handle the vehicle. Sometimes a few repetitions suffice to develop a sufficient mental representation.

When a person moves into a new home, they will go through a period of small adjustments. Initially, even turning the lights on or opening doors requires conscious effort. Yet, in a matter of days of weeks, one develops sufficient "fluency" in performing these actions that they become automatic. Instead of looking for the light switch, thinking about which light it turns on, which way the toggle goes to turn it on, the action of switching on the light is one thing. I walk into a room with the intention of turning the light on and my body will perform the action of turning the light on as if it were one. Repetition over time was all that was necessary, although the process will involve some amount of conscious reflection.

However, mere repetition is not sufficient to overcome the inevitable performance plateaus. After reaching a plateau, performance stops improving or may even degrade over time. But perhaps even more importantly, individuals will differ in their ability to build mental representations in a given field by mere repetition. This is perhaps what can best be described as aptitude or talent. While the basic tenet of deliberate practice is that within certain limits no general human skill is outside the reach of anyone engaging in practice aimed at developing mental representations, it would be hard to deny that individuals differ.

For instance, the development of reading or writing skills at all levels displays a large degree of variability across all levels from primary school to academic writing. Individuals exposed to by and large the same instruction develop differently. However, most people develop sufficient reading and writing skills through instruction that is primarily conceptual followed by procedural repetition with no focus on building mental representations. Yet, the requisite mental representations do develop in a large proportion of the population. But equally a significant minority will never develop sufficient fluency in reading and an even larger proportion will never become fluent writers. The disparity is even higher in analogous procedural skills such as mathematics or foreign languages, or even physical activities such as sports or musical performance.

What accounts for this variability is a matter of speculation but traditionally it has been ascribed to aptitude or talent. In reading and writing, often the disparity is a result of specific cognitive impairments such as dyslexia or SLI. However, even these limitations can be overcome with a structured approach to instruction where mental representations are cumulatively built up under the careful supervision of the teacher acting as coach.

In cases, where the variability cannot be explained by recognisable cognitive or physical limitations, the differences may be attributed to a difference in how repetition is engaged in by the individual in the given subject. Most likely explanation is a positive feedback loop between initial success of building the foundational mental representations leading to the accessibility of further practice. These individuals are more likely to engage in more reflection about the process which further enhances what may be externally perceived as mere practice through repetition.

This is important for developing academic or non-academic writing skills. Words such as talented writer are often used to talk about certain individuals. Yet, this often ignores the amount of reflection these individuals engage in in their practice and how it increase the amount of repetition which further develops these skills. Struggling writers will need to be reminded of this fact.

Time required for developing mental representations

Another issue not sufficiently addressed in the literature is the time required to convert component actions into chunks that can be operated on by working memory. The total number of 10,000 hours has been suggested [ref] as the typical amount of time required to achieve peak performance levels in certain fields. Ericsson suggests it takes multiple years, as many as 10, during which deliberate practice is the main focus, to achieve peak performance skill. This is clearly not realistic advice for most instances because it only focuses on peak performers and does not consider the fact that most people only need to develop a skill to a much lower level of performance in order for it to become useful for their aims.

However, no evidence-based guidelines exist for the amount of time required to develop such useful and usable mental representations. But we can turn to an estimate by Josh Kaufman [ref] who suggests that it takes 20 hours to develop basic mental representations in a field of effort. He presents several case studies of personal development where 20 hours over the course of about 4-6 weeks helped him develop usable skills in a domain. For instance, he reports learning to play four chords on a ukulele, learn yoga to improve his mobility, learn basics of a new programming language, and learn to type using a new keyboard layout.

While no research exists to confirm this estimate, we can triangulate this approximate time span by comparing similar instructional programmes. For instance, the famous running programme Couch to 5k (C25K) succeeds in developing running skills over the course of 9 weeks and a total of about 15 hours of practice. A summer school in a foreign language may offer students about 30 hours of meaningful practice over the course of two weeks. The Writing Process Reengineering programme discussed below is designed to run over 8 weeks with a total starting at about 20 hours. 20 hours over 8 weeks is also the approximate amount of instruction received by students in an academic subject over the course of a semester.

As a broad estimate, we that anyone engaging in 15 - 40 hours of deliberate practice over the course of 2 to 10 weeks can develop sufficient combination mental representations or chunks that can be operated on in working memory in order to allow them to legitimately perform in a field albeit where they previously had no competence at all. This performance will be significantly limited and it will take many repetitions of these learning sprints to achieve even moderate levels of expert performance. However, this gives a realistic estimate on the lower bound of time for deliberate practice. This can then be transformed into usable programmes for developing academic writing skills or other analogous procedural skills.