A few words on Cognitive Load Theory

In essence Cognitive Load Theory (CLT) is a simple idea: people have limited capacity for processing information; by designing learning experiences and materials in a way that respects those limitations, we can improve the process.

Though there is nothing fundamentally wrong with this idea, it risks distracting us from the things that really matter, when people learn.

Background:

Cognitive load theory draws on research suggesting that humans have well-established limitations in their ability to handle items of information simultaneously - for example, Miller’s (1956) research into the ‘7 item limit’ of short term memory (although in practice we typically only process 2-4 items concurrently), or limitations identified by Baddeley & Hitch’sWorking Memory Model (1974). The Working Memory Model suggests a number of cognitive ‘subsystems’ which operate under the direction of a ‘central executive’, such as a ‘visio-spatial scratchpad’ for reconstruction of visual scenes or a ‘phonological loop’ for handling of acoustic information.

The basic idea is that the ‘central executive’ system shuffles information in and out of these subsystems as it is being retrieved from long-term memory, or being stored in long-term memory – consequently ‘Working Memory’ can be thought of as a somewhat more sophisticated way of thinking about short-term memory.

The evidence for Working Memory is pretty sound: firstly, studies of brain-damaged patients will sometimes show that one system has been damaged, whilst another remains intact suggesting the systems are, indeed, distinct. Secondly, ‘dual attention’ tasks, in which people’s ability to do two things at once are tested, suggest cognitive systems which function independently – since some tasks can be undertaken concurrently and others cannot.

As a simple example of things that people struggle to do simultaneously:  people cannot both read a book and listen to a speech. You might wonder why this is since – at face value – reading requires vision whilst listening to a speech is an auditory activity. The Working Memory model suggests that the same system – the acoustic loop – is used to handle incoming sounds, as well as the sounds of the words that you are reading – so the two conflict. On the other hand you can probably drive and listen to the radio at the same time – suggesting that you employ separate systems to do these tasks. However (as you may already have thought) people will sometimes turn down the radio when driving becomes especially taxing – leading some psychologists to hypothesise that our cognitive systems have some combined capacity limitations. Whilst the debate continues, it is consistent with the everyday experience of being ‘overloaded’ – for example when you are trying to watch the television, while the radio is on, and someone else is attempting to have a conversation with you.

You can probably already see that this approach has some implications for education: for example it is a bad idea for an instructor to read text from a screen to students – since either the students are reading the text themselves (so cannot be listening to the instructor) or listening to the instructor (and ignoring the text). Since reading and listening are both automatic processes, it is likely that the two will interfere and information from one or the other lost. Smartphones may also be a distraction for students (although I will argue below that it would be far better to worry about whether the material is sufficiently interesting to warrant the student’s attention)

The original research into cognitive load theory (Sweller, 1988) focussed on the relative merits of worked examples vs practice problem-solving (e.g. in mathematical instruction), and key recommendations included breaking complex problems into smaller steps and ensuring that information that students will need to process together, is presented together.

Discussion:

Although Cognitive Load Theory has some worthwhile applications, it risks distracting us from more important variables affecting learning, some key concepts are poorly or completely undefined, and it is narrow in application. Overall it may lead us to focus on the presentation of the material we are teaching, rather than the learning process and the learner.

There are a number of things to consider in assessing Cognitive Load Theory:

Firstly, it is misleading to think of attention, short-term and long-term memory as if they are distinct systems – much like boxes into which information is transferred. The multi-store model of memory, for example, suggests that we transfer information from short-term to long-term memory via (acoustic) rehearsal – repeating it over and over. In real life, we don’t do this. A person arriving at work to be told, in hushed tones, by a colleague that their ‘skirt is tucked into their underwear’ will not need to repeat the information over and over in order to remember it. Indeed it is hard to think of any everyday occasion where we repeat information over and over to aid memory – with the obvious exception of phone numbers (though younger generations seemed to have stopped remembering these).

Though the working memory model has largely superseded the multi-store model, there is a more general point to make here: such techniques are used when information is insignificant (i.e. has no especial affective significance) . Rehearsal has received undue attention precisely because it is a technique to be used with meaningless information – the information in other words that our memory systems are designed to discard – but which we encounter a lot of in education. When considering cognitive load theory there is a very real risk that research and application are confined to the kinds of weird and non-generalisable situations that we encounter in education, rather than those we encounter when learning normally. To put it crudely: cognitive load may help us with education problems, but not with learning problems.

We still see some of this legacy thinking in the proposed role played the acoustic loop in transferring information to long term memory, in the Working Memory Model. To be clear, what I am saying is this: in normal circumstances we do not have to use acoustic repetition to transfer things to long-term memory. Use of this technique is almost entirely a phenomenon created by the peculiar situation we call ‘education’. Similarly, research into CLT has tended to look at ‘problem-solving’ contexts such as solving mathematical problems, rather than normal problem-solving contexts such as ‘Where can I get something to eat?’ or ‘How can I get Fred to like me?’ – and as you can probably immediately appreciate it is far less clear how ‘cognitive load’ applies in normal contexts (for which we are cognitively adapted).

Once more: attention, short-term and long-term memory are not distinct systems, with information being ‘transferred’ between them like golf-balls between shoe-boxes; instead we should think of cognitive processes as integrated and continuous with information being selectively stored from the range of things of which we are aware. In turn, what we are aware of will depend on what wehave stored. Short-term memory is a little like the handprints left in wet cement, long-term memory the imprints left as the cement dries.

The next issue to be aware of is the use of the ‘chunking’ concept within the theory. Early cognitive theories made use of the idea of a ‘chunk’ as a unit of information to be stored. This probably owes something to computational analogies, but unlike a computer it is much harder to define units of information when considering cognition: a ‘chunk’ can be virtually any size – for example ‘America’ or ‘restaurant’ or ‘Goldilocks and the three bears’, or ‘FBI’ or ‘13’. Cognitive theorists acknowledge this, frequently using the expression ‘schema’ to refer to the way in which people organise related information into a coherent pattern, which can then be treated as if it were a single unit. Cognitive Load Theory makes explicit reference to this – pointing out that unrelated information (i.e. not already integrated into a schema) will place more cognitive load on the individual than information that is already part of a schema stored in long-term memory. An everyday example of this might be navigating a supermarket: over time you develop a sort of mental map of items, allowing you to quickly skip to the ones you wish to find – but were someone to completely reorganise the shelves, this would slow you up considerably.

Whilst the concept of schema is useful, and the overall recommendation sensible, it has never been clear what schema are or how they work, and this introduces weakness into the approach. Cognitive load theory, in other words, doesn’t rest on a clear model of cognition. Some theorists imagine that schema are sets of semantic relations (e.g. a restaurant has tables and chairs), but there are good reasons for thinking that this cannot be the case (e.g. comparative psychology or AI attempts to implement this approach).

The Affective Context Model does provide a theoretical basis for understanding schema and cognition more generally: a schema is comprised of a related set of reactions to experiences. When we attend to experiences those aspects of our experience which elicit an affective response (‘reaction’) enter our awareness. Since we cannot possibly take in the entirety of our sensory experience, our reactions determine which objects are worthy of our attention. Of course what we react to will depend on our past experiences – for example if we are delighted by a certain era of architecture, we will notice these features as we stare out of the window of a train; the vast majority of information we have no reaction to, and therefore forget.

This explains why experiences to which we have an especially strong reaction may immediately be stored in long-term memory (such as a fight on the train) without the need for processing in short-term memory. It also explains why our attention is selective and shaped by past experience. In familiar environments we have little reaction to familiar elements, which are already encoded in our affective schema. This makes for a very efficient system, since only what is significant is stored, but is likely to result in a great many inaccuracies regarding specific details (as Loftus and others have found)

Returning to the problem-solving context, the difficulty for the learner is that if many of the problem components are new, then processing will be more effortful and will require more cognitive capacity, since they are not already stored. The affective context model therefore agrees with cognitive load theory in predicting that challenges of increasing complexity – i.e. which allow learners to create a ‘broad outline’ affective schema, before moving onto challenges of increasing sophistication and complexity, will work best. It suggests that where worked examples are used, they should not contain so may unfamiliar elements that the learner will struggle to grasp them all at once.

There are also a number of ways of reducing cognitive load that cognitive load theory does not consider – presumably because the theory is restricted in application to specific educational contexts. For example, checklists and guides (as Atul Gawande discovered) are highly effective at reducing cognitive load since they reduce the need to store extraneous information entirely and thereby free up cognitive capacity. It is far easier to think about your day at work if you are concurrently using a GPS device to navigate an unfamiliar city, than if you are not. Of course the reason cognitive load theory does not generally consider these applications is because of the implicit (educational) assumption that the learner's job is to store information in their heads. I mention this merely to point out that in the real world we are all familiar with techniques for reducing cognitive load – such as making a list, or using a pen and paper to help us solve a problem. These days, we are making increasing use of technology to reduce cognitive load.

I do not mean to suggest that the way we present a problem doesn’t make a significant difference: people familiar with the instructions that accompany flat-pack furniture will also be familiar with the problems caused by presenting the problem poorly – on one page a list of pegs of different sizes, labelled A10, A11, A12 – on another a list of panels labelled C2,C5, C8. On a third page instructions for assembling panels C5 and C8 using pegs A10, A12. We are now having to hold four unfamiliar items (‘chunks’) in our working memory as we review the assembly instructions – no wonder we get annoyed when people ask us questions!

The major objection to Cognitive Load Theory is not, therefore, that it is wrong – but that is it a distraction from more important aspects of the learning process. The cognitive effects that it describe are most applicable in a relatively narrow range of contexts created by the education system and therefore have limited applicability to real-world learning. Even worse, they risk distracting us from what is really happening when we learn.

The Affective Context Model argues that the central feature of learning is the affective significance of experiences to a given learner. Cognitive load will make some difference to problem solving but what really matters is whether or not someone cares about solving the problem. I imagine most of us are aware of this: if your car breaks down in the desert, the design of the repair manual may not be ideal, but you’ll figure it out nonetheless.

In educational contexts almost all of the affective context is assumed: we assume that the student respects the teacher, we assume that the student wants to learn, we assume that they wish to pass the test, we assume that they believe getting good grades will improve their future prospects – but these are the very things that determine the encoding of experience. With Cognitive Load Theory we are tinkering at the fringes of the process, rather than tackling learning itself i.e. the affective context.

Whilst it is true that an instructor reading text from a screen is sub-optimal, what really makes the difference is whether any of this stuff matters to the student – whether the instructor is able to relate the material to what the student cares about – or even just tell it as a story. As Bower and Clark showed, making something into a story can improve recall seven-fold. That’s a far bigger effect than anything CLT has to offer. It may be the case that smartphones distract the learner from the content – but this is generally a symptom of boredom. The root cause is that the learner cares more about what is on their smartphone than on the instructor’s screen. No-one is asking educators to become entertainers – I am arguing that all encoding of information depends on the reaction it causes in the learner – and unless we take the time to find out what matters to them, and relate the information accordingly, we are wasting our time however much we reduce cognitive load.

Defenders of CLT might accuse me of shifting the goal-posts; CLT is not primarily about recall but about problem-solving in specific contexts. If that is true then the range of application may really be relatively narrow – that is to say, once we have mapped and worked with all the affective aspects of the instruction (e.g. how important is the problem to the student, what existing concerns does it relate to, how does the problem statement relate to these concerns, what is the student-teacher relationship etc.) then cognitive load theory may prove useful – but certainly it is not the starting point for instructional design. The starting point for instructional design is the learner, and what matters to them.

In conclusion: 

Cognitive load is certainly something to consider, but being an educator who worries about cognitive load but not affective context is like being an athlete who worries about footwear but not their exercise routine.