I didn’t realise how many issues I had with Cognitive Load Theory until I saw teachers talk about it online.
Having wrestled with assumptions, terms, and proposed procedures of Sweller’s original theory for years.
It wasn’t until I saw teacher interpretations that got me diving deeper.
Despite the incredible contribution to learning science. The theory has too many myths for me.
- What cognitive load theory is.
- Proposed myths in cognitive load theory.
- Why cognitive load theory should be retired.
But first.
What is the theory?
Cognitive load theory is too vague
The concept of cognitive load isn’t unique to cognitive load theory.
Interchanging terms like mental load, mental effort, brain capacity, or energy have been used to refer to the same concept.
Surely there is a difference. Right?
In the 1970’s John Sweller and other researchers looked at student problem solving.
Their observation. Less thinking relates to better results.
The main question. How do we leverage this observation to get better results?
Science tests hypothesis. So they tested hypotheses.
So they tested and looked for mechanisms. Causes. Reasons for better performance they could then leverage.
Working memory was a growing theory around this time, and in the 1980’s that is what cognitive load referred to.
Cognitive load referred to working memory storage and processing.
This is true today.
That is why we see diagrams like this showing working memory at the centre of a persons head.
However, when someone refers to cognitive load theory, it might not be the cognitive load they are talking about.
As you can see by the above image, cognitive load is split into three. Intrinsic, extraneous, and germane.
But it wasn’t always like that.
Cognitive load is split for a reason
Originally there was intrinsic and extraneous load.
However, after a study using variability found that more variability leads to better learning transfer. So something had to change.
Extraneous load was seen as bad.
Adding variability would have been adding extraneous load, so therefore should have been bad.
But it wasn’t.
This meant some level of added load could be good. Germane load.
Most blog posts talking about cognitive load theory will show something like:
- Intrinsic load: Inherent difficulty of a task.
- Extraneous load: Difficulty added by how we learn.
- Germane load: Used for learning.
However, extraneous load was all the theory focused on up until the late 1990s.
Intrinsic load was fixed. So no reason to focus on it.
The variability study was published in 1994, so Germane load wasn’t considered until the late 90’s.
This meant reducing extraneous load was the sole focus.
Extraneous load is the most important load
Problem solving is where the research started.
In 1972 means-ends analysis was one method proposed to solve problems.
Trial-and-error.
Have a goal. Break it into sub-goals. Do things to achieve the sub-goals until you achieve the starting goal.
For humans. Problem-solving was suggested to be done in working memory.
All of that information and calculation build up a cognitive load.
In theory, if we reduce the extraneous load, the cognitive load would be decreased.
Remove the goal. That will lower the cognitive load.
Remove the problem. That will lower the cognitive load.
Thus, goal-free problems, and worked examples were used as ways to make problem solving easier.
To me, this sounds like a huge change in the problem-solving experience, thus the learning potential.
But let’s stick with the theory for now.
After testing goal-free problems and worked examples, researchers saw better participant performance.
Proposing effects which contributed to the better performance.
All the resulting instructional designs encouraged by cognitive load theory, are supported by these proposed effects.
The effects support everything
Each effect is attributed to a type of cognitive load. Intrinsic or Extraneous.
Germane isn’t included potentially due to it’s delayed inclusion to the theory.
Alternatively, Germane load isn’t included as Sweller says:
This means that as Germane load is ‘used for learning’, it wouldn’t be able to ‘demonstrate an effect’.
Germane load (I think) would be the desired effect.
The table below shows each suggested effect from Sweller’s 2011 chapter with the associated load type.
However, the foundational principles are arguably more important.
Foundational cognitive processing principles
If the cognitive load effects are ways to alter instructional design, the cognitive load principles are why the alterations should work.
But these principles are not facts.
Assumptions are a part of science.
However, assumptions in science can lead you to myths.
More on those later.
The figure below shows each suggested principle from Sweller’s 2011 chapter with an overview.
These ‘natural information processing principles’ are the roots of the theory termed ‘cognitive architecture’.
I have added the overview column to group principles with their reason. It helps me understand their purpose. It might not help you.
All this together means when someone refers to cognitive load theory, they could be talking about various things:
- Types of cognitive load.
- Cognitive load effects.
- Principles of cognitive architecture.
All of which are used as evidence for instructional design.
But this evidence is full of myths.
The myths in cognitive load theory
A myth could be:
- A story involving supernatural beings.
- A fictitious or imaginary person or thing.
- Widely held but false belief or idea.
- Misrepresentations of the truth.
Truth being a notoriously difficult word to define I am using true in this context as, supported by fact.
I am trying to avoid an epistemological debate in this article.
Let’s keep that word count down.
Then. As this is a theory.
I will also look to explore the theories to a logical philosophical conclusion.
My logical philosophical conclusion at least.
For me, a myth is something not supported by facts and is misrepresenting a commonly agreed philosophical belief. Which I believe.
Yes, you could argue these myths are assumptions.
In that case, maybe think of them as assumptions which I think should be retired.
So.
Cognitive load.
We can't measure cognitive load
A reminder that in this theory cognitive load refers to the demands of working memory.
The demand is built by interacting elements.
But unless I am missing something, there are no specifics on what an element is.
Just that lots of elements interacting is a high cognitive load.
And less interaction is a lower cognitive load.
Sweller saying that:
However, I don’t see any mention of some element interactions requiring more or less processing than others.
Instead of the number of elements interacting. The severity or difficulty of the interaction.
Without specifics. Elements and element interactivity are therefore highly ambiguous.
Not very factual.
The best measure I have found was from a study predicting mental workload.
It looks impressive.
But this doesn’t consider residual load.
Load that remains after the previous action before starting this measuredĀ task.
So the start isn’t 0, and the baseline likely wouldn’t remain constant either.
And more importantly, it’s built from questionnaires.
Likart scales of 1-9. How much effort was that task?
1-9 How difficult was that task?
Measuring a perceived feeling of energy expenditure after an action as a substitute for cognitive load.
But energy expended on a task includes physical expenditure.
Not just the working memory!
Even if we assume no energy was expended through the body. Which I will expand on in a bit.
I would assume long-term memory requires energy to retrieve, reject, and store information.
This would again make these energy measurements inaccurate and unreliable for working memory.
Some have have discussed learning efficiency as a substitute.
However, the effort is still subjective and not specific to working memory.
Yes, we use energy to complete tasks.
Yes, we can assume there is a cognitive load for completing tasks and solving problems.
However, referring to cognitive load as working memory can't be accurate.
A 2016 review of load theory discusses the splitting and differences of perceptual load and cognitive load.
In cognitive load theory, the theories environment organizing and linking principle suggests working memory works with long-term memory.
This means cognitive load includes energy expended through the demands of long-term memory as well.
Following the logical solution that cognitive load is therefore demands of all natural information processing.
We are still left with other energy expenditure.
This brings us to a philosophical question. What is cognition?
Cognition is more than the brain
The 4 E’s of cognition is an ongoing conversation in philosophy.
- Extended Cognition
- Embodied Cognition
- Embedded Cognition
- Enactive Cognition
Bounds of cognition could involve non-derived representational content, and be functionally similar to actual cases of internal, brain based human cognition.
Cognitive systems could be part of an integrated, persisting system, and it must causally contribute to a wide range of cognitive phenomena.
Science of the mind could belong to a subject in the sense of contributing to the intentional content of that subject’s personal-level states.
But without a consensus. Conversation is difficult.
However, it is agreed among most philosophers that cognition includes the body. Embodied cognition.
To what extent is debated. But cognition is still embodied.
Counting with your fingers is an example of using your body while thinking.
Thus energy expended while completing tasks is embodied. Not working memory.
This suggests cognitive load, as defined by Sweller, isn’t supported by facts and misrepresents common philosophical beliefs.
Thus cognitive load, in this context, is a myth. Or is it an assumption that should be retired.
Cognitive load effects don't say anything new
If we overlook the previous section, the effects are also questionable.
Effects ‘result in new instructional procedures with better test outcomes than a traditional procedure’.
I am not sure what counts as ‘traditional procedure’ but I would assume something not supported by cognitive load theory instructional design.
And or, something that has limited or no support in previous literature.
Variability in practice
The variability effect shows that adding variability increases intrinsic cognitive load resulting in increased learning if working memory resources are available.
For the ‘if working memory resources are available’ caveat, I will refer to the previous myth.
But the difference here is the grouped intrinsic load is for tasks over a certain period of time.
As Sweller mentions:
This is significant, and will come up later!
Opposite to the variability effect, is the isolated elements effect.
Isolating elements decreases the intrinsic cognitive load.
Considered together, variability and isolation sound like desirable difficulty.
Likely a factor for the use of the term in the 1994 study.
Hence, we should adjust the practice difficulty to suit the learner.
However, variability in movement coordination was introduced into learning science in the 1920/30s.
Bernstein’s more recent publication in 1967 (years before Sweller) discusses differences in expert and novice movements.
He found expert blacksmith hammer movement was not about producing the same movements, but achieving the same goal.
Hitting the hammer right.
Bernstein called this ‘repetition without repetition’.
Repetition without repetition sounds very similar to variability in practice…
Don't use goal free problems
Yes. Removing a goal reduces cognitive load.
But it also fundamentally changes the problem-solving to the point this effect is often ineffective.
Sweller suggests the worked example effect is more useful.
The worked example effect…
However, from what I have read and understand it’s not that students learn more or better, but rather quicker.
This 2009 study emphasizes ‘less time to process’, ‘solved more rapidly’ and ‘decrease in the number of mathematical errors’.
Opportunities to forget are limited in this study, so long-term learning is questionable.
Mistakes, challenging misconceptions and errors have also been shown to contribute to long-term learning in various contexts.
However, worked examples reduce those opportunities.
This suggests benefits are short term, not necessarily long-term.
Communicate information effectively
Physically integrating separate sources of information can reduce extraneous cognitive load.
Aiding understanding.
However, some information can be understood in isolation.
One article reconsidering cognitive load theory said:
Unintelligible in isolation sounds like issues with the amount of available information.
The modality effect is similar.
Presenting material using audio-visual rather than visual formats is beneficial.
This also sounds like it’s addressing the amount of information communicated.
It is worth noting that a reverse modality effect was found in 2004 and has been repeated.
Trying to avoid too much or too little for the learning experience.
But Fitts and Posner in 1967 proposed a three-stage-model of learning.
Those in the cognitive stage trying to work out what to do.
Educators were advised to reduce the amount of information given.
Sound similar?
In the associative and autonomous stages, more information is given and used as they work out how to perform skills.
Thus effective communication between educator and learner, or learner and learning materials will change with expertise.
How that differs from desirable difficulty, I am not so sure.
Learning changes everything
When we consider the redundancy and expertise reversal effects.
If the information says the same thing or something is already known, it is redundant and therefore is extraneous load and should be avoided.
But not always.
Practice involves repetition. That is what practice is.
When and how to include repetition is ambiguous.
The variability effect is over time. Practice over time.
I said this would come up later.
When the learner gains expertise, what was once useful could become redundant.
The element interactivity effect uses the aforementioned unspecified elements.
Sweller suggests these interactions are proposed to explain understanding.
I would personally say the depth of understanding rather than the level of element interactivity.
But as long as communication is clear, I don’t think there is an issue.
However, element interactivity as an effect relies entirely on the assumption, myth, that cognitive load is the demands of working memory.
Setting that aside for now, this effect suggests tasks with lower intrinsic load with added extraneous load may have fewer instructional consequences.
Simpler tasks allow for more complex instruction.
This aligns with the recommendations from Fitts and Posner.
Break down complex skills into simpler tasks.
Sweller goes on to say that:
This ‘increasing level of expertise’ is encapsulated by the suggested problem-fading effect.
Both problem completion and guidance fading are included here.
Higher levels of expertise require more element interaction from the task.
In other words. Desirable difficulty.
Make problems harder and or reduce guidance for those with higher expertise.
Remembering the right information
Humans forget things. Duh.
And the transient information effect suggests technology doesn’t always help.
As we will explore later, expertise in this theory is about information stored in long-term memory.
Thus, if the information is stored in technology, we don’t have it in long-term memory and therefore we are ‘less skilled’.
If. A big if. We don’t use the technology.
This brings us back to that philosophical question. What is cognition?
Extended cognition challenges this effect in more ways than one.
- Cognition including the technology.
- Cognition is supported by technology.
- Technology functioning like the brain.
Technology functioning like the brain is the important claim here.
Sweller saying that:
But his focus is on time, rather than the information.
So when Sweller says :
The medium focus is inaccurate.
Yes, you can go back and read text, or look at an image again.
But you can also do that with video. Audio-visual presentation.
Technology functioning like the brain. Recalling and replaying past events.
Practice is throughout time, and if technology can be used throughout time, it can and should be included in practice.
The focus here should be on how a learner remembers information. Not how they ‘receive’ information.
Remembering as a process is different from the event of having remembered something.
Ecological psychologists would suggest verbing your nouns.
But this theory is built from cognitive psychology.
So remembered it is.
But if some information goes straight into long-term memory because of its significance.
Transient information is ambiguous.
Now.
Processing is used to explain the imagination effect.
The effect proposes imagining procedures or concepts can be more effective than studying material.
But imagining what?
Using past information and expertise to complete a task.
Processing. Remembering.
Both are continuous throughout time.
But this sounds like desirable difficulty again.
Processing information is easier with expertise.
We should retire these effects
Each of the cognitive load effects is argued to get better outcomes and differ from traditional procedures.
My logic is that the new procedures would have limited or no previous supporting literature.
Considering the variability and isolated element effects.
Yes, variability can get better outcomes but the effects don’t, from what I can see, suggest when or how to do that in practice.
They rely on an immeasurable value. Working memory demand.
Variability in practice is also not new.
Granted, this might not have been deliberate, but the research and writing was and still is there.
Considering the Goal-free and worked example effects.
Yes, they get better outcomes. But only measured in the short term.
Long-term learning research suggests the ‘instructions’ these effects support would not be as effective.
Considering the split-attention, modality, problem completion, guidance fading, redundancy, expertise reversal and element interactivity effects.
They all align with desirable difficulty.
Before the term was coined, there were various instances of discussion about the differences in instruction for novices and experts.
Increase the difficulty of the task as the learner develops.
These effects specify how you could increase the task difficulty.
But this has once again been explored in various ways before cognitive load theory.
And to greater depths the effects in this theory don’t cover.
Considering the transient information and imagination effects.
The transient information effect makes claims that contradict extended cognition.
An aforementioned philosophical approach.
Due to the publish dates, we could logically assume the conflicting comments might be adjusted with more modern technology.
However, transient information and the imagination effects both rely on remembering information over time.
With ambiguous research, I am not convinced these effects are well supported. Nor do they result in new procedures or better test outcomes.
Apart from reinforcing practice variability and differentiating learner experiences for expertise, these effects don’t say anything new.
Claiming these effects result in new instructional procedures with better test outcomes than a traditional procedure is inaccurate.
Some overlook previous literature.
Some are misleading.
Most echo previous literature.
These effects are observations of practice. Not lawfully structured patterns.
Although not myths or assumptions worth retiring, implementing instructional design from these effectsĀ can be dangerous.
Emphasizing a point that Sweller made.
This comment seems to often be forgotten in the online discussions I tend to see online.
But the principles all these effects rise from are not that stable.
The unstable foundational principles
You may have seen these principles referred to as part of human cognitive architecture.
Status of knowledge is also included but that will be discussed afterwards.
I included an overview section in the above table because principles overlap.
But they shouldn’t!
Storing information in the brain
The long-term memory store is a well-documented theory in cognitive psychology.
One idea is that genomes in our DNA store information.
However, there is no agreed way to measure the size or information density of a genome.
de Groot’s research about chess grandmasters found better chess players were better at remembering chess positions than regular chess players.
But that wasn’t true for random positions.
This was put down to better players having lots of board positions in long-term memory.
Now, remembering things is obviously important. But what, where, and how it gets stored is a question that can't be checked effectively.
We can’t take out a long term memory store to see what was stored. It’s all in theory.
Brain scans are not specific enough. Yet.
Long-term memory stores are a theory needed for cognitive psychology.
Not a unique principle of cognitive load theory but an assumption required to do science.
However, this limitation is key!
Inputting information into the brain
The borrowing and organizing principle on the other hand has some issues.
Borrow suggests we can give information back. Which we can’t.
Acquire is the other word used.
But acquire suggests others have a store which we gain access to, somehow, to either make a copy or reorganize.
The theory suggests we reorganize the copy of information.
But if we have access to their store, why don’t we copy it?
Sweller references Bandura’s social learning theory here.
But observation isn’t explicit instruction!
But if we acquire information without interacting with people (observation), are we acquiring it from their long-term memory store?
Or.
Are we developing our expertise from our experience observing them?
That would suggest we are generating novel information.
Something cognitive load theory says we rarely do.
This theory argues schema, stored groups of information, is unique because of reorganization.
But if reorganizing information is unique, then it must be new. For us at least.
Expending energy to reorganize information into something new, sounds like generating new or novel information.
But that is a different principle. The Randomness and genesis principle.
But the theory doesn’t say how different a procedure needs to be, to be new.
There doesn’t seem to be a distinction between information and knowledge, or what different knowledge is, either.
Apart from the biological primary and secondary knowledge categorizations which I will get to later.
In maths, is a different number new enough? Negative numbers? Fractions? Decimals?
If you add numbers together, procedures will differ with each number. When if ever is it new? or not new?
Cognitive load theory seems to discourage ideas about constructivist approaches.
The idea that emphasizing people building their understanding, and learning from experiences generating new knowledge is less effective.
But the generation effect was an idea put forward in 1978, with supporting data, and with more recent support.
When we consider the borrowing and organizing principle. It mimics empiricism. Learning from experience.
But the randomness as genesis principle doesn’t.
With no reliable methods to measure how, when, or where information is sent to storage.
We must rely on philosophy. Specifically empiricism from epistemology. The study of knowledge.
Most empiricists follow two principles:
- We start with no knowledge. Tabula rasa or blank slate.
- Knowledge can only be gained from the senses.
Thus the randomness as genesis principle breaks the second principle of empiricism and is therefore a myth or an assumption that should be retired.
Processing information in the brain
The narrow limits of change principle focus on working memory.
Working memory is limited to 7 items for around 20 seconds for storage, then 3-4 items when processing.
But unless I am missing something, all we can do to test processing is to observe brain activation.
I hope the brain is active when we are thinking.
As to the item limit. What counts as an item?
A single number? A year? Which is 4 numbers. A phone number? Which is 10 plus numbers.
Schemas and chunks have been suggested as ways to group information.
Is one schema or chunk an item?
This ambiguity means although the limit of 7 might be accurate, 7 what's, will change for each person.
Logic would also suggest this would be impacted by an individual’s level of expertise.
The narrow limits of change principle form the roots of element interactivity and working memory demands.
However, when considering the relationship between working memory and long-term memory, this is where there are lots of different theories.
Long-term working memory has been suggested alongside other ideas about how these work together.
For those unaware, short-term memory theories have been rejected by most cognitive psychologists, and evolved into working memory.
But these can’t be proven either.
Introducing the environmental organizing and linking principle. Linking the external environment with the information store.
But empiricism’s second principle, knowledge can only be gained from the senses, suggests we need the external environment to gain knowledge.
Thus, working memory and long-term memory are both required for processing.
The narrow limits of change principle focus on working memory.
Whatever this processing thing is, it must be different from working memory because, as Sweller says:
With no reliable methods to measure how, when, or where information is sent to storage we must again rely on philosophy. Empiricism.
Thus the narrow limits of change principle breaks the second principle of empiricism and is therefore a myth, or an assumption that should be retired.
We should retire these principles
Each of the cognitive load principles is argued to be uniquely required for natural information processing.
However, without sufficient evidence and philosophical conflicts, I think some of these principles are myths and should be retired.
Assume long-term memory. Ok.
Assume learning from others. Ok. But that requires generating new knowledge.
We can’t copy someone’s memory store. Reorganizing sensed information is generating new knowledge.
Randomness as genesis requires sensed information which replicates the borrowing principle. These principles should be combined.
Assume working memory limits. Ok. However, a working memory item limit is ambiguous and falls victim to the cognitive load myth mentioned above.
In addition. Processing requires information from long-term memory.
The Environmental linking and organizing principle links working memory and long-term memory. This also considers working memory demand.
Narrow limits of change rely on a myth and replicate the environmental linking and organizing principle. These principles should be combined.
As five becomes three, we can see that these principles were unstable.
Now these principles are stable they mimic theories of indirect perception.
Not unique. And certainly not new.
Retire this categorization of knowledge
The status of knowledge is one of the newer additions to cognitive load theory.
Geary suggested knowledge could be split up into biological primary and secondary knowledge.
Biological primary knowledge relates to biological primary skills like listening or speaking.
The theory is that we learn to speak unconsciously, without explicit instruction, and are internally motivated.
Speech therapists were used as a caveat here but everyone has educators for speech.
Some require more help than others.
But some people don’t learn to speak at all because they use sign language. The deaf.
They are internally motivated to learn – a way of communicating – not necessarily speaking.
So when Sweller says:
I disagree. Other people are the reasons why we learn language skills.
Unconscious learning or implicit learning, happens in lots of experiences including reading, which is categorized as a biological secondary skill.
The theory argues biological secondary skills are unlikely to be learned without explicit instruction.
However, we learn from experiences, empiricism, so we can learn skills without explicit instruction.
The reason described for separating these skills is the ease of acquisition.
Another quote that I think is useful context here
However, home schooling is a thing.
Much like a parent teaching a child early year skills before going into formal education.
Or children learning skills outside of formal education.
If the example skills can fit into both biological primary and secondary knowledge categories, depending on the individual and their situation.
These categorizations aren’t lawful. They are ambiguous.
Why are we separating these skills or knowledge?
This is important because the theory focuses on secondary knowledge and skills.
The argument is the skills are acquired differently.
So how do you apply the theory to practice for primary skills, if they are acquired so differently?
An article reconsidering cognitive load theory said:
Yet instructional design principles work for so called primary skills like speaking.
If the instructional designs from the theory should only be applied to secondary skills, but works for primary skills. Why separate the skills?
With no evidence I can see supporting this separation of knowledge, again I am looking at epistemology.
If knowledge is:
- The sum of what is known.
- Information held in storage.
- Awareness or familiarity gained by experience.
The philosophical categorization can be discussed.
But applying the categorization to skills isn’t logical.
Through the lens of cognitive psychology, performing a skill or behaviour uses:
- What is known.
- Stored information.
- Awareness or familiarity with past experience.
Categorizing a skill or behaviour to align with categories of knowledge goes against common uses of knowledge for skill development.
Thus biological primary and second knowledge applied to developing skills, is a myth. Or is an assumption that should be retired.
Cognitive load theory should be retired
All frameworks are wrong some are useful.
Cognitive load theory has been useful for decades and will continue to be useful.
Empiricism. We learn from experiences – our past.
However, the meaning of cognitive load contradicts modern research suggestions and popular philosophy.
The effects are narrow and serve better as observations of phenomena.
Effects often lead to misinterpreted applications or discussions for instructional design which isn’t helpful.
Unstable principles which when stabilized, echo indirect perception add complexity to the literature.
And the categorization of knowledge applied to skill development is highly ambiguous.
Cognitive load theory adds unnecessary complexity to educational science.
I think we would be better off focusing on:
- Variability of practice.
- Representative learning experiences.
- Long-term skill development.
With clear and well stated objectives, learners can use natural learning to develop skills.
Creating clear well-stated objectives is a skill learners should develop in formal education.
But if people have set a goal, manage available resources, seek help and feedback while doing related tasks, ie practice. They will develop the skills.
Some have argued cognitive load theory is part of a cognitive science movement.
Even a social game as this author puts it:
Cognitive load theory clouds educators with assumptions applied as rules.
This is a theory, not fact.
We should treat it as such.
For me. That’s moving on.
