Cognitive Load Theory – Is it just a Load?

by Gary Woodill on August 5, 2010

Note: This item has been cross-posted to Workplace Learning Today (Aug. 6, 2010), where I also blog.

Recently, well known author Jane Bozarth wrote an article on “Brain Bandwidth – Cognitive Load Theory and Instructional Design” in Learning Solutions Magazine. Essentially, the argument in this theory is that “there is only so much new information the brain can process at one time.” This is called Cognitive Load Theory, and it is central to the work of such prominent researchers as Richard Mayer and Ruth Clark. One of the main pieces of evidence for the theory is a 1956 article by George Miller who “suggested that the largest number of discrete pieces of information the brain could manage was seven, plus or minus 2.”

Stephen Downes, in OLDaily, points to Jane’s article, but makes this critical comment: “I think cognitive load theory misrepresents how we acquire and store information. It supposes that information is atomic and symbolic, like a string of numbers.”

I have to agree with Stephen on this point.

In fact, the idea of a limited capacity of the brain to memorize a list of items goes way back to Hermann Ebbinghaus, a philosophy instructor at the University of Berlin in the 1880s. How he came up with this idea is documented in Frank Smith’s 1998 critique of learning theory, The Book of Learning and Forgetting, published by Teacher’s College Press. I quote at some length:

How could anyone make comparisons on any aspect of learning when people are so different, especially in the two things that make learning possible for anyone (according to the classic point of view) – interest and past experience? In the revealing language of science, interest and past experiences “contaminate” experiments and “invalidate” results. People who have a great interest in the topic or activity, and who have had a greater experience of it, are bound to learn more. And they ruin experiments. What experiments need is a method of control… so that the learning task is fundamentally the same for everyone.

…This was Ebbinghaus’s world-changing revelation: if you want to study how people learn without the involvement of interest and past experience — study how they learned nonsense. By definition, no one is interested in anything that makes no sense to them, and by definition, nothing in past experience can help anyone learned nonsense.

Ebbinghaus invented the nonsense syllable, a staple of psychological research ever since. He also described “the learning curve” which is that the ability to memorize nonsense syllables drops off around 10 items, and “the forgetting curve”, which is the memory of most of the nonsense syllables quickly drops off within a few hours.

As Kurt Danziger (one of my professors at York University in the late 1960s) points out in his book Naming the Mind: how psychology found its language (Sage, 1997), Ebbinghaus “defined memory in terms of the work of memorizing and not in terms of the experience of remembering. In this context ‘learning’ was used as a synonym for memorizing, and experimental investigations were designed to answer questions about the relative efficiency of different techniques of learning.” In other words, before Ebbinghaus, the word learning had several meanings in psychological, biological, and philosophical writings, but after, at least in North American psychological literature, learning became synomous with memorizing.

Rereading George Miller’s original article shows that he was talking about a limited kind of task – the ability to discriminate among different audio tones (also a nonsense task). Around 7 different tones, people start to make lots more mistakes. But he also suggests many ways of overcoming this seeming limit on working memory.

It seems that by adding more dimensions and requiring crude, binary, yes-no judgments on each attribute we can extend the span of absolute judgment from seven to at least 150. Judging from our everyday behavior, the limit is probably in the thousands, if indeed there is a limit. In my opinion, we cannot go on compounding dimensions indefinitely. I suspect that there is also a span of perceptual dimensionality and that this span is somewhere in the neighborhood of ten, but I must add at once that there is no objective evidence to support this suspicion.

This hardly seems to be hard nosed science, but it is often cited as “research” for “evidence-based learning”. We need to examine our concepts carefully and critically, and move away from research into nonsense as the basis of our instructional designs. (GW)

Nuts and Bolts: Brain Bandwidth – Cognitive Load Theory and Instructional Design | Learning Solutions Magazine | Jane Bozarth | 3 August 2010

{ 4 comments… read them below or add one }

Will Thalheimer August 6, 2010 at 3:19 pm

My comment cross posted at the brandonhall site…


Your critique makes a fundamental error. You focus on only one of the threads of research supporting the idea of cognitive load. You slam it, but you fail to acknowledge the many other areas of research that support the idea that our cognitive processing has limited capacity.

Even on the surface, the idea that we do NOT have cognitive limitations is just absurd. Be real dude!!

Try counting backward by 3’s and doing multiplication problems at the same time. Try writing while listening to music with lyrics. Try talking with your spouse while on the phone with your boss. Try talking on your cell phone, driving your car in traffic, and listening to Lady Gaga lyrics for deep meaning.

But don’t worry about your errors, you can blame it on cognitive load.


Chris Craft August 7, 2010 at 6:39 am


I must agree with Will that you focus exclusively on one aspect of cognitive load theory that was posited not by the theory’s authors but rather by Ruth Clark.

Either way, CLT takes into account the human cognitive architecture. We naturally have a limited working memory and a virtually limitless long term memory. I am surprised you would call that into question.

Even anecdotally, I recognize the limits of my working memory as I try to remember phone numbers. I find myself rehearsing them as I walk to the desk to write them down.

Yet the limited working memory is a fundamental assumption of the theory, not a portion of the theory itself. CLT deals with avoiding overload (which is when the working memory overloads, resulting in cognitive default).

I would be glad to share a good primer article with you that will better explain the concept of CLT. I fear you have not explored CLT enough to be bashing it.



Gary Woodill August 13, 2010 at 11:23 am

First, thank you Will for posting your comments to both sites where I had my original piece. Because the bulk of the comments were made at Workplace Learning Today, I am copying them here (and will avoid cross-posting in the future).

Paul Lefrere August 6, 2010 at 11:07 am [edit]
I applaud your suggestions re: looking carefully and critically at the basis for instructional design, and re-reading original articles to see what they actually said, rather than relying on summaries by others. Like self-reflection, scholarly investigation is out of fashion, even though it’s likely to be more reliable than the seductive short-cut of finding a summary on the internet. As to which, maybe you’d like to add your points to wikipedia to balance its article on cognitive load

Jared Stein August 6, 2010 at 11:56 am [edit]
A few points:

First, I’m not sure whether you’re confusing capacity to learn with capacity of working memory. Cognitive load describes the latter, whereas when you write “the idea of a limited capacity of the brain to memorize a list of items” I get the impression you’re describing the former.

Second, I Miller’s paper refers explicitly (as I recall) to chunking, which is not contradicted or refuted by cognitive load theory.

Finally, you conclude “[we need to] move away from research into nonsense as the basis of our instructional designs.” — is this a typo?

Clark Quinn August 6, 2010 at 12:16 pm [edit]
Sorry, Gary, I can’t let this go by. First, Cognitive Load Theory (CLT) is John Sweller’s work, and it’s got a solid body of evidence to support it. In short, simplify cognitive load to the right amount, and you get increased performance.

The phenomenon is real: most people can do two column addition in their head, most people can’t do multi-column addition in their head (as Jane’s article started with). Stephen’s right at one level, but at another level we have pretty reliable ‘chunks’ (c.f. Eleanor Rosch’s work on categorization). The point is, when designing learning, we pretty well should know what our learners do and don’t know, so we know what’s ‘chunked’ and what isn’t.

And if we are trying to design the most optimal learning experience (which we owe to our learners), we should be working hard to have the challenge level appropriate, and that includes adjusting the overall difficulty to be in just the right level of increment from beginning to appropriate level of performance. Right?

This introduction to the special issue on CLT and Instructional Design provides a nice overview:

Yes, it’s a bit ‘instructivist’, but I reckon we have to find a way to integrate empirical results across a body of approaches if we’re going to pragmatically design successful learning solutions. Which, at the end of the day, we need to do.

Gary Woodill August 7, 2010 at 12:31 am [edit]
Thanks for both the critical and supporting comments.

I certainly agree that there are cognitive limitations in terms of what we can think about at the same time. I was specifically questioning the George Miller article and its conclusions (the rule of 7) – and whether or not we can move from Miller’s fuzzy and speculative thinking in 1956 to concrete prescriptions for instructional design today. I still think that this is a big stretch.

BTW, Jared, Miller never uses the word “chunking” (as an action) in his article, but refers to “chunks” and adds “we are not very definite about what constitutes a chunk of information”. In the end he speculates that nonsense information that is recoded into words or pictures is more likely to be remembered. In other words, Clark, it is not a matter of simplifying to the right amount of information that works best, but making it meaningful.

As John Sweller has noted, limitation on working memory only applies to novel information that is not connected to previous experience (i.e., nonsense numbers, tones or words). When there is previous experience, then what is being learned is meaningful, and therefore can be quickly integrated into what is already known.

Jane Bozarth August 7, 2010 at 8:20 am [edit]
Gary, I appreciate the response but agree with Will that it seems you decided to take a slam-stance against the theory, and agree with Clark that the theory has sound elements. Also, this was not so much an ‘article’ as one entry in my monthly (first Tuesday) “Nuts and Bolts” column for Learning Solutions Magazine. The column is meant for those many who have found their way into eLearning design and purchasing (some from classroom training, others from IT, HR or other fields) without much benefit of formal instruction (hence, “Nuts and Bolts”). I am trying to provide 750-1,000-ish word overviews of ideas and concepts important to these folks without overwhelming or intimidating.

It’s no secret that much ‘eLearning’ just tries to do too much, and I hope the column made folks think about that in new terms. As you didn’t take issue with any of the solutions I proposed (use modules, eliminate extraneous material, chunk content in meaningful ways, attend to differences with novices v. experts) I’m assuming you still support the idea of not overwhelming learners.

In addition to Will’s suggestions (such as “try counting backwards by 3s and doing multiplication problems at the same time”) I ran into another good example while reading on a plane yesterday, one to which I bet everyone can relate. Author Bill Bryson, describing eating at a restaurant and being overwhelmed with choices: “If you order, say, a salad, the waitress reels off sixteen dressings, and I am not quick enough to take in that many concepts at once.”


Jane Bozarth´s last blog ..Look Inside Social Media for Trainers

Doug Holton August 7, 2010 at 5:32 pm [edit]
The research backing cognitive load theory is mixed, and ever since the theory was expanded in 1998 to include ‘germane’ cognitive load, it became unfalsifiable. Anything you add that helps performance is ‘germane’, anything you take away that helps performance was ‘extraneous’, and vice versa.

See volume 38, number 2 of Instructional Science for some reflections on cognitive load by Ton de Jong, Roxana Moreno and others. Here’s my summary:

Ultimately what we care about as instructional designers is, what are the implications of a cognitive (or other) theory for instructional design, for practice, and what new techniques does the theory inspire. The main instructional implication of (pre-germane) cognitive load theory was to simplify, simplify, simplify. Essentially spoon feed content to the learner in little bite-sized pieces. The main “new” technique to be affirmed by cognitive load theory was worked examples. Similar to the teacher’s version of a textbook, if you show students how to solve a problem, surprise, they’ll more quickly learn how to solve that problem and other very similar ones.

As I said in a comment to my post, be careful of research based on lab studies (like most cognitive load research) rather than in the classroom or in the field. There are alternative research approaches that have better generalizability to practice. Also be careful of justifying instructional design decisions based on cognitive load theory, as it (with germane cognitive load) can be used to justify anything after the fact, but biases you of course to the most simplified, piecemeal forms of delivery, which may not always be the best approach (and rarely is at least in highered, K-12 contexts at least).

Situated cognition/action is another theory which also had instructional implications and inspired or affirmed new instructional techniques such as game-based learning, problem-based learning, and so forth. Developed in the early 90s, it may be the reason why germane cognitive load was tacked onto cognitive load theory in 1998, because situated approaches take the opposite approach, making the learning environment ‘messy’ or ‘complex’ or ‘real-world’ at first, rather than starting with bite-sized pieces. You give students an exploratory simulation or field experience first, and then help them connect the dots with a lecture or presentation or reading assignment afterward, and they learn much more than in the opposite order.

There are other emerging psychological and social theories, as well, that also have instructional implications – phenomenology, embodied cognition, activity theory. I’m just a pragmatist, as I think most instructional designers are – I and you should pick and choose the best ideas and applications of various theories, try to do so in a coherent way so it’s not just a disconnected hodgepodge of instructional design techniques, and not get too religiously attached to one particular theory.

Carla Jacobs August 9, 2010 at 1:01 pm [edit]
It’s always interesting to see what happens to a common-sense idea when academics get together and decide to wring all the meaning out of it. (Also dirty pool to title the piece “Cognitive Load Theory — Is It Just a Load?” if you were really, as you say, only taking issue with Miller’s Rule of 7.) After all, to quote a sarcastic Twitter friend: “Any idiot can see that 749820613 is easier to remember than 749-82-0613.”

Gary Woodill August 13, 2010 at 11:10 am [edit]
Thanks for the additonal comments, everyone. I don’t think that there is enough debate in our industry, so having two sides of an issue aired is a good thing.

Jane, I didn’t critique your suggestions because 1) I admire your efforts in helping trainers sort out concepts, and 2) I’m an overly polite Canadian. For example, I think that any reduction of interest and engagement is not a good thing, as much e-learning is boring. I also want to be clear that I was not attacking you, but CLT.

Doug, thanks for the eloquent elaboration of some of the issues in CLT. You confirm that there ARE two sides to this debate.

Carla, all “common sense” ideas are not good just because they are common sense. It used to be common sense for most people that gays were mentally ill, but that is only common sense for some people now. I see nothing wrong with questioning common sense.

Chris, thanks for pointing me to additional resources on CLT. I plan to let this controversy die down, and come back to it at a later date when I have time to do justice to the topic.

Thanks again, everyone, for your contributions.


Gary Woodill August 18, 2010 at 10:19 pm

Additional comments on this issue, posted at Workplace Learning Today:

Paul Simbeck-Hampson August 18, 2010 at 7:24 am [edit]
Very interesting and lively discussion, learning alot

This discussion occurring over on Amplify may also be of interest to readers here…

Paul Simbeck-Hampson´s last blog ..How do we make training more effective and avoid cognitive overload by Dr Ruth Clark &lt Simple but so true!

Mary Thomas August 18, 2010 at 2:51 pm [edit]
Tangential Musings on The Learning Curve: Doctrine or Design Flaw?

Gary Woodill’s August 6, 2010 mention of “the learning curve” sparked musings on its basis as doctrine or design flaw. I too grew up on Ebbinghaus learning curve doctrine.

In contrast, however, I am reminded of the work of Murrary Sidman on errorless learning. Based upon this work, Sidman asserts that the learning curve is myth. Sidman’s rationale goes like this:
“Teaching a pupil all the prerequisites for a task will produce errorless learning. If errors do occur, they can be eliminated by identifying and teaching the missing prerequisites. Learning need not be a trial-and-error process for the pupil, although it may be for the teacher. Errorless learning indicates that the learning process is all-or-none; the learning curve becomes discontinuous, with any continuity residing in the teaching process.”

Sidman (1999, 2010) argued that learning need not be a trial-and error process for the learner. Learners need not make mistakes to learn. Instead, properly designed instruction can elicit errorless learning.

Can learning be errorless? Is the learning curve inevitable or bad design? Can the the learning process shift from a learning curve to an all-or-none event? Hummm. What if we were to entertain that:
1. The onus of responsibility shifts from the learner to the teacher.
2. The source of learning variance lies with the teaching process, not the learning process.
3. Readiness to learn is pivotal to errorless learning and a common source of failures.

Just thinking out loud.

Sidman, M. (1999). Coercion in educational settings. Behaviour Change, 16, 79-88.

Sidman, M. (1994). Reflections on behavior analysis and coercion. Behavior and Social Issues, 3, 75-85.

Sidman, M. (2010, May). Errorless learning and programmed instruction: The Myth of the learning curve. Paper presented at the Analysis of Behavior Association (ABA) Annual Conference, San Antonio, TX.

Gary Woodill August 18, 2010 at 10:11 pm [edit]
Because this discussion started by my agreeing with on Stephen Downes’ critical take on Jane Bozarth’s article on Cognitive Load Theory and Instructional Design, there have been related discussions on other sites. I wanted to point to these in order to give a complete picture of the debate.

Stephen Downes commented on my post, and has one reply.

Paul Simbeck-Hampson in Germany posted about cognitive overload, and the comments from that post linked to the debate on this blog.

If there are other comments floating around on this issue, please let me know.



Leave a Comment

{ 1 trackback }

Previous post:

Next post: