Question
Does cognitive load theory actually predict anything, or just describe failure?
The standard story is that cognitive load—working memory's finite capacity—explains why people struggle with complex tasks, bad interface design, confusing instructions. It's become a catch-all in product, education, UX. The problem is how unfalsifiable it's become. When something is hard, we say it's high cognitive load. When it's easy, it's low. We've described the outcome, not explained it.
What actually interests me is that the framework works better in reverse—as a diagnostic tool rather than a predictive one. You can look at a failed interaction and ask whether someone actually ran out of working memory or whether they lacked domain knowledge, got distracted, misunderstood the goal. The theory helps you *narrow down* what went wrong. But I'm skeptical it predicts much before the fact. Cognitive Load Theory literature is full of studies in controlled settings—memorization tasks, specific problem sets—where you can isolate variables. Real contexts are messier. Two people with identical working memory capacity handle the same task completely differently depending on whether they've seen the pattern before, whether they trust the system, whether they're tired.
There's also a historical parallel worth noting. The 1950s-60s had a similar moment around "channel capacity" and information theory—Miller's magical number seven, all that. Those were useful framings at the time. But we eventually realized the brain doesn't work like a telegraph. It's not that cognitive load theory is *wrong*, exactly. It's that it's narrower than we act like it is. It's a piece of the puzzle, not the puzzle. I'd be more cautious about using it to make confident claims about design decisions.
1 comment
Log in to comment.
You're right that CLT has become diagnostic rather than predictive, but I think there's a sharper distinction to make here. The theory *does* predict something specific—it predicts that performance degrades when task demands exceed working memory capacity *during that task*. That's testable and it holds up. The problem is we've expanded what we call "cognitive load" to mean basically any mental difficulty, which gutted the specificity.
When a student bombs a math problem because they're juggling too many intermediate steps at once, that's CLT doing real work. But when we say a website is "cognitively heavy" because it has too many colors or options, we're usually describing aesthetic preference or decision paralysis or information architecture problems—which might *involve* working memory but aren't the same constraint. A chess master can hold twenty positions in mind simultaneously. A novice can't hold three. That's not a working memory deficit; it's pattern recognition. CLT doesn't really account for the difference.
Your channel capacity parallel is apt. Where it breaks down is that CLT assumes working memory is this fixed-size bucket, but actual behavior depends so much on what's already *in* long-term memory—schemas, heuristics, domain knowledge. Two equally intelligent people process the same technical documentation at completely different speeds depending on what they've seen before. CLT on its own can't predict that. It's why the theory works fine in the lab with arbitrary nonsense syllables but falters once expertise enters the picture.