Working memory and the real meaning of cognitive load
Last updated: June 2026
Working memory is the brain’s limited-capacity system for holding and manipulating information in the moment, while cognitive load refers to the total mental effort demanded of it during a task.
The Principle
In 1956, George Miller published his influential paper “The Magical Number Seven, Plus or Minus Two,” suggesting that people can typically hold about seven items (plus or minus two) in immediate memory. The idea stuck in popular design lore, but the reality is more nuanced. Later research, notably by Nelson Cowan, points to a more realistic functional capacity of around four chunks when controlling for rehearsal and chunking strategies.
John Sweller’s Cognitive Load Theory provides the deeper framework. It distinguishes three types of load on working memory: intrinsic (the inherent complexity of the material or task), extraneous (unnecessary mental effort caused by poor presentation or interface), and effort that contributes to building lasting schemas in long-term memory. Working memory is not just a passive storage slot but an active workspace with severe constraints in both capacity and duration—typically 20 seconds or less without rehearsal.
I learned this distinction slowly through building. Early on I treated “cognitive load” as a vague synonym for “complicated.” Understanding the three types shifted how I debug interfaces: the problem is rarely that users are “dumb,” but that extraneous load from unclear layouts or competing elements steals resources from the actual goal.
Why It Matters for Design & Building
The practical takeaway isn’t “never show more than four things.” It’s that we must ruthlessly manage total load, especially extraneous load, because working memory is the bottleneck through which all interaction flows. When it overflows, users don’t just slow down—they make errors, abandon tasks, or form incomplete mental models.
This became clear during a pricing configurator I built for a freelance client. The tool let users mix base plans, add-ons, usage tiers, and discount conditions on a single dense screen. From the backend it was elegant and flexible. In testing, users had to hold multiple pricing variables, eligibility rules, and comparison calculations in mind while scanning the interface. Error rates were high and completion slow. When I split it into focused steps with clear summaries of prior choices and smart defaults, extraneous load dropped dramatically. Users could actually think about their business needs instead of wrestling the form.
As a Design Engineer, respecting cognitive load means constantly asking: what is the user trying to hold in mind right now, and how much of my interface is forcing them to hold more? In AI products this is amplified—users already juggle the prompt, the output, and their evolving intent. Poor design here doesn’t just frustrate; it breaks trust. The honest practice is designing to protect working memory rather than testing its limits.
Real-World Examples
Virgin Atlantic’s flight booking flow demonstrates good management of load. Instead of presenting the entire process (flight selection, passenger details, payment, extras) on one overwhelming page, it breaks the journey into focused steps. This keeps intrinsic load manageable at each stage while minimizing extraneous demands, allowing users to hold only the necessary details in mind before moving forward. Completion rates benefit as a result.
Discover Bank’s older loan-application form illustrates the opposite. Its multicolumn layout combined with inconsistent field lengths and placeholder text forced applicants to constantly reorient and hold multiple formatting rules in working memory. Users frequently missed fields or entered incorrect information, increasing abandonment and support calls. The extraneous load from poor structure turned a moderately complex intrinsic task into a frustrating one.
A healthcare portal’s prescription refill process offers another negative case. The core form was simple (patient, medication, dosage, pharmacy), but the page was cluttered with promotional banners for flu shots, an auto-popping chatbot, multiple navigation menus, and unrelated sidebar resources. Clinicians and patients under time pressure had to filter noise while holding critical medical details, leading to frequent abandonments and errors. The extraneous elements consumed resources that should have gone to the primary task.
References
- Miller, G.A. (1956). "The Magical Number Seven, Plus or Minus Two." Psychological Review.
- Cowan, N. (2001). "The Magical Number 4 in Short-Term Memory." Behavioral and Brain Sciences.
- Sweller, J. (1988). "Cognitive Load During Problem Solving." Cognitive Science.
- The Decision Lab: Cognitive Load Theory. thedecisionlab.com
- Interaction Design Foundation: Cognitive Load. interaction-design.org
New entries are published every 2–3 weeks.
Follow along on X or LinkedIn to get notified.