Working memory runs everything. And holds almost nothing.
Alan Baddeley's multicomponent working memory model (1992) establishes the limits. Humans maintain and manipulate approximately four discrete chunks of information. Simultaneously. In active consciousness.
Individual items? Decaying within 20-30 seconds. Without rehearsal.
This limited-capacity cognitive workspace—distinct from passive short-term storage—actively processes information. Through executive control. Phonological rehearsal. Visuospatial manipulation.
Making interface designs respecting these constraints? Essential for complex task completion. Without cognitive overload.
The principle: Four chunks maximum. Externalize the rest. Respect the limits.
Baddeley and Hitch's seminal 1974 model revolutionized memory research by demonstrating that working memory comprises multiple specialized components rather than a single unified system. The central executive coordinates attention and processing across two subsidiary systems: the phonological loop (maintaining verbal information through rehearsal) and the visuospatial sketchpad (manipulating visual and spatial information). This architecture explains why people can simultaneously track visual layouts and remember verbal instructions without total capacity interference—different working memory subsystems handle different information types.
Cowan's meta-analysis (2001) refined Miller's original estimates, finding working memory capacity averages 4±1 chunks with 95% confidence interval, with cognitive load exceeding this threshold resulting in 67% increase in error rates and 43% longer task completion times.
Cowan's critical 2001 revision updated Miller's classic "7±2" estimate, demonstrating through rigorous experimentation that true working memory capacity approximates four chunks—not seven—when controlling for rehearsal and grouping strategies. His research showed that Miller's higher estimates reflected chunking abilities rather than raw capacity. Cowan's finding has profound design implications: interfaces simultaneously presenting five or more discrete information elements exceed typical user capacity, forcing either information loss or reduced processing efficiency.
Nielsen Norman Group's practitioner research translates working memory science into interface design guidance. Their studies demonstrate that users consistently struggle when interfaces require maintaining more than four pieces of information simultaneously—whether navigating complex hierarchies, comparing product specifications, or completing multi-step workflows. Successful interfaces externalize information to environmental memory (visible displays, persistent context) rather than burdening working memory with temporary storage demands.