Interleaving Effect

In the AI Story

The interleaving effect illuminates the architecture of professional expertise more clearly than perhaps any other finding in Bjork's framework. Experts do not merely possess more solutions than novices; they perceive problems differently, seeing structure and category where novices see undifferentiated difficulty. This perceptual expertise—the capacity to look at an ambiguous situation and rapidly, accurately categorize it before acting—is built through varied, interleaved exposure to problem types in unpredictable sequence. Blocked exposure, even extensive blocked exposure, cannot build this capacity because it never requires the discrimination.

AI tools provide blocked solutions by design. Each query produces a type-specific response: describe a bug, receive a bug-specific fix; describe a legal question, receive a legal framework; describe a strategic problem, receive a strategic analysis. The user receives correct, well-organized, domain-appropriate answers without performing the categorization that would determine the domain. Over hundreds of such interactions, the user develops skill in evaluating presented solutions but not in the prior, harder cognitive operation: determining what kind of problem she faces before selecting a framework for addressing it.

The 2006 Taylor-Rohrer geometry study became Bjork's paradigmatic demonstration of interleaving's power because it measured exactly what professional contexts require: not memory for facts but flexible application of procedures to novel cases. The students practicing interleaved problems were not asked to recall formulas—they had access to formulas during the test. They were asked to determine which formula applied to which problem. This discrimination, not memorization, determined performance. And discrimination is precisely the skill that AI-first workflows systematically fail to develop, because the AI performs the discrimination for the user.

Organizational training programs face the interleaving implementation challenge at scale. A junior consultant learning financial modeling techniques will learn faster in the short term through blocked practice (a week on discounted cash flow, a week on comparable company analysis, a week on precedent transactions). She will develop better long-term judgment through interleaved practice (daily rotation through all three techniques). The first approach produces specialists who execute within known categories. The second produces generalists who can determine which category applies. AI has made execution cheap and categorization expensive—exactly the inversion that makes interleaved training more valuable and more difficult to justify on quarterly timelines.

Origin

The interleaving effect was documented in motor learning research decades before cognitive psychologists identified it in conceptual domains. John Shea and Robyn Morgan's 1979 study of movement sequences established that random practice orders produced better retention than blocked orders. Bjork recognized the finding's broader significance: the same principle governing motor programs governs cognitive procedures. The benefit arises not from the specific domain but from the structural requirement that varied, unpredictable presentation forces discrimination between response types.

Kelli Taylor and Doug Rohrer's 2006 geometry study—published in Applied Cognitive Psychology—brought interleaving to educational attention by demonstrating effect sizes impossible to ignore: students practicing interleaved problems performed three times better on delayed tests despite performing worse during practice. The finding could not be attributed to motivation, interest, or instructional quality—the students, the material, and the total practice time were identical. The only difference was sequence, and sequence determined outcome with a magnitude that educational interventions rarely achieve.

Key Ideas

Discrimination is the skill that transfers. Professional judgment requires determining what kind of situation you face before selecting a response—a capacity built only through varied exposure that forces categorization, never through blocked exposure that makes categorization unnecessary.

Blocked practice feels better, works worse. Students prefer blocked practice, report higher confidence after blocked practice, and perform better during blocked practice—every subjective and immediate measure favors blocking, while every long-term and transfer measure favors interleaving.

AI provides blocked solutions. Each AI query produces a type-specific response, performing the problem categorization for the user and delivering a domain-appropriate answer—bypassing the discrimination step that interleaving builds and that judgment depends on.

Junior practitioners need interleaving most. Early-career professionals are building the discriminative architecture that will govern their judgment for decades; AI-first workflows that eliminate categorization challenges during this formative window prevent the architecture from developing.