The Monthly Dispatch — What's New in Learning Science? — June 2026

Key Takeaways

• Retrieval practice fails under high cognitive load for complex texts
• Rational‑thinking training shows no transfer to argument evaluation
• Brief exposure creates overconfidence exceeding that of novices
• Removing learned cognitive tools harms performance more than adding them helps
• Retracted Wang & Fan meta‑analysis undermines AI‑in‑education evidence base

Pulse Analysis

Recent learning‑science research is sharpening the contours of what works in the classroom. A study by Redifer et al. demonstrates that when students grapple with dense, concept‑heavy material, the extra mental effort required for retrieval practice cancels its usual testing effect. This adds to a growing list of boundary conditions—feedback timing, material type, and now cognitive load—suggesting teachers must calibrate active recall to the complexity of the content rather than applying it indiscriminately. Similarly, Grimm and Richter’s experiments confirm that brief rational‑thinking training improves performance on practiced tasks but does not generalise to evaluating arguments, underscoring the long‑standing challenge of far‑transfer in education.

Equally striking are findings on the psychological side effects of shallow learning. Hong, Son and Kim reveal that brief exposure to new information can inflate learners’ confidence beyond that of true novices, creating a false sense of mastery that hampers subsequent instruction. Florean et al. further show an asymmetry in cognitive offloading: withdrawing a tool that students have grown reliant on—whether calculators, AI assistants, or other supports—damages performance far more than the initial introduction of the tool improves it. These insights call for deliberate scaffolding strategies that fade external aids and for curriculum designs that avoid seductive, irrelevant details that disproportionately distract low‑inhibitory‑control learners.

The most consequential development for ed‑tech stakeholders is the retraction of Wang and Fan’s highly cited meta‑analysis claiming ChatGPT boosts learning outcomes. Coupled with newer meta‑analyses that flag substantial publication bias, the evidence base for AI‑enhanced education appears fragile. While AI can raise task completion rates and lower stress, as shown in a recent RCT on programming courses, it does not deepen conceptual understanding. Policymakers and marketers must therefore temper claims of AI’s transformative impact and prioritize rigorous, independent assessments of genuine learning gains over superficial performance metrics.