“Why would you still bother memorising something if you can just look it up?” That thought has long felt almost self-evident. Smartphones, calculators, and ChatGPT make it tempting to simply Google anything you don’t immediately know. However, according to a recent publication by Barbara Oakley and colleagues, convenience might be one of the root causes of something much more fundamental: a creeping memory erosion quietly undermining our collective ability to think.

Their analysis starts from a striking paradox. As we seem to be getting smarter thanks to the technologies around us, average IQ scores in Western countries have been declining for decades. The well-known Flynn Effect – the steady rise in IQ throughout the 20th century – has reversed. And that seems more than coincidental. The authors point to a link between this reversal and our growing reliance on digital memory aids, combined with educational practices that have sidelined knowledge in favour of teaching “skills”.

Oakley et al. state that the widespread opposition between knowledge and skills doesn’t hold up. Neuroscientific evidence shows that without knowledge in your head, your skills start to fall apart, too. Knowledge isn’t just something you store and occasionally fetch from a drawer – it’s the scaffolding that supports how you think, reason, and spot errors. Without mental frameworks – what researchers call schemata – you can’t recognise patterns, make predictions, gain insight, or notice when something goes wrong.

And this doesn’t just apply to school subjects. The distinction between declarative memory (facts and concepts) and procedural memory (skills) is fundamental. Real expertise only emerges when knowledge is understood and internalised – when it moves from knowing to feeling, from declarative to procedural. And that requires practice. Not just reading or looking something up once, but repeating, applying, and reflecting.

When we outsource too quickly, relying on calculators or letting AI explain, we skip that crucial step. We fail to build robust internal representations. And that comes at a cost. Without sufficient internal knowledge, there’s no mental model to compare an outcome. No prediction means no prediction error, and therefore no learning. One example the authors give is that of a nurse who hasn’t memorised multiplication tables, who might not even notice when she makes a grave mistake in a dosage calculation. She trusts her device, but her brain has no internal check. No alarm bell. No correction. No learning.

What makes this article especially powerful is how deeply it dives into the role of schemata and so-called neural manifolds – organised patterns in brain activity that help structure knowledge efficiently. These don’t emerge on their own. They require repetition, variation, reflection, and – yes – effort. The more schemata we have, the faster and more flexibly we can process new information. But if we only look things up without integrating them, every new problem becomes an isolated puzzle. We recognise nothing, start from scratch each time, and overload our working memory. Instead of building a well-stocked mental toolbox, we end up with a messy pile of bookmarks and search results. The AI might know where something is, but we no longer do.

The authors are critical of constructivist educational models – a now quite fashionable position – in which explicit instruction is minimised. Not because students shouldn’t be active learners, but because without a strong foundation of knowledge, active learning simply doesn’t get very far. Especially biologically secondary knowledge, like reading, math, or scientific concepts, requires clear instruction, repeated retrieval, and guided practice. They call for a revaluation of memory, not as a goal but as a foundation. Spaced repetition, retrieval practice, interleaving – these aren’t outdated tricks but proven ways to train the brain. Technology can support that process only if it supplements internal memory, not replaces it.

This study is a powerful argument for putting memory back at the centre of learning, when we’re tempted to think we no longer need it. Outsource everything, and you build nothing. And in a world where everyone can look up everything, the real difference lies in what you already know. Memory isn’t a burden. It’s your cognitive engine room.

And perhaps, it’s also the foundation of our shared intelligence.

Abstract of the pre-print:

This chapter offers the first neuroscience-based theory linking the reversal of the Flynn Effect—documented declines in IQ scores across high-income countries since the 1970s—to the growing prevalence of cognitive offloading and pedagogical trends that minimize direct knowledge acquisition. Drawing on cognitive neuroscience, learning theory, and memory systems research, we argue that widespread underuse of the brain’s declarative and procedural systems has weakened internal representations—schemata—that are essential for reasoning, intuition, and expertise. We examine how shifts toward digital dependency and constructivist educational models have reduced opportunities for repeated retrieval, proceduralization, and the formation of robust engrams and neural manifolds. These trends disrupt the consolidation and automatization of biologically secondary knowledge, impairing schema-driven prediction, error correction, and transfer. We propose that the weakening of these memory systems—long before neurodegeneration sets in—may explain recent cognitive declines more plausibly than purely environmental or genetic accounts. The chapter closes by outlining implications for cognitive theory, educational practice, and AI-era learning environments. Rather than viewing memory as outdated in a world of instant information access, we argue that internal knowledge remains foundational for deep learning and that cognitive augmentation requires—not replaces—biological memory.

Show less