The Cognitive CollapseSubstack Icon

3786 words

For most of the 20th century, humanity grew smarter on paper. Average IQ scores steadily climbed—an observation known as the Flynn Effect, reflecting gains of roughly 3 to 5 IQ points per decade across many countries (Trahan et al., 2014). Better nutrition, education, and healthcare were often credited as drivers of this long-term rise in cognitive performance.

But something has gone fundamentally wrong. In recent decades that upward trend has stalled or even reversed. Norwegian studies found IQ gains stopped after the mid-1990s and declined in numerical reasoning (Sundet et al., 2004). Danish conscripts showed a 1.5-point decrease between 1998 and 2004 (Teasdale & Owen, 2008). In Britain, Flynn himself found that average 14-year-old IQ scores dropped by more than two points between 1980 and 2008, with the upper half showing even worse performance—a six-point decline (Flynn, 2009). Bratsberg & Rogeberg's landmark 2018 study demonstrated that both the original Flynn Effect rise and subsequent decline occurred within families, confirming environmental rather than genetic causes. This isn't a statistical anomaly—it's a systematic weakening of cognitive capacity precisely when economic competition demands cognitive strength.

The timing couldn't be worse. Just as human intelligence begins declining, economic forces are dramatically raising the bar for human relevance. Professional services firms—the traditional training grounds for cognitive workers—are abandoning their century-old "pyramid" structure where armies of junior staff learned through grunt work. Goldman Sachs analysts predict AI could expose up to 300 million jobs worldwide to automation, with banks already considering cutting analyst hiring by two-thirds as AI handles tasks like updating financial models (Business Insider, 2024). Law firms report AI tools reducing human contract reviews by 40%, while Jamie Dimon predicts AI could automate 60-70% of work in coming years (Hindustan Times, 2024). What emerges is a “Diamond-shaped” organisation—narrow at the bottom, wider in the middle, still narrow at top—that simply cannot train the next generation of experts. The staircase to expertise is being dismantled for short-term efficiency gains, creating a talent pipeline crisis that threatens entire professions.

We are witnessing a perfect storm: declining cognitive capacity colliding with rising cognitive demands.

The Root Cause: The Memory Paradox

A compelling explanation for this crisis has emerged from neuroscience and education research. Barbara Oakley and colleagues (2025) call it “The Memory Paradox”: the idea that in an age when information is instantly accessible, we risk weakening the very mental faculties that made us so cognitively formidable. The paradox is that in a world of information abundance, we feel less need to remember facts or methods—after all, why memorise something that Google or an AI assistant can retrieve in seconds? But this very convenience may be making us cognitively weaker.

As Oakley and colleagues put it, “The more we offload knowledge to external aids, the less we exercise and develop our own cognitive capacities.” This isn't merely about losing individual facts. Human memory, comprising of both declarative (facts and concepts) and procedural (skills and habits) knowledge, needs regular use to strengthen the neural circuits that underpin reasoning. When those circuits go underused, our ability to think through novel problems and learn new ideas suffers. Over-reliance on calculators, search engines, and now generative AI means we practice using our own memory and problem-solving muscles far less than previous generations did.

The Memory Paradox shifts the focus from attention to retention. Many commentators worry that social media is shortening attention spans. But even if one's attention is captured in the moment, learning won't stick without engaging memory processes. A concept understood today can be completely forgotten next week if not revisited and encoded in long-term memory. We've gone so overboard on the idea that understanding is the only important thing in learning that we forget that if you can't remember what you've learned, you still really don't understand it.

True understanding depends on memory. If knowledge evaporates after exposure, there is no foundation for higher-order thinking. Neglecting memory-building creates a brittle intellect: an intellect only capable of momentary flashes of insight at best, but lacking the deep wells of knowledge needed for sustained reasoning or creative problem-solving. This memory deficit explains how we can have an information-rich society that paradoxically produces individuals who struggle with complex reasoning: we have emphasised short-term exposure and external reference over internalised knowledge, and now we are seeing the consequences.

How Education Abandoned Memory

The shift away from memory-based learning has been decades in the making. In the late 20th century, educational reformers increasingly criticised traditional memorisation—often deriding it as rote learning that produced mindless fact regurgitation. The pedagogical fashion swung toward teaching critical thinking skills, problem-solving strategies, and broad exposure to many topics without dwelling on details. Mastery of content and memorisation of fundamentals came to seem old-fashioned, even counterproductive, in many Western schools (save for areas of law and medicine where the use Anki and other flash card mechanisms are still the hallmark of the best students).

Instead, the idea was that students should learn how to learn, with facts being less important than the ability to find information on the fly. This well-intentioned shift fixated on breadth over depth: curricula expanded to cover more subjects and skills, but with less time to drill down and truly absorb any single area. Educators began focusing on hitting "tick-box" outcomes (ensuring each item in a standards list was introduced) rather than ensuring students had thoroughly internalised the knowledge. Many schools started treating initial exposure, not lasting mastery, as the goal of instruction in each unit.

The result was that students encountered a lot of content, but much of it was only superficially processed and quickly forgotten. As Oakley and coauthors argue that the very educational reforms that promised to make us smarter—particularly the widespread rejection of memorisation as “rote learning” and the shift toward “critical thinking skills”—may actually be undermining our cognitive abilities. Modern classrooms often emphasise project work, using tools, and collaborative problem-solving. Those are valuable activities, but when they come at the expense of foundational knowledge (e.g. vocabulary or scientific facts), students end up with shaky scaffolding.

This is sometimes referred to as “skill signalling”: students become good at displaying skills or following procedures taught in workshops, yet they lack deeper content fluency or the ability to reason independently. A classic example is a student who can Google any formula or use an app to solve equations but struggles to set up a simple algebraic problem without those aids. Education researcher David Didau sums it up bluntly: "It doesn't matter what pupils understand if they can't remember it… everyone understood everything [during the lesson] and yet no one seems able to recall any of it next lesson."

Over years, students educated under this paradigm accumulate exposures but not durable competencies. They may have been taught critical thinking frameworks, but without factual knowledge those frameworks have nothing to grab onto. This phenomenon directly links to the IQ declines: as core skills like literacy, numeracy, and factual knowledge receive less emphasis, performance on cognitive assessments that rely on those basics will inevitably suffer.

The Technology of Cognitive Offloading

Layered on top of educational trends is a technological transformation in how we handle information. We live in an era of cognitive offloading, where tasks once done mentally are now routinely offloaded to gadgets and software. Need to do arithmetic? Pull out the calculator app. Can't recall a historical date or scientific term? Ask Google or ChatGPT. Unsure whether a fact is true? Reply “Is this true, @grok?”. Don't want to compose an email from scratch? Use an AI writing assistant.

Cognitive offloading in itself is not new, but its scale and ubiquity today are unprecedented. When we consistently outsource mental tasks, we sacrifice the practice that our brains require to build skill and understanding. As Oakley notes, “when we consistently outsource thinking to technology, we effectively bypass our brain's natural reinforcement learning mechanisms—the very pathways that build and strengthen neural manifolds supporting sophisticated thinking.” By letting external tools take over, we short-circuit the process by which we normally internalise knowledge and gain intuition.

Consider the "Google effect" (also dubbed digital amnesia). Studies have found that people are less likely to remember information if they know it can be easily looked up online (Sparrow et al., 2011). Our memory systems subconsciously decide, "I don't need to encode this; I can always find it later." What seems like reasonable efficiency—why memorise what's on Wikipedia?—in aggregate leads to poorer recall ability and general knowledge. Similarly, reliance on GPS turn-by-turn directions can weaken a person's natural spatial memory. Many of us can barely remember phone numbers or birthdays now, tasks that were routine a generation ago.

This habitual offloading undermines three critical cognitive processes:

  • Automaticity: The ability to perform skills or recall knowledge automatically, with little conscious effort, due to extensive practice. Automaticity allows a person to read words or do basic arithmetic without thinking about it, freeing up working memory for higher-level tasks. Achieving automaticity requires repetition and memorisation—exactly the activities many modern curricula minimise. Without basic facts made automatic (like multiplication tables or common vocabulary), students' brains get overloaded when tackling complex problems.
  • Schema Construction: In cognitive psychology, a schema is an organised knowledge structure or mental model. Building schemas is how we connect individual facts into a bigger picture of understanding. Strong schemas are what make someone an expert: what a novice perceives as separate, confusing pieces, the expert perceives as a coherent whole. However, schema construction doesn't happen without focused learning and memory of details. If students only ever engage in shallow exposure or rely on reference tools, they never build the rich interconnected schemas in their own minds.

By leaning heavily on external aids, we inadvertently stunt these processes. A calculator can give you the answer but won't contribute to your number sense or mathematical schema. Wikipedia can provide facts but won't build the mental framework that an expert possesses. The net effect is a kind of learned helplessness: people appear proficient as long as the tool is at hand, but when faced with a problem that requires independent reasoning or recall, they falter.

It's telling that the decline in IQ scores aligns chronologically with the digital revolution. Oakley notes that the introduction of handheld calculators in the 1970s corresponded with a drop in students memorising basic math facts, which in turn hindered the development of intuitive understanding in mathematics. Each technological leap has further reduced the need to mentally grapple with lower-level tasks. The reversal of the Flynn Effect can be interpreted as a warning that we have pushed cognitive offloading too far, creating a generation of people who are superficially skilled but fundamentally weaker in raw cognitive endurance.

The AI Slide and Workplace Transformation

This cognitive decline is coming at the time when more and more of our tasks are becoming automated. One particularly stark manifestation is the "AI Slide": the removal of the lower rungs of the staircase to expertise in professional development thanks to AI automation. Traditionally, most professions had an apprenticeship model: new entrants started with simple, routine tasks and gradually took on more complex responsibilities: step by step, Junior analysts would clean data and make charts before learning to craft insights; step by step, junior lawyers would proofread contracts and summarise case law before arguing in court; step by step, entry-level engineers would fix small bugs and write test cases before designing entire systems.

These humble tasks weren't glamorous, but they served as steps on both the professional ladder and steps up the educational stairs towards competency: they built the foundational skills and institutional knowledge that, over years, transformed novices into capable practitioners. Now, imagine smoothing the bottom steps off the stairs. That's essentially what happens when AI-driven workflows augments and automates the basic tasks.

As more tasks get automated, the organisational structure looks less like a pyramid and more like a diamond. In consulting, law, and tech, leaders are noticing a collapse of the junior layer of talent. Fewer entry-level hires are being made, because managers figure those tasks can be automated. Some firms are responding creatively—Goodwin Procter implemented intensive simulation-based training for new associates since they won't get practice on routine tasks (LexFusion, 2024), while PwC tried mandatory AI training for all 75,000 U.S. employees (Going Concern, 2023). The broader trend is clear: if entry-level positions dry up, there will be no next generation of experts.

The AI Slide is essentially the workplace mirror of the memory paradox. At an individual level, offloading tasks to AI can prevent a person from developing skills; at an organisational level, replacing entry-level work with AI prevents the development of talent. In both cases, the staircase that connects novices to experts is being dismantled. This creates an illusion of competence: a company might function well for a while with a lean, automated operation, just as an individual might function well using AI helpers to compensate for lack of knowledge. But when novel challenges arise that the AI isn't tuned for, or when current experts retire, there is a collapse—a gap where capable humans should be.

The diamond model of staffing might be efficient, but it is fragile. It lacks the resilience that comes from having a pipeline of rising talent who have been tempered by gradually increasing challenges. In the long run, removing the bottom of the ladder doesn't just hurt those who can't get on—it destabilises the entire structure above.

The Economic Reality of Cognitive Fitness

People often dismiss these concerns with the familiar refrain: "Memorisation is outdated: we have Google for facts. Objections like this fundamentally misunderstand how expertise actually develops and, more critically, ignore the economic reality that cognitive fitness now determines career survival.

The "understanding without memory" fallacy is perhaps most dangerous. Try explaining calculus to someone who hasn't memorised basic arithmetic, or asking a lawyer to craft a complex argument without knowing legal precedents by heart. True understanding emerges from the interplay between memorised knowledge and reasoning—not from the ability to quickly look things up.

More urgently, the economic bar for human value is rising rapidly. As AI handles routine cognitive tasks, the human premium shifts to capabilities that require deep, internalised knowledge: novel problem-solving, creative synthesis, ethical judgment, and the ability to effectively direct AI systems. The choice isn't between memorisation and understanding; it's between cognitive independence and professional obsolescence.

Consider what's already happening in professional services. Law firms like Allen & Overy have deployed AI assistants to all 3,500 of their lawyers, letting them generate drafts and research at lightning speed (Reuters, 2023). The value isn't in the tool; it's in the expert wielding it.

I am not advocating for memorisation instead of thinking, but for memorisation as the foundation of thinking**.** Without a rich knowledge base stored in memory, critical thinking frameworks have nothing to operate on. Without automatic recall of basic facts and procedures, working memory gets overwhelmed trying to juggle low-level details while attempting higher-order reasoning. The opposition to memory work isn't progressive, it's economically dangerous in an age where cognitive capability determines professional survival.

Building Educational Ladders for the AI Age

The trends described above paint a worrisome picture, but they are not irreversible. If the root of the problem is that we've devalued memory and overvalued external aids, the solution lies in rebalancing. We need to build "educational ladders"—structured pathways that deliberately integrate memory, retrieval practice, and simulated problem-solving as scaffolding for long-term cognitive strength.

The educational research provides a clear roadmap. Bloom's 2-Sigma Problem, demonstrated through decades of studies, shows that individualised, mastery-based instruction produces learning gains of two standard deviations—meaning the average tutored student performs better than 98% of students in conventional classrooms (Bloom, 1984). This isn't about returning to old methods; it's about scaling what we know works: deeply individualised learning that ensures genuine mastery before advancing. The challenge has always been implementation at scale—which is precisely where thoughtfully designed AI systems can transform education from an industrial sorting mechanism into a true talent development engine.

In educational policy and practice, curricula can reintroduce spaced retrieval practice and cumulative review of core knowledge. Research shows that actively recalling information is one of the most potent ways to cement learning (Karpicke & Roediger, 2008). Simple techniques like low-stakes quizzes or flashcard reviews can significantly boost retention compared to passive review. We should encourage teaching methods that force students to remember and apply knowledge from previous lessons, not just cram for tests and forget. We are creating a mental library that students can draw upon for genuine critical thinking.

Knowledge and skills acquisition should be treated as a stepwise progression again, rather than all-at-once exposure. Educational ladders might involve requiring students to write a research essay without AI assistance before teaching them how to use AI as a helper, so they appreciate and retain the underlying writing skill. Or in coding education, having participants manually code small algorithms and debug them before introducing heavy use of AI code generators, so they internalise basic programming logic. The idea is to simulate the conditions of expertise development in a scaffolded way: create safe environments where learners must rely on their own knowledge and effort, thus strengthening their cognitive muscles. Once that foundation is laid, then bring in the power tools to accelerate them further.

In professional settings, the concept of educational ladders means investing in training programs and junior roles even if AI could theoretically handle the work faster. Consider it an investment in resilience: by letting a human learn the ropes (perhaps with AI as a guide, not a crutch), the organisation gains adaptive, creative problem-solving ability that no static automation can provide. Some forward-thinking companies are already rotating new hires through "old-school" assignments to ensure they grasp fundamentals before granting them AI tools to boost productivity.

Underlying all these reforms is a philosophical shift: recognising that human memory and knowledge matter immensely, even in the AI era. As Barbara Oakley and colleagues argue, the optimal path forward is not to reject technology, but to use it wisely. They advocate strategies that "harness technology as a complement to – rather than a replacement for – robust human knowledge." This means encouraging students and workers to first attempt problems using their own reasoning before turning to AI, and using AI more as a tutor or to double-check work rather than as a first-pass solution.

Cognitive Independence or Professional Obsolescence

The cognitive crisis and economic transformation are converging into a perfect storm. Just as human intelligence begins declining, AI advancement is raising the bar for human relevance. The traditional apprenticeship model—where novices climbed a staircase of gradually increasing challenges—will be replaced by an "AI Slide" where the bottom step simply disappear, preventing learners from climbing. Firms, under economic pressures, will choose efficiency over expertise development, creating immediate competitive advantage while quietly undermining their future talent pipeline.

We need to secure our cognitive futures. Countries and institutions that rebuild cognitive fitness through memory-based learning, retrieval practice, and genuine mastery will develop workforces capable of directing AI rather than being displaced by it. Those that continue down the path of cognitive offloading will find themselves with populations unprepared for an economy where human value depends on irreplaceable cognitive capabilities.

The Flynn Effect's reversal is an early warning system. It tells us that something in our approach to learning and thinking has gone awry at a fundamental level. By reintegrating memory training, practice, and staged skill-building into our education systems and workplaces, we can fortify the cognitive foundation of the next generation. We can build ladders alongside our slides. We can ensure that the astonishing new AI tools lift us higher rather than cause us to slip downward.

The age of Artificial Intelligence, ironically, confronts us with a very old truth: our intelligence need exercise. Knowledge that lives in our own heads, skills honed to automaticity, and rich schemas of understanding are not antiquated relics; instead, they are the bedrock of human intellect. If we rebuild that bedrock, we can halt the cognitive collapse and perhaps even restart the upward climb of human intelligence.

The choice is stark: reclaim our memory, rebuild our mental strength, and rise above our artificial partners—or slide toward a future where human intelligence becomes a relic of history. It's time to reclaim our memory, retrain our minds, and rethink education so that human and artificial intelligence can grow together, rather than at the expense of one another. As we navigate this balance, we must remember that the ultimate goal of education is not just to know more, but to think better—and thinking better begins with what we carry inside our minds.


References

Bloom, B. S. The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, vol. 13, no. 6,, pp. 4–16. JSTOR. DOI: https://doi.org/10.2307/1175554.

Bratsberg, B., & Rogeberg, O. (2018). Flynn effect and its reversal are both environmentally caused. Proceedings of the National Academy of Sciences, 115(26), 6674-6678. DOI:10.1073/pnas.1718793115

Business Insider. (2024, April). AI job cuts: Finance Wall Street investment banking analysts hiring. Retrieved from https://www.businessinsider.com/ai-job-cuts-finance-wall-street-investment-banking-analysts-hiring-2024-4

Flynn, J. R. (2009). Requiem for nutrition as the cause of IQ gains: Raven's gains in Britain 1938–2008. Economics and Human Biology, 7(1), 18-27.DOI: 10.1016/j.ehb.2009.01.009

Going Concern. (2023). AI is moving so fast PwC couldn't even commit to a year-long training program. Retrieved from https://www.goingconcern.com/ai-is-moving-so-fast-pwc-couldnt-even-commit-to-a-year-long-training-program

Hindustan Times. (2024). JP Morgan CEO Jamie Dimon predicts 3.5-day work weeks thanks to artificial intelligence. Retrieved from https://www.hindustantimes.com/htcity/jp-morgan-ceo-jamie-dimon-predicts-3-5-day-work-weeks-thanks-to-artificial-intelligence-living-the-ai-dream-101732625573178.html

Karpicke, J. D., & Roediger, H. L. (2008). The critical importance of retrieval for learning. Science, 319(5865), 966-968. DOI:10.1126/science.1152408

LexFusion. (2024). Education in the age of Gen AI: The old way of training law firm associates just doesn't work. Retrieved from https://www.lexfusion.com/post/education-in-the-age-of-gen-ai-the-old-way-of-training-law-firm-associates-just-doesnt-work

Oakley, B., Johnston, M., Chen, K. and Jung, E. & Sejnowski, T. (May 11, 2025). The Memory Paradox: Why Our Brains Need Knowledge in an Age of AI. Retrieved from SSRN: https://ssrn.com/abstract=5250447 or http://dx.doi.org/10.2139/ssrn.5250447

Reuters. (2023, February 15). OpenAI-backed startup brings chatbot technology to first major law firm. Retrieved from https://www.reuters.com/legal/transactional/openai-backed-startup-brings-chatbot-technology-first-major-law-firm-2023-02-15

Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776-778. DOI: 10.1126/science.1207745

Sundet, J. M., Barlaug, D. G., & Torjussen, T. M. (2004). The end of the Flynn effect?: A study of secular trends in mean intelligence test scores of Norwegian conscripts during half a century. Intelligence, 32(4), 349-362.

Teasdale, T. W., & Owen, D. R. (2008). Secular declines in cognitive test scores: A reversal of the Flynn Effect. Intelligence, 36(2), 121-126.

Trahan, L. H., Stuebing, K. K., Fletcher, J. M., & Hiscock, M. (2014). The Flynn effect: A meta-analysis. Psychological Bulletin, 140(5), 1332-1360.

© Oisín Thomas 2025