The extinction of human skills. Nobody takes inventory

What we stop doing disappears. Mental arithmetic, spelling without a checker, the memory of phone numbers, orientation without GPS —each of them atrophied in a decade or less for most of the urban population. The evidence isn't only intuitive. Eleanor Maguire and colleagues published in 2000, in PNAS 97, a neuroscientific study on London taxi drivers: those who had learned "The Knowledge" —exhaustive memorization of streets— showed larger posterior hippocampi. Later follow-up studies showed the symmetrical part: when taxi drivers stop memorizing, the effect reverses. Brain plasticity is real and bidirectional. Cognitive skills that stop being exercised atrophy, not metaphorically but organically. AI is going to accelerate the list: arithmetic, autonomous writing, sustained reading, critical analysis, linguistic formulation from scratch. What atrophies isn't easily recovered. And nobody takes inventory.

The neuroscientific evidence

It's worth being precise with the evidence, because it's frequently cited imprecisely.

Maguire et al. (PNAS 97, 2000) — Navigation-related structural change in the hippocampi of taxi drivers. Compared the hippocampi of London taxi drivers with a control group. Found greater volume in the posterior hippocampus in taxi drivers; a correlation with years of experience. Follow-up studies (Maguire et al. 2003, 2006) confirmed and refined it, showing that the effect was associated specifically with active route learning, not with other variables of the job.

Maguire et al. (PNAS 2006) and later work —Recall and recognition memory in taxi and bus drivers. Compared taxi drivers (flexible spatial use) with bus drivers (fixed routes). The former showed better spatial memory; the latter didn't show the hippocampal effect. The refined conclusion: the active exercise of the spatial skill produces the structural change, not the driving profession in general.

Wolbers & Hegarty and others (2010s-2020s) — research on GPS use. Studies with young adults comparing active navigation (route decision, environment reading) with passive navigation (following GPS directions) showed lower hippocampal activation in passive navigation, with effects on subsequent spatial memory. The studies are consistent with the disuse-atrophy hypothesis.

Sparrow, Liu & Wegner (Science 333, 2011) — Google Effects on Memory. Documented that when participants expected to be able to look up the information later, their memory of the content was worse, while their memory of where to find the information was better. The conclusion: the brain reallocates resources according to the availability regime of external information.

Risko & Gilbert (Trends in Cognitive Sciences, 2016) — Cognitive offloading. A theoretical framework on how the systematic use of external cognitive resources —notes, reminders, GPS, assistants— modifies internal cognitive resources. A distinction between occasional offloading (no persistent effect) and systematic offloading (with an effect on baseline capacities).

MIT Media Lab Your Brain on ChatGPT (preprint, 2025). An EEG study on students writing essays with an LLM, with a search engine, or with no assistance. Preliminary results suggest reduced neural connectivity and executive activation in LLM use versus unassisted writing. The study is a preprint, not peer-reviewed in final form, and its interpretation admits nuance; the direction is consistent with the literature on cognitive offloading.

The five sources are consistent on one substantive idea: the brain adapts its structure and function to the use regime. What's exercised is maintained or grows. What's delegated atrophies.

The historical catalogue

It's worth listing documented or widely reported atrophies in recent decades, with the caution that the evidence is uneven for each case:

• Mental arithmetic. With calculators since the seventies and spreadsheets since the eighties, the practice of figuring in your head —even order of magnitude— has contracted massively. Studies on numeracy report a decline in some developed countries; others show stability. Significant cultural variability.
• Spelling without a checker. Word processors with autocorrect since the nineties; predictive correctors on phones since the 2000s. Unassisted spelling is reported as worse in younger cohorts. The evidence is uneven and entangled with pedagogical changes.
• Memory of phone numbers. Before digital address books, most people remembered dozens of numbers. Today, most don't remember more than five. The change is massive and almost without public discussion.
• Orientation without GPS. As seen in Maguire and successors, the skill of orientation with maps, urban landmarks and spatial memory has atrophied in cohorts that grew up with mobile GPS.
• Sustained reading. Wolf, in Reader, Come Home (HarperCollins, 2018), documents how the regime of fragmented digital reading erodes the capacity for deep, sustained reading. The neuroscientific and pedagogical evidence converges.
• Capacity for waiting and for boredom. The continuous availability of digital stimulus reduces tolerance for silence and emptiness, with consequences on creativity, reflection and emotional development. Documented by cognitive and educational psychology.

Each case admits nuance. The aggregate pattern is clear: there are cognitive skills in regression over the last three decades, in temporal correlation with the growing availability of tools that made them unnecessary for everyday use.

The next ones on the list

As of 2026, the cognitive skills at risk of accelerated atrophy through mass AI use, identifiable without speculating too much:

• Autonomous writing. Writing from scratch, without an assistant, without autocomplete, without rephrasing. The microsecond of thought between intention and word that autocomplete eliminates.
• Structured analysis. The capacity to break down a problem, order its parts, rank them, synthesize a conclusion. Replaceable by an assistant with summaries and proposals.
• Long, sustained reading. Already in decline from the prior digital context; the availability of AI summaries accelerates the pattern.
• Order-of-magnitude mental arithmetic. The skill of quick judgment about whether a reported figure is plausible. Replaceable by a query to an assistant, but it loses the error-detection function.
• Formulation of one's own questions. When the assistant suggests questions or rephrases the user's, the capacity to formulate an original question is exercised less.
• Memory of basic facts. Dates, names, frequently used data. Replaceable by immediate query; consequences for reasoning that requires the rapid combination of several facts.
• Decision under uncertainty without support. The practice of deciding with incomplete information, without consulting an external source, shrinks with an assistant available.

The list isn't exhaustive. Each item admits the nuance: the effect depends on intensity of use, on age, on prior education, on the presence of an alternative practice.

The asymmetry of cost

There's an operational property of cognitive atrophies worth being precise about. The atrophy is gradual, asymptomatic and cheap in its production. Recovering the atrophied skill is slow, costly and, in some cases, partially irreversible.

The asymmetry has an economic consequence. In a market or system without an external cost, individuals delegate skills to external tools until something changes: the tool stops being available, the context changes, the profession demands the skill. When that moment comes, recovering the atrophied skill takes years. In between, people live with an operational dependence on the tool.

The asymmetry structurally favors atrophy. Without a conscious policy that exercises specific skills, the natural direction is to delegate the delegable.

The inventory that doesn't exist

There's a serious political question almost nobody asks. Which human cognitive skills are an operational condition for democratic participation, individual autonomy, social resilience? If they were identified, they could be defended specifically —in the educational curriculum, in cultural policies, in tool design. The silence around the question means the atrophy happens without discussion.

Compared with what we do take inventory of:

• There's an inventory of biological species at risk (IUCN). There's public investment to protect them.
• There's an inventory of intangible cultural heritage (UNESCO). There are preservation programs.
• There's an inventory of endangered languages. There are revitalization efforts.
• There's an inventory of traditional knowledge in some contexts.

What doesn't exist: a systematic inventory of the human cognitive skills society considers worth preserving. Neither at the international level (UNESCO), nor national, nor professional. The question hasn't been seriously posed.

This is personal opinion, but documentable: the asymmetry between the magnitude of the aggregate cognitive change underway and the absence of an institutional framework to monitor it is probably one of the most significant lags between reality and institutional response in 2026. The consequences will accumulate over decades before manifesting as a politically salient problem.

The question of what's gained

To avoid falling into a purely declinist reading, it's worth recognizing the symmetry. What's lost when delegating skills coexists with what's gained by having free cognitive capacity available. If mental arithmetic atrophies, you gain capacity for reasoning about the problem. If memory of facts contracts, you gain capacity to process relationships between facts. If sustained reading shrinks, you gain capacity to process larger volumes of fragmented information.

The gain is real but not automatic. It requires the brain to reallocate what's freed to something else of value. If the reallocation is going to consume more fragmented content, navigate feeds, receive more digital stimulus, the gain vanishes. The empirical question of whether delegation produces productive reallocation or derived consumption is central, and the available evidence suggests the second scenario is at least as frequent as the first.

Wolf, Carr and the convergence

Maryanne Wolf and Nicholas Carr are two authors who have documented the phenomenon from different angles: Wolf from the neuroscience of the reading brain (Proust and the Squid, 2007; Reader, Come Home, 2018), Carr from the cognitive culture of the digital environment (The Shallows, 2010; The Glass Cage, 2014). Their theses converge: the technological mediation of central cognitive operations reorganizes the brain and produces aggregate effects on cognition, attention and judgment that are not neutral.

The academic critique of both authors has been vigorous. They've been accused of selecting evidence, of extrapolating prematurely, of adopting a declinist tone. The critiques have merit in concrete details. What survives the critique is the central thesis: the direction of cognitive change isn't neutral, and systematically delegating cognitive operations has measurable consequences that deserve systematic study.

The society without measurement

The consequence worth naming is the aggregate absence of measurement. A society without an inventory of its cognitive capacities in use operates blind to its own human capital. It doesn't know what it loses before losing it, doesn't gauge the cost of the loss, doesn't design preventive intervention. When the cost manifests —in productivity, in autonomy, in resilience to AI service interruptions, in democratic capacity— the loss is already consummated and reversing it is more expensive than preventing it.

Systematic measurement would exist if there were public funding for longitudinal research, institutions charged with inventorying skills, metrics comparable across cohorts. None of those elements exists at an adequate scale in 2026. The political pressure to create them is modest. The asymmetry between present cost (modest, individualized) and future cost (aggregate, partially irreversible) guarantees that inaction holds until the future cost becomes politically salient.

Definitions

• Brain plasticity: the nervous system's capacity to modify its structure and function in response to experience. Bidirectional: use strengthens, disuse atrophies.
• Cognitive offloading: the practice of delegating cognitive operations to external resources —notes, tools, AI systems. The theoretical framework of Risko and Gilbert.
• Disuse atrophy (cognitive): regression of cognitive capacities through lack of exercise. Documented for several specific skills.
• Posterior hippocampus: a brain region associated with spatial memory. The subject of the Maguire study on London taxi drivers.
• The Knowledge (London): the traditional exam for London taxi drivers that requires exhaustive memorization of streets and routes. A skill documented as producing hippocampal change.
• Deep reading (sustained reading): prolonged, deep, uninterrupted reading, with elaborative processing. A skill documented in regression in cohorts with intensive digital use.
• Numeracy: the capacity to operate with basic quantitative concepts in everyday contexts. A metric used in PISA and OECD studies.

References

• Maguire, E. A., Gadian, D. G., Johnsrude, I. S., Good, C. D., Ashburner, J., Frackowiak, R. S. J. & Frith, C. D. "Navigation-related structural change in the hippocampi of taxi drivers." PNAS 97(8), 2000.
• Maguire, E. A., Woollett, K. & Spiers, H. J. "London taxi drivers and bus drivers: A structural MRI and neuropsychological analysis." Hippocampus 16(12), 2006.
• Sparrow, B., Liu, J. & Wegner, D. M. "Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips." Science 333(6043), 2011.
• Risko, E. F. & Gilbert, S. J. "Cognitive Offloading." Trends in Cognitive Sciences 20(9), 2016.
• MIT Media Lab. Your Brain on ChatGPT. Preprint, 2025.
• Carr, N. The Shallows: What the Internet Is Doing to Our Brains. W. W. Norton, 2010.
• Carr, N. The Glass Cage: Automation and Us. W. W. Norton, 2014.
• Wolf, M. Proust and the Squid: The Story and Science of the Reading Brain. HarperCollins, 2007.
• Wolf, M. Reader, Come Home: The Reading Brain in a Digital World. HarperCollins, 2018.

Going deeper

• Norman Doidge. The Brain That Changes Itself. Penguin Books, 2007. An accessible synthesis of the literature on neuroplasticity; needed to understand why disuse atrophy is real and why recovery is possible though costly.
• Alvaro Pascual-Leone. Publications on brain plasticity and rehabilitation. Clinical research on how deliberate practice modifies the functional organization of the brain; applicable to the design of preventive interventions against atrophy.
• Daniel Schacter. The Seven Sins of Memory. Houghton Mifflin, 2001. The classic taxonomy of human memory dysfunctions; useful for identifying which types of memory are most vulnerable to offloading and which are robust.

You may also like

• Learning as accelerated evolution
• New cognitive skills
• The cost of thinking less
• What it means to know something today

Elsewhere

• IUCN — Red List of Threatened Species
• UNESCO — Intangible Cultural Heritage