Why Test Scores Are Dropping in Developed Countries: Demographics, Not AI or Cell Phones
Woke intellectuals will tell you what you want to hear: worse teachers, cell phones, COVID, and AI; these have modest impacts compared to the real-time human capital shift.
This doesn’t surprise me at all… and it shouldn’t surprise anyone who understands evolutionary trait distributions and isn’t fully brainwashed by Blank Slate anti-hereditarian shell games.
Cue the woke intellectual rationale:
Teacher quality is getting worse!
COVID really set these kids back!
Cell phones are clearly the cause!
Kids are losing critical thinking skills cuz of AI!
Schools aren’t getting enough funding!
These aren't entirely wrong — they're just secondary. Teacher quality has taken a hit (DEI prioritization hasn't helped), COVID accelerated everything-digital, and kids are overstimmed on caffeine and distracted by screens. But none of these explain the core pattern.
If you're working with students who have lower innate intellectual potential, the best move is probably sports and exercise for half the day, then basic reading, writing, arithmetic once they're burnt out and willing to focus — zero cell phones. We already know One Child Per Laptop was an epic failure — this is a genetic capacity (substrate) issue.
"Schools aren't getting enough funding" is the most moronic take possible. Schools are massively overfunded. The "best schools" merely correlate with high-IQ kids who get schools well-funded because they were born to high-IQ parents. When we "diversify" top schools via legal mandates — test scores tank, teachers quit, parents pull their kids — and wokes wonder: What went wrong?! Must be COVID and cell phones.
My guess is you could axe the budgets of the “best schools” to almost nothing and if you kept the kids the same, they’d still crush it academically because it was never about the money.
Test Scores are Falling in Developed Countries
The underlying data is brutal.
The OECD’s PISA 2022 results show the OECD math average falling a record ~15 points (489→472) between 2018 and 2022 — triple any previous decline. Reading dropped 10 points, double the prior record.
The PIAAC 2023 results for US adults tell a similar story with important caveats: literacy fell ~12 points from 2017 to 2023, with ~28% now at Level 1 or below — functionally unable to use a library search engine.
White adults at lowest literacy rose from 12% to 16%; Hispanic adults at Level 1 jumped from 31% to 45%. But the PIAAC comparison is methodologically messy — the 2023 round switched to all-tablet, changed scoring, and for the first time included respondents who couldn’t speak English or Spanish via a “Doorstep Interview” protocol.
With 1 in 4 Americans now first- or second-generation, the “within-group” Hispanic decline likely reflects a compositional shift; more recent immigrants in the 2023 sample, not the same Hispanics getting dumber. PIAAC is directionally supportive of the thesis, not dispositive.
Finland, the global education darling, collapsed 64 points in math from its 2006 peak (548→484). Germany dropped 39 points. Norway fell 33 in a single cycle. The Netherlands shed 26.
Conventional explanations: COVID, smartphones, TikTok brain, bad teachers, AI offloading, insufficient funding — are secondary accelerants.
They’re the explanations you’re supposed to say out loud, what intellectuals and policymakers offer because they’re socially acceptable, imply fixable problems, and keep the gov money flowing... they’re what everyone wants to hear.
Are Cell Phones the Problem?
France banned phones in 2018 — before PISA 2022 was even administered — and posted its lowest math score in PISA history.
A King’s College London analysis found countries with more bans actually scored lower — every 10% increase in schools banning phones correlated with a 9.4-point PISA decrease.
Even in OECD schools with bans, 29% of students still use phones daily.
Enforcement is downstream of the same population traits that produce the test scores: conscientiousness, impulse control, respect for authority.
The variable isn’t the phone. It’s who’s holding it.
Is AI (Cognitive Offloading) the Problem?
Some allege that AI (cognitive offloading) may be to blame for the massive decline in test scores… but that makes little sense as a primary driver.
Why?
PISA 2022 was administered in spring 2022 (October–November in the US).
ChatGPT launched November 30, 2022.
Claude and Bard (now Gemini): March 2023.
The big drop-off from 2014 to 2022 occurred before AIs launched.
NAEP 2022 was administered January–March 2022, 8 months before ChatGPT — math down 8 points, largest drop in NAEP history, entirely pre-AI.
NAEP 8th grade reading at the 10th percentile had fallen 10 points between 2013 and 2019 — before COVID, before AI, before any of it.
Japan and Korea have the same smartphones, TikTok, and AI access. Japanese teens average 4–6 hours/day online. Both countries’ scores held or improved.
The answer has to be something present before 2022 that differs between countries that declined and countries that didn’t.
Demographic changes
The primary driver of test score decline across the developed world is demographic change; replacement of higher-scoring populations with lower-scoring ones through unselected immigration and differential fertility.
The secondary driver is dysgenic fertility for IQ within all groups; the least cognitively adept (within all racial/ethnic groups) are having the most kids.
Other accelerants include: cell phones, COVID, teacher quality erosion, anti-intellectual curricula, and stimulants (Yerkes-Dodson imbalance; suboptimal for focus).
The Natural Experiment: Homogeneous Countries vs. Diversifying Countries
If demographics change is the primary driver behind test scores plummeting in developed countries, there’s a clear prediction: high-performance countries that remained demographically homogeneous should have roughly held their scores, while countries undergoing rapid demographic change should have crashed.
That’s pretty much what happened.
Eight systems — all gained or held steady while the OECD average cratered. They have the same smartphones, TikTok, and AI access as the West — Japanese teens average 4–6 hours/day online, 70% use TikTok — and they had COVID.
Yet only 5% of Japanese students report phone distraction in math class vs. 30% OECD-wide.
The one variable these countries don’t share with the declining West is mass unselected immigration.
Every time PISA drops, think tanks ask: “What can we learn from Japan, Singapore, Estonia?”
Nobody asks the obvious question: What if you don’t need their methods — just their people?
Japan didn’t gain because it discovered a pedagogical secret.
Estonia doesn’t outperform Germany at half the spending because Estonian teachers are twice as good. The variable is DNA.
Estonia maintains mostly Estonian ethnicity, whereas Germany heavily imported non-German ethnicities.
Now compare to the countries that diversified rapidly:
Germany — immigrant share doubled in one decade; native-immigrant gap is 59 PISA points (32 after SES controls). The 1.1M refugees from 2015 came from Syria (~83 IQ), Iraq (~87), Afghanistan (~84).
Sweden — 11% first-generation in PISA sample; top sources: Syria, Afghanistan, Somalia, Eritrea, Iraq. Sweden’s own PIAAC data: college-educated migrants from Arab states and Sub-Saharan Africa scored at the same numeracy level as low-education native Swedes. 76% of immigrant students don’t speak Swedish at home. African-origin students have the lowest grades and have declined over time.
Finland — smaller immigrant share but steepest decline of any OECD country (−64 from peak). Same origin profile as Sweden: Somalia, Iraq, Afghanistan.
The UK — least trustworthy data. Immigrant students rose 13% to 20% but PISA classifies second/third-generation Pakistanis, Somalis, and Caribbeans as “native” while current immigrants skew toward selected Indians and Chinese. GCSEs show convergence but had g-loading halved by Blair-era reforms. Every other measure — CAT3, PISA, university outcomes — shows substantial gaps. Cremieux confirmed: same sample, GCSE parity but significant IQ gap.
Canada — the “selective immigration” everyone praises. 34% immigrant PISA sample. Math dropped 35 points since 2003 — worse than Germany. Manitoba fell 58. Canada’s “points system” is ~58% “economic class” including unscreened spouses and children; the actual screened applicant is under a third of intake. Even that is gameable — 20,000 Indian students enrolled in 2024 without attending class.
The correlation is clear. AI, phones and COVID cannot explain this gradient.
The US: Arithmetic That Nobody Wants to Do
The compositional math in the US is straightforward.
Non-Hispanic White share (Census Bureau):
69.1% (2000) → 57.8% (2020). White population actually declined in absolute numbers 2010–2020 — first time in US history.
In public schools: White students dropped from 61.2% to 44.7%; Hispanic nearly doubled from 16.4% to 28.7%.
NAEP scores by race (8th grade math, 2022): White ~275, Hispanic ~257, Black ~244, Asian ~295. These gaps map onto the Roth et al. (2001) meta-analysis of 6.25 million test-takers (B-W gap: 1.1 SD; H-W: 0.72 SD).
The arithmetic: Replace 16.5 percentage points from a group averaging ~275 with one averaging ~257. Mechanical effect: ~3–4 NAEP points. Add English Learner growth and within-group compositional shifts, and demographics plausibly accounts for 4–6 of the ~9-point 8th grade math decline from 2013–2022.
Tom Loveless documented this precisely: 2009–2019, national 4th grade NAEP math was flat while every subgroup gained. Simpson’s Paradox — the aggregate hides the compositional shift.
The US immigration system is ~two-thirds family-based, ~17% employment-sponsored, plus 11M+ unauthorized.
The Swiss natural experiment is the cleanest causal evidence: switching from low-skill guest workers to EEA free movement improved immigrant PISA scores 43 points in a decade — Cattaneo & Wolter (2012) found ~75% was due to changed immigrant characteristics. Not better schools. Different people.
Richwine (Harvard, 2009): third-generation Hispanic immigrants still ~0.5–0.75 SD below the White mean. Environmental uplift produces gains, not convergence.
The Double-Whammy: Replacement + Dysgenic Fertility
Two simultaneous forces are pulling the cognitive mean of every Western nation downward, and no educational reform will reverse either.
Between-group replacement: Higher-scoring populations (below-replacement fertility) are shrinking while lower-scoring immigrant-origin populations (above-replacement fertility) are growing.
Within-group dysgenic fertility: In every racial and ethnic group, the least cognitively able are having the most children.
US fertility by education (2019 CDC/NCHS): no diploma TFR 2,433 vs. bachelor’s 1,396 — a 1.74x fertility advantage for the least-educated. Since education correlates with IQ at r ≈ 0.55–0.65, this is direct dysgenic pressure.
By race, the pattern is a near-perfect inverse correlation between cognitive scores and fertility:
Asian — NAEP 8th grade math ~295, TFR ~1,519 (lowest fertility, highest scores)
White — NAEP ~275, TFR 1,610
Black — NAEP ~244, TFR 1,774
Hispanic — NAEP ~257, TFR 1,939 (highest fertility, near replacement)
The two highest-scoring groups have the lowest fertility. The two lowest-scoring are nearest to or above replacement. Within the White population, the most educated women (bachelor’s+, TFR ~1.4) are being outbred nearly 2:1 by the least educated (no diploma, TFR ~2.4). This pattern holds within every group.
Kong et al. (2017, PNAS), using deCODE genetics on 129,808 Icelanders, showed polygenic scores for educational attainment declining ~0.30 IQ points per decade — genetic deterioration documented at the genomic level in one of the most homogeneous populations on earth.
A study many cite to rebut demographic changes is Bratsberg and Rogeberg (2018) — within-family IQ decline in Norwegian brothers.
Three problems: (1) the birth order confound (first-borns score 2–3 points higher — you can’t separate this from “born later in time”), (2) the paternal age effect (~1–2 de novo mutations per year — a genetic mechanism), and (3) magnitude (0.2–0.3 points/year vs. Germany’s 59-point native-immigrant gap). B&R is a footnote cited as if it refutes a tsunami; the situation is worse than demographics alone.
High IQ Kids = High Test Scores
Educational quality is downstream of population quality, not the other way around.
The causal chain runs like this:
High-IQ parents cluster in affluent suburbs through sorting on income, which is itself substantially determined by cognitive ability.
They fund schools through high property taxes, creating per-pupil spending that dwarfs urban districts (even adjusting for cost of living).
Good teachers want to teach there because the students are teachable, the parents are engaged, the classrooms aren’t behavioral management nightmares, and the pay is competitive.
Students absorb the material because they have the cognitive substrate to do so, reinforced by home environments where educated parents supplement schooling.
Test scores are high. Everyone attributes this to the great teachers and great schools.
But the causal arrow runs primarily from student quality → teacher quality → school quality, not the reverse.
Put the same “great teachers” in Baltimore or Detroit and scores aren’t gonna budge.
Finland is the proof case in reverse. Finland got elite teachers (top 17% of applicants, master’s degrees required) because it had a homogeneous, high-IQ population that was genuinely teachable.
Teaching was a high-status profession because the work was intellectually rewarding.
As Finland’s demographics shifted and classrooms became harder to manage, teaching enrollment has started to decline — the same pattern the US experienced decades ago, just on a lag.
In the US, education majors now score 28 points below the SAT average and have the lowest GRE Quantitative scores of any graduate field (149 vs. 161 for engineering).
Teaching enrollment is down 33% since 2010. Smart people with options don’t sign up for work that increasingly consists of behavioral management for students who lack the cognitive capacity to absorb the material.
The “good teacher” is mostly a byproduct of the good student — and both are a byproduct of the genetic stock of the parents and the community that funds the school.
AI-Assisted Learning Bifurcation
The same technology is producing radically different outcomes depending on the cognitive substrate of the user.
NAEP data: the 10th percentile of 8th-grade readers fell 10 points between 2013 and 2019 (before COVID) while the 90th percentile held steady.
Khan Academy’s Khanmigo exploded from 68,000 to 700,000 users in one year, with meaningful gains for engaged students.
But the Khan Academy CLO actually reports: some students engage exactly as designed, “answering questions and posing their own to deepen understanding.” Many others respond “I don’t know” or “Bro, IDK.”
High-ability kids use AI tutors, Claude, ChatGPT, and self-directed platforms to learn at accelerated pace, often routing around the school system entirely. A motivated 14-year-old with a 130 IQ can learn calculus or Python with no teacher and no per-pupil spending. The technology amplifies existing cognitive capacity.
Alpha School — Compresses all academics into two hours of AI-driven adaptive learning per morning. Their students learn 2.6x faster on average, top 20% at 6.5x, scoring 99th percentile on MAP across nearly all grades and subjects. First graduating class: 11 of 12 to four-year universities.
Mentava does the same for gifted early readers. AI-driven phonics with human coaching for high-ability kids whose needs aren’t met by age-normed classrooms. These schools work because $75K tuition selects from the right tail of the cognitive distribution, and AI removes the artificial pace constraint. The technology unleashes innate ability that was already there.
When Arizona greenlit Alpha’s charter model for public funding (“Unbound Academy”), Pennsylvania rejected the identical application: “untested and fails to demonstrate alignment.”
They’re probably both right. It works for Alpha’s population. It may not work for the population Pennsylvania is worried about; “fails to demonstrate alignment” is a laughable take but probably for the best.
Low-ability kids get the same technology and respond “Bro, IDK.” The tool won’t create cognitive capacity that isn’t there. Same technology, different outcomes, determined entirely by substrate. The schools where AI tutoring “works” were already producing good outcomes because of the selected students.
Spend More on Education = Money Incinerator
The US spends $16,500+ per pupil nationally, 38% more than the OECD average. Lighting money on the fire… lighting more on fire… and think that lighting even more money on fire will somehow eventually produce a positive ROI.
This is like giving midgets NBA-level coaching and training and thinking they’re gonna make the pros one day — or like effective altruists throwing money into developing countries and thinking the newly funded systems will remain sticky (rather than requiring infinite money).
The only solution is targeting biology: gene modification, embryo engineering, embryo selection, etc. — for higher IQ.
Estonia: ~$8,500/pupil → 1st in Europe, 6th globally in PISA math
Japan: ~$10,000/pupil → +9 points, now 1st in reading
The United States: ~$16,500/pupil → 34th in PISA math, declining every cycle
Estonia gets 2x the output per dollar because the median Estonian absorbs the instruction. Past a relatively minimal threshold, additional dollars have zero ROI because the constraint is biology.
The Kansas City experiment (1985–1997) settled this empirically.
A federal judge invited educators to “dream” — forget about cost.
Over 12 years, nearly $2 billion: 15 new schools, Olympic swimming pool with underwater observation room, TV/animation studios, robotics lab, 25-acre wildlife sanctuary and zoo, model United Nations with simultaneous translation, field trips to Mexico and Senegal.
Student-teacher ratio fell to 12–13:1, lowest of any major district. Highest per-pupil spending of the 280 largest districts.
Result: Test scores unchanged. Black-white gap unchanged. Dropout rate went up. Nonwhite enrollment rose from 73% to 80%+. “For decades, critics said ‘You can’t solve educational problems by throwing money at them.’ Educators replied, ‘No one’s ever tried.’ In Kansas City, they tried.”
Washington DC: ~$31,000/student — highest in the nation. 8th graders score two grade levels below national norms. DC schools are 64% Black, 19% Hispanic. Results track the population, not the expenditure.
LeBron James’ I Promise School (Akron): $1.4M/year from the Foundation on top of district funding; additional tutors, smaller classes, extended days/year, summer tutoring, wraparound services, guaranteed free college tuition. Everything the education establishment says it needs.
Result: 2 out of 75 seventh-graders passed state math (2.7% vs. 50% state average). Inaugural class: 0% math proficiency three years running. 125th lowest of 3,318 Ohio schools.
Fordham Institute found I Promise students performed worse than matched peers at regular Akron schools. The NYT celebrated the school in 2019. They haven’t written about its test scores since.
Jalen Rose Leadership Academy (Detroit): 97% graduation rate. Math proficiency: 2.42%. You can graduate every kid, you cannot make them proficient.
Head Start: ~0.2 SD initial IQ gains fade out completely by first grade, undetectable by third.
Perry Preschool: same fade-out by age 8. The most replicated finding in educational research — and the one every policy discussion ignores.
The OECD finds spending explains ~54% of performance gaps up to ~$75,000 cumulative spending (ages 6–15), but above that threshold the relationship vanishes. The US is far above.
The ROI is negative: Every dollar poured into students who can’t absorb instruction is diverted from students who can. Kansas City spent $2 billion and moved the needle zero. LeBron provided everything money can buy and got 2.7% proficiency. Money buys results only when the substrate can convert it into learning. When it can’t, you get Olympic swimming pools and 0% proficiency.
The Elites Tell You What You Want to Hear… Not What You Need to Hear
No economist, education researcher, or policy analyst can say “test scores are falling because we imported millions of people with lower cognitive potential and the least capable members of every group are having the most children” and keep their job.
So they find real but secondary mostly bullshit factors: COVID, phones, teacher shortages, AI offloading — and present those as the story.
The audience smiles, nods, agrees. An op-ed gets published. Gov comes up with a big “AI education spending initiative to promote equity” (or something).
Not once does anyone mention the demographics of the tested populations dramatically changed.
And never will they highlight Japan (with cell phones, AI, TikTok, YouTube, multitasking, gaming, etc.) improved during the same time horizons because it didn’t change its demographics.
The New York Times will never run “It’s the Demographics, Stupid” because the implications are socially unacceptable — even if they’re empirically unavoidable and obvious to truth-seekers.
What will actually fix test scores?
Besides making all tests absurdly easy so that everyone “does well”; this is test inflation… and the worst aspect is that declines are happening with rampant test inflation.
If you are serious about fixing this… we need rapidly scaled intelligence upgrades for humans that didn’t evolve to function in the recent cognitively-demanding times.
1. Highly selective immigration. This is the single highest-leverage policy. The US family-reunification system and Europe’s humanitarian intake produce the opposite. Every percentage point of the population shifted from high-scoring to low-scoring groups moves the national cognitive mean downward.
2. Stop wasting money on “closing gaps” that are biologically constrained. Education spending should be redirected from futile attempts to equalize outcomes toward efficiency and discipline. For lower-performing populations, the focus should be on vocational training, behavioral discipline, and practical skills — not on forcing academic content that isn’t being absorbed. For high-performing populations, remove obstacles and let AI-assisted technology amplify their ability.
3. Phone/social media restriction in low-performing schools. The highest-ROI near-term intervention. PISA 2022 data shows 15 points of benefit. Japan’s low distraction rates (18% vs. 59% OECD) correlate with maintained scores. This doesn’t fix the structural problem but slows the environmental bleed in low performers.
4. Bioenhancement — the only real long-term solution.
Embryo selection with Herasight et al.: current technology could plausibly produce 5–15 IQ point gains per generation across all groups. This raises the cognitive floor for everyone. Eventually use WGS… and eventually embryo engineering.
Somatic gene editing for intelligence/cognition once the relevant variants are characterized and delivery mechanisms mature. I have a protocol that I will release soon and by the 2030s-2040s people can test if they want.
Age reversal (Operation Senolysis) to extend the productive lifespan of high-human-capital individuals. Every high-IQ person who dies at 75 instead of 150 is an enormous loss of productive capacity and institutional knowledge.
These are the only interventions that address the underlying biological constraints rather than trying to paper over them with spending that demonstrably doesn’t work.
The Human Capital Cliff Is Here
The developed world built institutions, economies, and technological infrastructure on a population with a certain cognitive profile — and that population is being replaced via unselected immigration and differential fertility.
The new population has a different profile, and predictably, test scores are falling.
You can take phones away, fire bad teachers, hire quality teachers, pour another $200 billion into schools and none of it will reverse the tide. Why? The tide is about the human capital of subsequent generations.
The only serious move to prevent a Dark Age in developed countries is: age reversal + embryo engineering + somatic IQ upgrades in adults + AGI/ASI robot diffusion.
Most other things are just rearranging deck chairs.
Note: This is why I think Brad Gerstner’s “Invest America” idea is well-meaning but misses the mark. It increases government entitlements (spending) at a time when the U.S. is in massive debt (with massive interest on that debt) and assumes the new demographics won’t just “cash out” and blow the investment the minute they turn 18. Does skin in the game matter if it isn’t your skin? And does it work if the median doesn’t have the innate ability to delay gratification? I suspect it’ll have zero effect on the growing demand for socialism in the younger, more Hispanic generation.
