What Does Humanity Pay Attention To?

Wikipedia is the closest thing humanity has to a collective brain. Every pageview is a vote — a tiny signal of what someone, somewhere, chose to learn about. Stack up billions of those signals and you get a portrait of what humanity actually cares about.

We pulled the top 1,000 most-viewed Wikipedia articles across five languages — English, Spanish, German, French, and Portuguese — covering roughly 2.5 billion internet users. Then we categorized every article, cross-referenced with national education data, and tested whether "smarter" countries read "smarter" things.

The answer is not what you'd hope.

Entertainment dominates everywhere. Across all five languages, celebrity news, films, TV shows, and pop culture consume 34–48% of substantive Wikipedia attention. English Wikipedia is the most entertainment-heavy at 48%.

Science, history, and the arts — the content that arguably enriches understanding — ranges from 8% to 14% combined, depending on the language. The average across all five is about 11%.

The cultural fingerprints are interesting. Portuguese and Spanish Wikipedias have high sports attention (28% and 21%) — football culture in action. German Wikipedia has the highest politics share (28%), likely driven by the 2025 federal election. But the entertainment floor is universal: at least a third of all substantive attention goes to pop culture, everywhere.

We computed a "depth ratio" for each language — intellectual content (science + history + arts) divided by passive content (entertainment + sports + taboo). Then we plotted it against each country's Human Capital Index.

There is no meaningful correlation. Spearman rho = −0.20, p = 0.75. France has the highest HCI but the lowest depth ratio. Brazil has the lowest HCI but a middle-of-the-pack ratio.

Wikipedia attention follows a textbook power law with R² > 0.93 in all five languages. But the concentration varies dramatically. In English, the top 100 articles get 32% of views. In French, the top 100 get 65%. In German, 60%. The power law shape is universal; the steepness is not.

English has the flattest distribution — attention is most evenly spread (Gini 0.38). French and German are most concentrated (Gini 0.63–0.68), driven by massive main-page traffic. But the shape is remarkably similar across all languages.

Carlo Cipolla's Basic Laws of Human Stupidity classifies all human actions into four quadrants based on whether they benefit or harm the actor and others. We applied this framework to attention itself:

Cipolla's third law states that the fraction of "stupid" people is constant across all groups — independent of education, wealth, or status. If that applies to attention, the ratio of passive to intellectual attention should be roughly the same everywhere.

The data partially supports this. "Helpless" attention ranges from 13% to 38% of views — a wide spread (CV = 38%). But the ratio is structurally consistent: for every 1 view of intellectual content, there are 2.4–4.5 views of entertainment across all five languages.

We pulled Google Trends data for conspiracy-related searches across the same five countries. The patterns differ sharply from Wikipedia attention:

Brazil leads in conspiracy search interest (avg 20.0 across all terms, 28.6 excluding zero-scoring ones), driven by "fake news" (97/100) and "illuminati" (67/100). The US has the most diverse conspiracy diet — flat earth, QAnon, deep state, illuminati all score above 30.

Notably, "5G conspiracy" has died to zero everywhere. "Fake news" dominates in every country except the US, where multiple terms compete.

Wikipedia might attract a particular kind of reader. Books require effort. Podcasts are niche. But YouTube? YouTube is where everyone goes. It's the world's second-largest search engine, and its trending page is the closest thing to a real-time vote on what humanity wants to watch.

We pulled the top 200 trending videos in all five countries via the YouTube Data API. YouTube assigns its own categories — Music, Gaming, Education, Science & Technology — so we don't need regex. The platform tells us directly what each video is.

The result is staggering. Across all five countries, 100% of trending videos are entertainment. Music dominates everywhere (34–66% of views), followed by Gaming and Film trailers. Not a single Education or Science & Technology video made the trending list in any country.

The cultural variations are there but minor: the US skews toward Gaming (39%), while Spain and Brazil are Music-heavy (62–66%). France leads in Film & Animation (26%), driven by movie trailers. But the macro picture is uniform: zero educational content trends anywhere.

Trending is algorithmic — it shows what YouTube pushes. But what happens when people search? We sampled 6,063 videos across five countries using the YouTube Data API, filtering for the most-viewed content published since 2024.

The US stands out. When people actively search, the US shows 27% of views going to content labelled "Education" or "Science & Technology" — compared to just 5–9% in Europe and Brazil. Entertainment still dominates everywhere (63–82%), but the gap in content tagged as educational is real and persists after methodological normalisation.

This categorisation problem is not unique to YouTube. The same issue applies to book data, where self-help, alternative medicine, and pop psychology sit alongside rigorous non-fiction under "education." Any taxonomy based on creator or publisher labels will conflate genuine intellectual content with content that merely borrows its aesthetics. We flag this as a cross-cutting limitation of the entire investigation.

YouTube is the extreme case. But how does the pattern look across all platforms we've measured? From the most frictionless (YouTube) to the most intentional (GitHub):

There's a friction gradient. YouTube trending (zero friction) is 100% entertainment. YouTube search (low friction — you type a query) drops to 63% in the US, 77% in Europe. Wikipedia (low-medium friction — you have to read) drops to 48%. Open Library (medium friction — you borrow a book) drops to 25%. GitHub (high friction — you need to code) is just 2%.

The more effort a platform requires, the less entertainment dominates. Podcasts are the odd case — US Apple Charts show 36% "educational," but this includes business and self-help. The pattern is consistent: ease of access predicts entertainment share.

Stack Overflow — arguably the most knowledge-intensive community online — shows the same hierarchy. Its most-viewed question isn't about algorithms or design patterns. It's "Why is processing a sorted array faster than processing an unsorted array?" (2M views). Fascinating, but it went viral because it's a puzzle, not because people needed the answer.

Meanwhile, Physics Stack Exchange's top question is about cooling coffee with a spoon. Philosophy's top question is whether math is invented or discovered. History's is about Hitler's strategic blunders. The attention economy rewards accessibility and narrative over depth, even in expert communities.

GitHub offers a unique window into developer attention over time. All-time, the most-starred repos are balanced: 34% educational (awesome-lists, tutorials), 20% AI/ML, 11% frameworks.

But filter to repos created in 2024 or later and the picture transforms: AI/ML surges to 60%, educational drops to 12%, and frameworks nearly vanish. Developer attention has undergone a phase transition — AI is now the dominant attractor.

Everything above is a snapshot — what people read, watch, and click now. But has it always been this way? Google's Ngram Viewer tracks word frequency across millions of books published since 1800. It's a 220-year record of what authors thought was worth writing about.

The plot twist: in books, science has always beaten entertainment. "Science" appears 6× more often than "entertainment" in published books — and the gap has been consistent for two centuries. "Technology" exploded from near-zero to its peak around 1990, then retreated. "Religion" dominated in 1800 but has been falling steadily for 200 years.

This is the opposite of what we see on YouTube and Wikipedia. Why?

The second chart reveals something subtler. "Knowledge" has held remarkably steady at 150–200 ppm for 220 years — it's the bedrock of published thought. "Learning" surged after 1970, tripling in frequency. But "curiosity" collapsed from 34 ppm in 1800 to 5 ppm by 1980, before a recent recovery.

"Pleasure" tells the starkest story: it fell by two-thirds from 1800 to 1980, then started climbing again in the internet era.

There's an elephant in the room. The original question behind this investigation was: does average human interest reflect average intelligence?

IQ, for all its controversy, follows a normal distribution with a mean around 100. About 68% of people score between 85 and 115. The tails — the very bright and the very limited — each contain roughly 2–3% of the population. This shape is one of the most replicated findings in psychology.

Now look at the attention data. Entertainment captures 34–48% of substantive Wikipedia views. Intellectual content — science, history, arts — gets 8–14%. For every one person reading about quantum mechanics, three to five are reading about a Netflix series. Stack Overflow's most-viewed questions are puzzles, not paradigms. The most-read books on Open Library are thrillers, not textbooks.

But here's where it gets more nuanced. Education doesn't shift the ratio. Germany (mean IQ estimates ~99–102) and Brazil (~87–89) show strikingly similar entertainment-to-intellectual ratios. France, with Europe's highest Human Capital Index, has the lowest depth ratio in our sample. If IQ determined attention, we'd expect a clean gradient. We don't see one.

This suggests something subtler: it's not that people can't engage with deep content. It's that at the population level, they don't choose to — regardless of ability. The psychologist Stanovich calls this the "dysrationalia" gap: the difference between intelligence (can you solve this?) and rationality (do you seek things worth solving?). IQ measures the engine. Attention data measures where people drive.

The bell curve of intelligence and the power law of attention are different distributions describing different things. But they converge on one insight: the average is where most of the mass lives, and the average human, given the choice, reaches for entertainment over enrichment. Not because they must, but because they prefer to.

Five findings survived our analysis:

1. Entertainment dominance is universal and cross-platform. It's not a Wikipedia artifact. Not a Western phenomenon. Not a function of education. YouTube trending is 100% entertainment across all five countries — zero education, zero science. Even YouTube search, where users actively choose, is 63–82% entertainment. Wikipedia shows 48%, books 25%. The less friction, the more entertainment wins.

2. Education doesn't predict attention allocation. Countries with higher Human Capital Index scores do not spend proportionally more Wikipedia attention on science, history, or the arts. The depth ratio shows no correlation with education metrics (p > 0.7). Whatever drives intellectual curiosity at the population level, it's not captured by formal education measures.

3. The power law is the same everywhere. Attention concentrates at the top following a power law with remarkably similar exponents (alpha 0.59–0.65) across all five languages. There is no "fat tail" of intellectual content in better-educated countries. The distribution shape is a structural feature of how large populations allocate attention, not a cultural variable.

4. Attention follows fashion. GitHub's 2024 data shows AI/ML capturing 60% of new-repo stars vs 20% historically. Developer attention is just as susceptible to hype cycles as general attention is to entertainment trends. The object of attention changes; the concentration pattern doesn't.

5. Friction is the only brake. YouTube trending (zero friction) = 100% entertainment. YouTube search (low friction) = 63–82%. Wikipedia (low-medium) = 48%. Open Library (medium) = 25%. GitHub (high friction, requires coding) = 2%. The effort required to access a platform is a better predictor of its intellectual content share than any education metric we tested.

What the data can say with confidence: the structure of collective attention is strikingly similar across languages and education levels. The ratio of entertainment to intellectual content is a feature of how humans allocate curiosity at scale, not a bug that education fixes.

Cipolla might have been onto something.

Data: Top-1,000 Wikipedia articles by pageviews from the Wikimedia REST API (2024–2025). Education metrics from the World Bank Open Data API (2020–2024). Google Trends data via pytrends (2024). Each dataset has SHA-256 checksum receipts.

Categorization: Articles were classified into 10 categories using regex pattern matching on titles, with language-specific patterns for Spanish, German, French, and Portuguese. Coverage: 56–75% of articles categorized (remainder as "other"). This is a known limitation — many articles about specific people or events don't match keyword patterns.

Statistical tests: Spearman rank correlations (n=5, acknowledging very low statistical power). Power law fit via OLS on log-log (acknowledging this is an approximation; Clauset et al. 2009 MLE would be more rigorous). Gini coefficient for inequality measurement. All p-values reported honestly; no cherry-picking.

Limitations: n=5 languages provides no real statistical power. Country mapping is simplistic (English Wikipedia ≠ USA). Categorization is pattern-based with systematic errors. No synthetic data was used. No causal claims are made.