The Price of Understanding

Complaints that used to fill two pages now fill twenty, and four in ten people avoid the news. Is the world really getting harder to understand? What can be measured, and what cannot.

In November 2026 The Guardian reported that British councils, MPs’ offices, employers and tribunals were being buried in complaints, requests and appeals their staff believe were written with chatbots. Complaints to councils used to run to about two pages; according to Deborah Evans, who leads the association of council lawyers, most are now ten times longer, and the longest she knew of ran to about 200. “It started as a trickle in 2024 and now it’s an onslaught,” said a caseworker in an MP’s office. One council, Wiltshire, is now exploring AI tools to summarise the long emails and draft the first replies.[1]

That is the comedy of Per My Last Email turned into public administration: one machine inflates a message, another deflates it, and the meaning has to survive both trips.[a] It is also one small case of a larger feeling, which I hold and suspect many readers share: that understanding the world we live in is getting harder. This essay asks whether there is evidence for it.

42%of people in 48 markets say they sometimes or often avoid the news, against 29% in 36 markets in 2017

62%worry about what is real and what is fake online, up from 56% in 2023

18×as many researchers needed to double chip density as in the early 1970s

11 of 27countries where adult literacy fell significantly since 2012; 2 improved

Four claims

What “harder to understand” could mean

The feeling hides four different claims, and they need different evidence.

The signal is noisier

There is more to read, less of it can be trusted, and it is harder to tell what is real. The world may be no more complicated; the view of it is foggier.

The world is more complex

There is more knowledge to master before reaching the edge of any field, and more rules to follow in ordinary life.

The things we use are closed

What we own and depend on is easier to use and harder to open: understanding it has moved from the owner to the maker.

We are less able

Reading, attention and reasoning are getting worse, so the same world is harder to take in.

No survey has followed people’s understanding of the world over decades, so the claim as a whole cannot be measured directly. Each of the four parts can, some better than others.

The noise

More to read, less to trust

The economist Herbert Simon saw the problem in 1971: “a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it”.[2] What is new is the scale, and the evidence that people are reacting to it. The Accelerant looked at what the feed did to attention, trust and the people in the middle who used to filter what reached everyone; this section asks only what the surveys show.

The Reuters Institute surveys about 97,000 people in 48 markets every year about how they use news. In 2017, 29% said they sometimes or often avoid it; in 2026, 42% did (Figure 1).[3][4] In 2024, 39% said they felt “worn out” by the amount of news, up from 28% in 2019.[5] Among those who avoid the news, 9% gave as a reason that it is too hard to understand, and 14% of those under 35.[3]

Figure 1. Share of people who say they sometimes or often avoid the news, all markets surveyed, each year’s published figure. The set of countries grew from 36 in 2017 to 48, and the 2026 report describes its 42% as flat on the year before, so the last step reflects the change of sample. The 2019 figure was read second-hand; 2022 is implied by the 2023 report. Source: Reuters Institute Digital News Report.[3][4][5][6] Drawn by scripts/understanding.py.

Trust is falling with it. The share who trust “most news most of the time” held at 40% from 2023 to 2025 and fell to 37% in 2026, the lowest since 2015, with falls in 29 of 48 markets. The share worried about what is real and what is fake online rose from 56% in 2023 to 62% in 2026.[3][4][6] Portugal is better placed than most on trust, at 51% in 2026, but among the most worried about what is real; 37% of Portuguese say they avoid the news.[4] In the United States, 51% of adults said in 2025 that they generally find it difficult to determine what is true and what is not when they get news.[7] From Bulletin to Feed follows how news became a product sold for attention, which is part of how that trust was spent.

Generative AI adds a new kind of noise: text that costs almost nothing to write and the same as ever to read. In a 2025 survey of 1,150 American employees, 40% had received AI-generated work in the previous month that looked finished but was not, and each case took nearly two hours to sort out.[8] Is AI making us lazy or are we just using it wrong? describes the same cost inside companies: the work saved in writing comes back as work in checking. Transport for London received 4,771 freedom-of-information requests in 2025-26, its busiest year on record and 19% more than the year before; appeals rose from 154 to 224, with a “noted increase” in appeals written with AI “which often contain highly technical and often inaccurate assertions and assumptions”.[9] The UK’s information commissioner warned public bodies in May 2026 that such requests can be “very long, repetitive and contain unrelated comments or misdirected demands”.[10]

On the reading side, people increasingly take the summary instead of the source. In a study of 900 Americans’ browsing, those shown an AI summary at the top of a Google search clicked through to a result on 8% of visits, against 15% without one, and clicked a source cited in the summary on 1%.[11][b] A summary is a reasonable answer to too much text. It also means that fewer people see the original, and that what they understand depends on what the summary kept.

The frontier

Further to the edge

The second claim is about knowledge itself, and here the evidence is good. A study of 709,577 abstracts from 123 scientific journals, published between 1881 and 2015, found that “the readability of science is steadily decreasing”: the share of abstracts scored as harder than college-graduate level rose from 14% in 1960 to 22% in 2015. The authors attribute much of it to “a growing use of general scientific jargon”, words like “furthermore” and “consequently” that fill papers in every field.[12]

It also takes longer to reach the frontier. The economist Benjamin Jones studied 2.1 million US patents granted between 1975 and 1999. The age at which inventors made their first invention rose by about 0.66 years per decade, inventors became more specialised, and teams grew from an average of 1.73 inventors to 2.33. His explanation is “the burden of knowledge”: each generation has more to learn before it can add anything, so it learns later and in narrower slices, and has to work in teams to cover a problem.[13]

Once there, progress costs more. Nicholas Bloom and three colleagues found that “the number of researchers required today to achieve the famous doubling of computer chip density is more than 18 times larger than the number required in the early 1970s”, a fall in research productivity of 6.8% a year. Across the US economy, they estimate research productivity has fallen by a factor of 41 since the 1930s. “Everywhere we look,” they write, “we find that ideas, and the exponential growth they imply, are getting harder to find.”[14]

This is real, but it is a difficulty for specialists. Nobody needs to read chip-design papers to live in a world with chips. A complex society works by dividing the understanding up: you trust the pharmacist, the engineer and the statistician to know their part. The question for everyone else is whether that trust is holding, and the falling trust in news suggests it is under strain.

The fine print

The rules we are expected to know

Ordinary life also comes with more rules. The longest consistent count is American: the Code of Federal Regulations grew from 54,834 pages in 1970 to 188,343 in 2021, about 3.4 times as long, or some 2,600 pages a year (Figure 2).[15] Pages are a crude measure: a long rule can be clear and a short one baffling. But nobody reads 188,000 pages, and the share any citizen or small firm can know shrinks every year.

Figure 2. Total pages of the US Code of Federal Regulations. Totals before 1970 include revisions and supplements and are not strictly comparable with later years. Source: Office of the Federal Register.[15] Drawn by scripts/understanding.py.

Outside the United States the counts are harder to compare, and the one I found for Portugal points the other way. The Observatório da Legislação Portuguesa counted 2,203 acts published in the first series of the Diário da República in 2000, 2,546 in 2007 and 822 in 2018.[16] That is the number of new acts each year, not the stock in force or its length, and I found no series of either for Portugal. For the European Union, a tracker built on the official database counted 46,438 acts in force on 10 November 2026, with no consistent series behind it.[17] The American count is the only long one, and it is evidence for the United States.

The same is true of the private rules we accept with a click. In 2008 Aleecia McDonald and Lorrie Cranor estimated that reading the privacy policies of the websites a typical American visits would take about 244 hours a year, or 40 minutes a day; at eight hours a day, that is some 30 working days.[18] That was before cookie banners, app permissions and terms for AI assistants. Nobody reads them, so everybody consents to things nobody understands.

Some of that length is not an accident, and the unreadable document is much older than the chatbot. Lawyers, and not only lawyers, have long produced text that is long and hard to read, and it is fair to ask how often that is its purpose. A study of the terms that the 500 most popular American websites bind their users to at sign-up found them, on average, about as readable as articles in academic journals; its authors note that “numerous scholars have suggested that consumer contracts are indeed written in a way that dissuades consumers from reading them”.[19] Companies do it with their own accounts. Across American annual reports, those “of firms with lower earnings are harder to read”: longer, and in more difficult language.[20] The worse the news, the thicker the report. It is a quiet relative of what Agnotology describes, ignorance made on purpose, and of the organised suspicion in Conspiracy Theories as Epistemic Systems; The Ministry of Doubt tells the same idea as a novel.

Purpose is harder to show than effect. When researchers at MIT tested lawyers, the lawyers too understood and remembered plain contracts better than contracts in legal style, judged them just as enforceable, and preferred them.[21] A later study found that people with no legal training fall into the same convoluted style when asked to write a law, which its authors read as a way of signalling authority, not a plan to confuse.[22] Habit, ritual and caution explain part of it. But whoever writes the document gains when it goes unread, and has little reason to shorten it.

The sealed box

Easier to use, harder to open

The fourth claim is about the things themselves. A car is far easier to drive than it was in 1970, and far harder to understand. My father could change the brakes and the oil and fix small mechanical faults himself. Now it takes a computer just to know what is wrong. By 2009 a premium car already ran on about 100 million lines of code, spread over dozens of processors.[23] When something fails, the fault is read through a diagnostic port with the maker’s software, and the information needed to repair it is the maker’s to give. Laws have had to prise it open: the European Union requires carmakers to give independent repairers standardised access to diagnostic and repair information,[24] and in 2020 about three-quarters of voters in Massachusetts approved a measure giving owners and independent garages access to the data their cars send to the maker.[25]

It is not only cars. In January 2025 the US Federal Trade Commission and two states sued the tractor maker Deere: according to the FTC, the only software tool able to carry out every repair on Deere equipment is made by Deere, which gives it only to its authorised dealers.[26] In July 2026 the FTC announced a proposed settlement, under which Deere must give farmers and independent repairers the same repair resources as its dealers, for ten years.[27] In the European Union, a directive that applies from 31 July 2026 says that manufacturers “shall not use any contractual clauses, hardware or software techniques that impede the repair” of the goods it lists.[28] Nobody forbids what nobody does.

The newest case goes a step further: a product that its makers cannot fully open either. “People outside the field are often surprised and alarmed to learn that we do not understand how our own AI creations work,” wrote Dario Amodei, the head of Anthropic, in April 2025.[29] Grown, Not Built explains why: these systems are trained, not written line by line, so there is no line to read.

The interface that makes the car easy to drive also moves the understanding of it from the person who owns it to the company that made it. That is a different way for the world to become harder to understand: not more noise, but fewer things that can be opened and looked inside. Nothing to Spare follows the same thread from the car to the networks we all depend on, and asks whether they have become too fragile.

The readers

Are we reading less well?

The last claim is the most uncomfortable, and the evidence is mixed. In 2023 the OECD tested the skills of about 160,000 adults aged 16 to 65 in 31 countries, representing 673 million people. Adult literacy “has stagnated or declined in most participating countries” since the first round in 2012 or 2015: of the 27 countries with both rounds, it fell significantly in 11, held steady in 14 and rose only in Finland and Denmark (Figure 3).[30] Across the OECD, 26% of adults now read at the lowest level or below, which means they can at most understand short, simple texts.[30]

Figure 3. Change in adults’ mean literacy score between the first round of the OECD Survey of Adult Skills (2012 or 2015) and 2023, in points. Small changes are within the margin of error: the OECD counts 11 significant declines and 2 significant rises. Poland’s result is flagged by the OECD for unusual response patterns. Portugal took part for the first time in 2023 and has no change to show. Source: OECD.[30] Drawn by scripts/understanding.py.

The averages hide who is falling. “Declines in average proficiency are largely due to falls among the lowest-performing adults,” the OECD writes. Literacy fell significantly among the weakest readers, the bottom tenth, in 16 countries, and the gap between the weakest and the strongest tenth widened in 17. Among adults who did not complete secondary school, scores fell in all but six countries, by between 12 and 46 points; among graduates they fell in 13 and rose only in Finland.[31] The loss is widest at the bottom, among the people for whom a long letter was already hard.

Portugal took part for the first time in 2023, so it has a level but no trend. The level is low: a mean literacy score of 235, against an OECD average of 260; 42% of Portuguese adults at the lowest level or below, against 26%; and only 4% at the two highest levels, against 12%. The young do better: Portuguese aged 16 to 24 averaged 258.[30][32] That is a country catching up through its schools, not a country in decline, but it means that for four in ten adults a long, dense letter is hard going, before anyone has made it twenty pages long.

Schoolchildren show a sharper fall, with a clear cause attached. In PISA 2022 the average 15-year-old in OECD countries scored almost 15 points lower in mathematics and about 10 lower in reading than in 2018, drops the OECD called unprecedented; it links them mainly to the pandemic, but notes they may not be due to it alone.[33]

Reading for pleasure, which is how most people practise long reading, is falling too. In the United States, the share of people who read for pleasure on a given day fell from 28% in 2004 to 16% in 2023, according to the American Time Use Survey, with the gaps by education and income widening.[34] In the United Kingdom, 32.7% of children and young people aged 8 to 18 said in 2025 that they enjoyed reading in their free time, the lowest since the question was first asked in 2005.[35]

And in Norway, where conscripts have taken the same test for decades, IQ scores rose until the cohorts born around 1975 and then began to fall. Comparing brothers, Bernt Bratsberg and Ole Rogeberg showed that the rise, the turning point and the decline “can all be fully recovered from within-family variation”: the cause is something in the environment, not who was having children.[36] What in the environment is not known.

One popular piece of evidence should be left out. The claim that human attention spans have fallen to eight seconds, shorter than a goldfish’s, comes from a 2015 Microsoft marketing report that took its numbers from a website that could not say where they came from; researchers say an average attention span is not a meaningful measure, since attention depends on the task.[37][38] The Accelerant reached the same conclusion about that number.

The old complaint

Every generation says so

The strongest objection to the whole idea is that people have always felt it. In the first century Seneca warned a friend that “in reading of many books is distraction”.[39] In 1545, a century after the printing press, the Swiss scholar Conrad Gessner complained of the “confusing and harmful abundance of books”, in the historian Ann Blair’s translation, and called on kings and princes to solve it.[40] Simon wrote his line in 1971. The feeling, measured, is high but not obviously rising: in 18 countries surveyed by Ipsos, 75% agreed in 2013 that “the world today is changing too fast”, and 76% in 2023.[41] In the United States, 68% said in 2018 that they were worn out by the amount of news, and 66% in 2020.[42]

Misunderstanding is not new either. In 2017 Gapminder asked about 12,000 people in 14 rich countries twelve simple questions about the world, on poverty, population and health. They got an average of 2.2 right, fewer than the four that random guessing would give.[43] People were not well informed before chatbots. They were misinformed in a steadier, more familiar way.

The case against

More access than ever

There is a fair case that understanding has never been easier. More people have more years of school than at any time in history. An encyclopaedia, a translator, a course from a good university and an assistant that will explain a contract in plain words are free to anyone with a phone. For someone reading in a second language, or with dyslexia, these tools can be the difference between understanding a letter from the tax office and not.

Complexity is also not the same as incomprehensibility. A car is far more complicated than it was in 1970 and far easier to drive. Most of the growing complexity of the world is hidden behind interfaces that work, and nobody needs to understand it until the interface fails or misleads. And specialisation, the burden of knowledge, is how societies have always handled complexity: by dividing it up and trusting each other. The sealed box is the limit of that argument: there the trust is asked for, and the means to check it are withheld.

These are good arguments, and they show that the claim needs narrowing. The tools for understanding have improved. What has got worse is the environment they work in: more noise, less trust, longer rules, fewer people practised at reading anything long, and fewer of the things we own that we can open and understand.

Dearer, not closed

Understanding has got more expensive

Put together, the evidence supports a narrower version of the claim than the one I started with.

The signal is noisier

Good evidence. News avoidance up from 29% to 42% since 2017, in a growing set of countries; worry about what is real up; trust at its lowest since 2015; AI-written text now arriving in quantities that public bodies cannot read.

The frontier is further away

Good evidence, for specialists. Science is harder to read, inventors start later and in narrower fields, and ideas take many more researchers to find.

The rules are longer

Good evidence that they are longer in the United States, where the only long count exists; weaker that they are harder, since length is a crude measure, though some of the length and difficulty serves the writer. Portugal publishes fewer new acts a year than in 2000.

The things we use are closed

Suggestive evidence: cases, not a count. Cars and tractors can be diagnosed only with the maker’s software, and right-to-repair laws are the response; the makers of AI systems say they do not understand how their own work. Easier to use, harder to understand.

We are less able

Mixed. Adult literacy falling or flat in most rich countries, with the losses concentrated among the weakest readers; children’s scores down after the pandemic, reading for pleasure falling; but schooling and access higher than ever, and in Portugal the young read better than their elders.

The world as a whole is harder to understand

Not measured. The feeling is widespread but has been high for a long time.

The version I would defend is this: understanding the world has become more expensive. It costs more attention to filter what reaches us, more effort to check what can be trusted, and more years to reach the edge of any field, and more of what we depend on can be understood only by the people who made it. Our ability to pay that cost is not rising to match, and in some places it is falling.

Generative AI pushes in both directions at once. It makes explanation cheap, which helps; and it makes text cheap, which floods the channels explanation has to travel through. A council that answers AI-written complaints with AI-written summaries, as Wiltshire is considering, reaches the logical end of that: machines writing to machines, and the person who wanted to be understood, and the person who has to understand, both at one remove from what was said. It is also the public answering in a language it was taught. The long document that nobody is meant to read used to need a lawyer; now anyone can produce one, and the institutions that wrote such documents for decades are the ones complaining that they cannot read them.

What would help is unglamorous. Short originals, kept alongside every summary. Sources that can be followed. Limits on length where a public body must answer, as some ombudsmen are now setting,[1] and the same discipline in what public bodies and companies send to us. Time spent reading something long, all the way to the end. And a survey that measures, over years, whether people can explain the world they live in, so that the next essay on this question can answer it directly.

On method and tools

This essay was written collaboratively with Claude Opus 5.5 (Anthropic) on 10 November 2026: Luis Matos Ferreira’s question and view, and his editorial direction; the machine’s research, analysis and drafting. Claude research agents checked the sources against the originals that day. Some were read second-hand: the Guardian report from the text Luis supplied, since the site blocks automated reading; the information commissioner’s guidance as quoted by Local Government Lawyer; the 2019 news-avoidance figure from a summary of that year’s report; the OECD’s adult-skills report through a copy published by Spain’s education ministry, Portugal’s detailed figures through the OECD country note, and the count of significant changes by country through Ireland’s Central Statistics Office, which cites the OECD; the PISA figures through the OECD’s online report; the attention-span debunk through summaries of the BBC’s report; the IEEE Spectrum article on software in cars through its published summary; Gessner through Ann Blair’s article and Seneca through a transcription of the Loeb translation; and the two FTC announcements on Deere through an automated reader, which is why they are paraphrased and not quoted; the two MIT studies of legal language through MIT’s own reports and summaries of the papers; and the study of annual reports through its abstract. The study of website terms was read in its SSRN version. Where the Guardian gives 4,769 freedom-of-information requests to Transport for London, the essay uses Transport for London’s own figure of 4,771 received, of which 4,769 were answered in time. Figures and the few calculations are made by scripts/understanding.py (results in docs/understanding-results.json). The cover is computed by scripts/the_price_of_understanding_cover.py: one column for each decade of the US Code of Federal Regulations, one grey line per thousand pages, and in each column one teal line, the rule you need, which stays the same size while the rest grows.

Authored by: Luis Matos Ferreira — Physicist, Developer, Writer

Updates and corrections

Last updated 10 November 2026.

1. 10 November 2026. After a review of the published text: the closed products now have their own section (“The sealed box”), with tractors, the EU repair directive and AI systems added to cars, and the essay speaks of four claims instead of three; older complaints about too many books (Seneca, Gessner) added; the distribution behind the literacy averages added; a count of Portuguese legislation added, which does not show growth; the council in the opening named; and the change in the news-avoidance sample noted in the summary figures.
2. 10 November 2026. Added to “The fine print” and to the closing: text that is long and hard to read on purpose is older than the chatbot (website terms, company reports), with the evidence that habit and ritual explain part of it. Links added to earlier essays on this blog that cover the same ground.

Go deeper

Ideas this essay uses without stopping to explain them, and where this blog does. The letters in the text point here; the links go straight to the section.

[a] How meaning gets lost

Why summaries drop the middle, overstate, flatter and take jokes for facts. Per My Last Email, “Why the meaning gets lost”.

[b] The answer instead of the link

What happens to the web when search engines answer instead of sending readers on. The Dead Internet, Measured, “The answer instead of the link”.

Related essays on this blog

1. Nothing to Spare — whether our connected, just-in-time systems have become too fragile, or we have forgotten how they were.
2. Per My Last Email — a comedy about an email inflated to eighteen pages and summarised back, and what is real in it.
3. The Accelerant — what social media changed in attention, trust and the way facts are settled.
4. Agnotology: How Ignorance Gets Manufactured — ignorance that is made on purpose, from tobacco to climate.
5. Conspiracy Theories as Epistemic Systems — how suspicion is organised when trust fails.
6. The Dead Internet, Measured — who reads, writes and answers on the web now.
7. The Manufactured Public — what happens to public debate when messages cost nothing to produce.
8. What We Leave Behind — how fast the web forgets, and who keeps the rest.
9. Mostly Machines — this blog’s own readers, counted three ways.

Sources

1. J. Partridge, “‘AI clog’: UK public bodies battle onslaught of chatbot-generated complaints”, The Guardian, 8 November 2026.
2. H. A. Simon, “Designing Organizations for an Information-Rich World”, in M. Greenberger (ed.), Computers, Communications, and the Public Interest, Johns Hopkins Press, 1971, pp. 37–72 (quotation pp. 40–41).
3. N. Newman et al., Reuters Institute Digital News Report 2025, Reuters Institute for the Study of Journalism, June 2025 (48 markets, 97,055 respondents).
4. Reuters Institute for the Study of Journalism, Digital News Report 2026, June 2026 (48 markets, 97,520 respondents), including the Portugal country page.
5. N. Newman et al., Reuters Institute Digital News Report 2024, Reuters Institute for the Study of Journalism, June 2024.
6. N. Newman et al., Reuters Institute Digital News Report 2023, Reuters Institute for the Study of Journalism, June 2023.
7. L. Wang and J. Haner, “Many Americans say they often come across inaccurate news – and have a hard time knowing what’s true”, Pew Research Center, 29 October 2025 (survey of 5,153 adults, August 2025).
8. K. Niederhoffer, G. Rosen Kellerman, A. Lee, A. Liebscher, K. Rapuano and J. T. Hancock, “AI-Generated ‘Workslop’ Is Destroying Productivity”, Harvard Business Review, 22 September 2025.
9. Transport for London, Audit and Assurance Committee, 9 September 2026, item 10, “Freedom of Information Update”, public reports pack, pp. 399–402.
10. Information Commissioner’s Office, “Freedom of Information (FOI) and Artificial Intelligence”, 6 May 2026, as quoted by Local Government Lawyer, 7 May 2026.
11. A. Chapekis and A. Lieb, “Google users are less likely to click on links when an AI summary appears in the results”, Pew Research Center, 22 July 2025.
12. P. Plavén-Sigray, G. J. Matheson, B. C. Schiffler and W. H. Thompson, “The readability of scientific texts is decreasing over time”, eLife 6, e27725, 5 September 2017.
13. B. F. Jones, “The Burden of Knowledge and the ‘Death of the Renaissance Man’: Is Innovation Getting Harder?”, Review of Economic Studies 76(1), 283–317, 2009 (author’s version).
14. N. Bloom, C. I. Jones, J. Van Reenen and M. Webb, “Are Ideas Getting Harder to Find?”, American Economic Review 110(4), 1104–1144, 2020.
15. Office of the Federal Register, “Code of Federal Regulations: total pages 1938–1949, and total volumes and pages 1950–2021”, 2023.
16. Observatório da Legislação Portuguesa (CEDIS, NOVA School of Law), Boletim n.º 8, December 2019, table I, p. 16.
17. M. Ovádek, “Legislative Output of the European Union”, tracker drawing on EUR-Lex, read 10 November 2026.
18. A. M. McDonald and L. F. Cranor, “The Cost of Reading Privacy Policies”, I/S: A Journal of Law and Policy for the Information Society 4(3), 2008 (authors’ version).
19. U. Benoliel and S. I. Becher, “The Duty to Read the Unreadable”, Boston College Law Review 60, 2255–2296, 2019 (SSRN version).
20. F. Li, “Annual report readability, current earnings, and earnings persistence”, Journal of Accounting and Economics 45(2–3), 221–247, 2008 (abstract).
21. E. Martínez, F. Mollica and E. Gibson, “Even lawyers do not like legalese”, PNAS 120, e2302672120, 2023, read through MIT’s report and summaries.
22. E. Martínez, F. Mollica and E. Gibson, “Even laypeople use legalese”, PNAS 121, e2405564121, 2024, read through MIT’s report, 19 August 2024.
23. R. N. Charette, “This Car Runs on Code”, IEEE Spectrum, February 2009 (read through the article’s published summary).
24. Regulation (EU) 2018/858 of the European Parliament and of the Council of 30 May 2018 on the approval and market surveillance of motor vehicles, chapter XIV and annex X, on access to vehicle OBD information and repair and maintenance information.
25. “Massachusetts voters approve wireless vehicle data access measure, reject ranked-choice voting”, Ballotpedia News, 4 November 2020.
26. Federal Trade Commission, “FTC, States Sue Deere & Company to Protect Farmers from Unfair Corporate Tactics, High Repair Costs”, 15 January 2025.
27. Federal Trade Commission, “FTC, States Secure Settlement with Deere & Company, Advancing Farmers’ Right to Repair”, 8 July 2026.
28. Directive (EU) 2024/1799 of the European Parliament and of the Council of 13 June 2024 on common rules promoting the repair of goods, articles 5(6) and 22.
29. D. Amodei, “The Urgency of Interpretability”, April 2025.
30. A. Schleicher and S. Scarpetta, Survey of Adult Skills 2023: Insights and Interpretations, OECD, December 2024, as published by Spain’s Ministry of Education.
31. OECD, Do Adults Have the Skills They Need to Thrive in a Changing World? Survey of Adult Skills 2023, December 2024, pp. 22–23, 101, 110–112 and 117.
32. OECD, “Survey of Adults Skills 2023: Portugal”, country note, December 2024.
33. OECD, PISA 2022 Results (Volume I): The State of Learning and Equity in Education, December 2023.
34. J. K. Bone, F. Bu, J. K. Sonke and D. Fancourt, “The decline in reading for pleasure over 20 years of the American Time Use Survey”, iScience 28(9), 113288, 20 August 2025.
35. National Literacy Trust, “Children and young people’s reading in 2025”, 3 June 2025.
36. B. Bratsberg and O. Rogeberg, “Flynn effect and its reversal are both environmentally caused”, PNAS 115(26), 6674–6678, 2018.
37. S. Maybin, “Busting the attention span myth”, BBC News, 10 March 2017, read through summaries of it; on Microsoft Canada, “Attention spans”, Consumer Insights, 2015.
38. N. A. Bradbury, “Attention span during lectures: 8 seconds, 10 minutes, or more?”, Advances in Physiology Education 40(4), 509–513, 2016.
39. Seneca, Epistulae morales ad Lucilium 2.3 (“Distringit librorum multitudo”), trans. R. M. Gummere, Loeb Classical Library, 1917, in the Wikisource transcription.
40. A. Blair, “Reading Strategies for Coping with Information Overload, ca. 1550–1700”, Journal of the History of Ideas 64(1), 11–28, 2003, p. 11, citing C. Gessner, Bibliotheca universalis, Zurich, 1545.
41. Ipsos, “Ipsos Global Trends 2023 highlights”, 21 November 2023.
42. “Almost seven-in-ten Americans have news fatigue, more among Republicans”, Pew Research Center, 26 February 2020.
43. Gapminder, “Gapminder Misconception Study 2017”, retrieved 10 November 2026.