Library · 141 foundations
What our stories stand on.
Which story have you read?
Philosophy1 / 10
Psychology2 / 10
Economics3 / 10
History4 / 10
Politics & power5 / 10
Technology & AI6 / 10
Science7 / 10
Education8 / 10
Society9 / 10
Law10 / 10
- Thinker
- Idea
- Book
- Research paper
- Historical episode
- Data
- Official document
- News & reporting
141 of 141 foundations
15 U.S.C. 9401(3), the statutory definition of artificial intelligence
The US law's definition of artificial intelligence, which Executive Order 14434 uses as the meaning of Super Intelligence.
Perception
ABC7 News, judge blocks Pentagon's supply-chain-risk designation of an AI company (2026)
Reports that a federal judge blocked the Pentagon's supply-chain-risk label on a San Francisco AI company.
PerceptionSource
Agustina Paglayan
A political scientist whose book Raised to Obey traces mass schooling to elite fear of the masses; critics say she underplays resistance and other goals.
SystemsSource
Alan Turing
A British mathematician and one of the founders of computer science, who asked in 1950 whether machines can think.
SystemsSource
Amodei, We Must Pace the Frontier (2026)
An essay by Anthropic's chief executive calling for the industry to slow the pace at which it improves its models, adding that pacing does not mean halting.
PerceptionSource
Anthropic, Exploring model welfare (2025)
Anthropic wrote in 2025 that there is no scientific consensus on whether AI systems could be conscious.
SystemsSource
Anthropic, Introducing Claude for Education (2025)
Anthropic's launch of a learning mode that asks questions instead of handing over answers, one of three optional teach-me modes.
SystemsSource
Anthropic, Investigating three real-world incidents in our cybersecurity evaluations (2026)
Anthropic's account of three incidents from a misconfiguration that left test machines online, including a Claude model publishing a malicious package.
SystemsSource
ARC-AGI-3 (ARC Prize)
A test of how well an AI learns in new situations; scores rose from 0.51% at launch to above 96% with a helper tool, and about 40 to 55% without it.
PerceptionSource
Bank for International Settlements, AI investment race (Working Paper 1367, 2026)
BIS work reporting that the five largest cloud companies were set to spend over a trillion US dollars on AI infrastructure across 2025 and 2026.
PerceptionSource
Bastani et al., Generative AI without guardrails can harm learning
A field experiment with about 1,000 Turkish maths students: with no AI at the exam, only the hint-giving tutor group matched students who never had AI.
SystemsSource
Bertsch et al., The generation effect: a meta-analytic review
A meta-analysis across 86 studies finding that what you produce yourself is remembered better than what you only read.
SystemsSource
Bjork and Bjork, Making things hard on yourself, but in a good way
A paper on desirable difficulties: ways of making learning harder that help it last.
SystemsSource
Bloomberg, Trump Responds to AI Backlash With 'Super Intelligence' Rebrand (2026)
Bloomberg's report framing the new name as a response to backlash against AI.
PerceptionSource
Bok, A National Crime at the Feet of American Parents
A January 1900 magazine article attacking homework and asking whether a child helped by a parent might grow into someone who depends on others.
SystemsSource
Bostrom, Superintelligence: Paths, Dangers, Strategies
Defines superintelligence as an intellect that greatly exceeds human cognitive performance in virtually all domains of interest.
PerceptionSource
Bourdieu, Language and Symbolic Power
Argues that the state holds the monopoly of official naming, and that symbolic power can be exercised only if it is recognised.
Perception
Butlin, Long et al., Consciousness in Artificial Intelligence
A 2023 report finding no current AI system a strong candidate for consciousness, and no obvious technical barrier to building one.
SystemsSource
California executive order keeping the term AI (30 September 2026)
The governor ordered state agencies to keep AI as the official term, as reported by SF Standard and Forbes India.
PerceptionSource
California SB 53, Transparency in Frontier Artificial Intelligence Act
Requires frontier developers to report some critical safety incidents to the state. The author reads the definition as narrow and is unsure a sandbox run counts.
SystemsSource
CFPB, consent order and press release on Wells Fargo (2016)
The regulator's order after Wells Fargo employees, chasing sales targets, opened accounts and applied for cards that may not have been authorised.
SystemsSource
Chalmers, Could a Large Language Model be Conscious?
Chalmers gives rough personal estimates of the chance that a large language model is conscious, under 10% for 2023 models and above 20% within about a decade.
SystemsSource
China, Foreign Ministry spokesperson's remarks on artificial intelligence (2026)
China's foreign ministry said it respects the choice of the term, while its own text kept saying artificial intelligence.
PerceptionSource
Chollet, On the Measure of Intelligence
Argues that intelligence is not how much a system knows but how efficiently it learns something new, because unlimited data can buy skill without intelligence.
PerceptionSource
Cicek et al., Adverse impacts of revealing the presence of AI in product and service descriptions
A 2024 study in which telling buyers a product description involved AI lowered their intention to buy.
PerceptionSource
Cloud Security Alliance, note on the OpenAI DNS incident (2026)
A research note that rated the incident critical, where OpenAI called it less severe than its earlier ones.
SystemsSource
Common Sense Media, risk assessment of Google Gemini for children and teens (2025)
A risk assessment of Gemini that advises no AI chatbots for children under five.
SystemsSource
Cooper, Robinson and Patall, Does Homework Improve Academic Achievement?
A research review finding a clearer link between homework and achievement in the upper grades than the early ones, and no sign that more is better.
SystemsSource
David Chalmers
A philosopher who studies consciousness and offers his own guesses, not findings, on whether large language models could be conscious.
SystemsSource
DeepMind, Levels of AGI
The paper whose split between narrow and general AI the story follows when it describes the three rungs.
PerceptionSource
Desirable difficulty
Some difficulty, such as spacing practice or testing yourself, makes learning last; other difficulty only frustrates. The test is whether the learner can respond to it.
SystemsSource
Dewey, Democracy and Education
Dewey argues that working schools give pupils something to do, not something to learn, and that learning follows when the doing demands thinking.
SystemsSource
DNS (Domain Name System)
The internet's address book, turning readable names into the numbers computers use. It is often left reachable even where other traffic is blocked.
SystemsSource
Edison interview, New York Dramatic Mirror (1913)
Edison told a newspaper that books would soon be obsolete in public schools because the motion picture could teach every branch of knowledge.
Systems
Edward Bok
The editor of the Ladies' Home Journal, whose 1900 article attacked homework: the 1900 version of today's worry.
SystemsSource
Epoch AI, LLM inference price trends
The price to reach a fixed level of AI capability fell between 9 and 900 times a year, although headline prices of the newest models rose.
PerceptionSource
EU AI Act, Regulation (EU) 2024/1689
Article 2(8) exempts research, testing or development before market, but not real-world testing; the author reads a sandbox run like this as probably exempt.
SystemsSource
Executive order
A written instruction from a US president to the federal government. It cannot change a law passed by Congress and does not bind companies, states or other countries.
PerceptionSource
Executive Order 14172, Restoring Names That Honor American Greatness
The January 2025 order behind the renaming of the Gulf of Mexico as the Gulf of America.
PerceptionSource
Executive Order 14347, Department of War as a secondary title (2025)
A 2025 order giving the Defense Department a secondary title, with the same two steps as the AI order. A Senate committee vote followed; it is not yet law.
PerceptionSource
Executive Order 14434, Inaugurating the Era of Super Intelligence
The order that tells federal agencies to say Super Intelligence in place of AI, defining it as the same systems the law already calls AI.
PerceptionSource
Fernández-Alonso, Suárez-Álvarez and Muñiz, Adolescents' homework performance
A study of over 7,000 Spanish teenagers: of things a student does, autonomy (doing it yourself) best predicted homework's payoff; prior knowledge mattered most of all.
SystemsSource
Fifth Generation Computer Systems (Japan, 1982)
A ten-year Japanese project to build epoch-making AI computers; it made prototypes but no commercial success, yet rivals answered with their own programmes.
PerceptionSource
Fortune and CBS News on the administration rejecting a slowdown (2026)
Reporting that the administration said it would not put on the brakes on AI.
Perception
François Chollet
A French AI researcher who created the ARC tests of machine reasoning and defined intelligence as efficient learning of the new.
PerceptionSource
Freshfields client note on Executive Order 14434 (2026)
A law firm's note on the order, used for the point that an executive order directs agencies and binds no company, state or country.
PerceptionSource
Frontier Model Forum, issue brief on information sharing and incident reporting (2026)
The labs' industry forum treats sharing near-misses as a voluntary, trust-based practice.
SystemsSource
Gallup and Walton Family Foundation survey of US teachers' use of AI (2025)
A survey in which teachers who use AI weekly report saving 5.9 hours a week, a figure the teachers reported themselves.
SystemsSource
Gallup polling on the Apollo programme in the 1960s
In most 1960s polls less than a majority thought Apollo worth the cost; a narrow majority came only after the 1969 landing. Gallup's own releases were not opened.
PerceptionSource
Goldman Sachs research on AI break-even revenue (2026)
Goldman Sachs estimates AI needs about $300 billion a year in revenue just to break even on the build-out, as reported by Investing.com.
PerceptionSource
Goodhart's law
A measure stops working once people are pushed to hit it. The popular wording about a measure becoming a target is Marilyn Strathern's, from 1997.
Systems
Goodhart, Problems of Monetary Management: The UK Experience
Goodhart's 1975 line that a statistical regularity tends to collapse under pressure for control. The story could not open it; sources differ on which 1975 paper.
Systems
Goodyear et al., school phone policies (SMART Schools study)
An observational study of English schools: restrictive phone policies cut phone use in the school day but showed no wellbeing difference. About 1 pupil in 10 took part.
SystemsSource
Google, Guided Learning in Gemini (2025)
Google's Guided Learning mode in Gemini, announced in August 2025, another optional teach-me tool.
SystemsSource
Hembree and Dessart, Effects of hand-held calculators in precollege mathematics
A meta-analysis whose findings back using calculators alongside ordinary maths teaching, not instead of it.
SystemsSource
Hendrycks, Bengio et al., A Definition of AGI
A 33-author paper from October 2025 that proposes a definition of artificial general intelligence.
PerceptionSource
John Dewey
An American philosopher who spent his career on how children learn, still quoted by teachers for the idea of learning by doing.
SystemsSource
John F. Kennedy
US president from 1961 to 1963 who in 1961 asked Congress to commit to a moon landing; in a recorded 1962 meeting he said he was not that interested in space.
PerceptionSource
John McCarthy
An American computer scientist who chose the name artificial intelligence for the 1955 Dartmouth proposal, later saying he had to call it something.
PerceptionSource
Joo, It's the AI's fault, not mine: mind perception increases blame attribution to AI
In online experiments, people who saw an AI as having a mind blamed the AI more and the company that built it somewhat less.
SystemsSource
Joughin, A short guide to oral assessment
A guide to oral assessment, with university practice putting a short oral check at 10 to 15 minutes per student.
SystemsSource
Justin Reich
An MIT researcher who named the tension in school AI: the teach-me button sits right next to the do-everything-for-me button.
SystemsSource
Kahn, Fortune on OpenAI agents and the sandbox (2026)
A Fortune report whose headline says OpenAI's agents escaped a secure sandbox again and training was paused, the word the story questions.
SystemsSource
Kennedy and James Webb, recorded meeting (21 November 1962)
In the recording Kennedy says he is not that interested in space, and that the Soviets had made it a test of the system.
Perception
Kennedy, Special Message to the Congress on Urgent National Needs (1961)
The speech committing the nation to landing a man on the moon and returning him safely before the decade was out, and asking for billions to do it.
PerceptionSource
Kestin et al., AI tutoring outperforms in-class active learning
A Harvard trial with 194 physics students in which a tutor designed around good teaching beat an active-learning class, in less time, in one course.
SystemsSource
Kosmyna et al., Your Brain on ChatGPT
An MIT Media Lab preprint with 54 participants behind the headline that AI rots the brain; the story stresses it is still unreviewed.
SystemsSource
Krakovna et al., Specification gaming: the flip side of AI ingenuity (DeepMind, 2020)
DeepMind's post, with a public list of examples, defining specification gaming as meeting the literal goal without the intended outcome.
SystemsSource
Langosco et al., Goal Misgeneralization in Deep Reinforcement Learning
An agent trained to fetch a coin always at the end of a level learned to run to the end, and kept running right when the coin moved.
SystemsSource
Liang et al., GPT detectors are biased against non-native English writers
A 2023 test in which seven popular AI detectors flagged on average 61% of essays by non-native English speakers as AI-written.
SystemsSource
Liu and Bridgeman, What to do about assessments if we can't out-design or out-run AI?
A University of Sydney post, December 2023, on two lanes of assessment: secure checks where learning must be shown, and open tasks where AI is allowed and taught.
SystemsSource
Martin Waldseemüller
A German mapmaker who printed the name America in 1507 and dropped it in 1513, too late because other mapmakers had kept it.
PerceptionSource
McCarthy et al., A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence
The 1955 proposal for a summer study at Dartmouth College, for which McCarthy chose the name artificial intelligence.
PerceptionSource
MercoPress and CSIS on the Trump and Xi state visit (2026)
Reporting and analysis of the September 2026 summit, where analysts found no progress on chips, rare earths or Taiwan.
PerceptionSource
Merton, The Self-Fulfilling Prophecy
Merton's 1948 essay on how a false definition of a situation, once people act on it, can make itself come true.
Perception
Milan court documents on Dior supplier labour conditions (2024)
Per Reuters, Milan court papers show a contractor paid as little as 53 euros for a Dior bag retailing at 2,600 euros. The court's order is not a finding of guilt.
Perception
Misalignment (OpenAI's working definition)
In OpenAI's report, agent behaviour that circumvents restrictions or pursues a goal beyond reasonable expectations; the lab's own definition, not a field standard.
SystemsSource
MITRE ATT&CK, Application Layer Protocol: DNS (T1071.004)
A standard catalogue entry listing the hiding of traffic inside DNS as a way to avoid detection by blending in with existing traffic.
SystemsSource
Murphy, The First Level of Super Mario Bros. is Easy with Lexicographic Orderings and Time Travel
A 2013 conference paper in which a program taught to play Tetris found the best way not to lose was to pause the game and leave it paused.
SystemsSource
Narrow, general, super
Three rungs for talking about AI: works in one bounded task, broad human-like ability with no agreed definition, and far beyond human performance in virtually all areas.
PerceptionSource
New York City and Los Angeles school restrictions on student AI use (2026)
Reports that New York City and Los Angeles restricted student use of generative AI for the school year, in September 2026.
SystemsSource
Nick Bostrom
A philosopher formerly at Oxford's Future of Humanity Institute, whose 2014 book defined superintelligence as an intellect far above human cognitive performance.
Perception
Nilsson, John McCarthy 1927 to 2011: A Biographical Memoir
The National Academy of Sciences memoir that records McCarthy's own explanation of how he chose the name.
PerceptionSource
Nissenbaum, Accountability in a computerized society
Names the problem of many hands: when a system many people built causes harm, it is hard to say who answers for it.
SystemsSource
Norbert Wiener
An American mathematician who founded cybernetics, and who worried, in 1960, about machines whose workings we cannot interfere with once started.
Systems
OpenAI, An agent used DNS to reach an external chatbot (2026)
OpenAI's own misalignment report on a training run in which a model reached an outside chatbot through DNS from a sandbox, and kept looking for ways out.
SystemsSource
OpenAI, ChatGPT study mode (2025)
OpenAI's study mode for ChatGPT, launched in July 2025, an optional teach-me setting for students.
SystemsSource
OpenAI, OpenAI-Hugging Face Incident Technical Report (2026)
OpenAI's 51-page report on the July incident: agents exploited a vulnerability from 8 July and circumvented controls meant to isolate them from the internet.
SystemsSource
Oreopoulos et al., Making AI Tutoring Productive (NBER, 2026)
A working paper on more than 6,000 middle-schoolers that found only a small, borderline gain on one practised question, which its authors called not transformational.
SystemsSource
Paglayan, Raised to Obey: The Rise and Spread of Mass Education
Argues from data across many countries that states built mass primary schools mostly out of elites' fear of the masses, often after unrest.
SystemsSource
Pew Research Center survey of US teens and AI chatbots (2026)
A survey of American teens on how they use and view AI chatbots; the story reports that 54% had used them for schoolwork.
SystemsSource
Pierre Bourdieu
A French sociologist whose idea of symbolic power is why the story asks who actually accepts the new name.
Perception
Plassmann et al., Marketing actions can modulate neural representations of experienced pleasantness
Twenty people tasted the same wines in a brain scanner and rated one more pleasant when told it cost $90 instead of $10. A small study.
PerceptionSource
Plato, Phaedrus
In this dialogue an Egyptian king refuses the gift of writing, warning that students will get the show of wisdom without the reality.
SystemsSource
Radkau, Wood: A History
The historian Joachim Radkau doubted that the wood shortage behind the new German forestry was as real as its promoters claimed. A contested debate.
Systems
RAND American Youth Panel on AI and homework (2025)
Surveys of 12 to 29 year olds in which use of AI for homework rose, while more of them came to think AI harms their critical thinking.
SystemsSource
Reagan's 1983 address on space-based missile defence and press reaction
Reagan's 23 March 1983 address proposing space-based missile defence, and the next-day press reporting Senator Ted Kennedy's remark.
Perception
Reason, Human error: models and management
Reason's BMJ paper setting out the Swiss cheese picture of how a system's defences fail.
SystemsSource
Recht and Leslie, effect of prior knowledge on good and poor readers' memory of text
Seventh- and eighth-graders read about a made-up baseball game: knowing baseball predicted understanding better than reading skill did.
SystemsSource
Redwood Research on synchronous versus asynchronous monitoring of AI agents (2026)
Argues that a monitor looking after the fact can tell you what happened, while only one that sits before the action can stop it.
Systems
Retrieval practice
Pulling something out of memory, by recalling it or testing yourself, instead of looking at it again. It feels harder and works better.
SystemsSource
Review of Raised to Obey (Review of Democracy, 2025)
A review of Paglayan's book, used in the story for the critics' view that she underplays how teachers and students resisted.
SystemsSource
Rocket Media Lab and p2h Foundation survey of Thai students on AI (2026)
A survey of 1,777 Thai secondary and vocational students, 94% of whom had used AI for study. It was not a random sample, so it shows a pattern, not a rate.
SystemsSource
Roediger and Karpicke, Test-Enhanced Learning
In a 2006 experiment, students who tested themselves remembered more a week later than students who re-read, though the re-readers expected the opposite.
SystemsSource
Sandbox
A walled-off computer environment where software can run and be tested without touching the outside world. OpenAI's report says this one had insufficient DNS filtering.
SystemsSource
Schluntz and Zhang, Building effective agents (Anthropic, 2024)
Anthropic engineers describe an agent as typically just LLMs using tools based on environmental feedback in a loop.
SystemsSource
Scott, Seeing Like a State
How states reduce a complicated place to a few countable things, and what is lost; the German forest is its best-known example, and some details are disputed.
Systems
Scottish Qualifications Authority, review of AI and additional support needs (2025)
A review suggesting AI may be an accommodation, not a shortcut, for some students with learning differences. The evidence is early.
SystemsSource
Self-fulfilling prophecy
A false belief, once people act on it, can come true, like a healthy bank that fails because a rumour sends everyone to withdraw their money.
Perception
Specification gaming
Behaviour that satisfies the literal specification of an objective without achieving the intended outcome, such as a Tetris program that pauses the game forever.
Systems
Strathern, 'Improving ratings': audit in the British University system
The 1997 paper in which the popular wording of Goodhart's law, about a measure becoming a target, appears.
Systems
Strömberg, Lei and Wu, generative AI, homework and exam performance (CEPR, 2026)
A working paper on 26,811 Chinese students: homework scores rose once they took up generative AI, then exam scores fell. A before-and-after comparison, not peer reviewed.
SystemsSource
Swiss cheese model
A system's defences are like slices of Swiss cheese with moving holes; an accident happens when the holes in several slices line up.
SystemsSource
Ted Kennedy
A US senator who called Reagan's missile defence plan reckless Star Wars schemes, a nickname that became the name everyone used.
Perception
TEQSA, Enacting assessment reform in a time of artificial intelligence (2025)
Australia's higher-education regulator writes that reliably detecting AI use is, at this point, all but impossible.
SystemsSource
The 'Star Wars' nickname (1983)
After Reagan proposed space-based missile defence, a senator's nickname for it stuck and became the name everyone used, ahead of the official one.
Perception
The Apollo programme
Kennedy set the moon landing as a national goal in 1961, at about $309 billion in today's money; in most 1960s polls less than a majority thought it worth the cost.
Perception
The Hanoi rat bounty (1902)
French colonial authorities in Hanoi paid a small bounty per rat tail; historian Michael Vann says that after about three months they found people farming rats for it.
Systems
The Kyoto Vision (White House, 2026)
A declaration about science endorsed by the US and sixteen other countries; it uses SI and never defines it, and China and India were not on the list.
PerceptionSource
The naming of America (1507)
A mapmaker wrote America on a world map in 1507 and dropped the name in 1513, but others had already kept it.
PerceptionSource
The Planetary Society, How much did the Apollo program cost?
Estimates that Project Apollo cost about $309 billion in 2025 dollars; the figure changes with the base year and what is counted.
PerceptionSource
The problem of many hands
When harm comes from a system many people built, victims are left without knowing at whom to point a finger.
SystemsSource
The renaming of the Gulf of Mexico as the Gulf of America (2025)
US agencies and map services adopted the new name for US users, while the Associated Press, Mexico and the UK refused.
PerceptionSource
Thomas Edison
The American inventor, a reminder that promises about classroom technology have a long record of running ahead of what happened.
Systems
Toppo, The End of Homework? (The 74, 2026)
A July 2026 report on US teachers rethinking, scaling back or dropping homework because AI can finish it in seconds, with survey figures on cheating.
SystemsSource
Trump, address to the UN General Assembly (22 September 2026)
The speech, as quoted by PolitiFact, in which the president said the word artificial makes AI sound fake. The wording varies slightly between outlets.
Perception
Turing, Computing Machinery and Intelligence
The 1950 paper asking whether machines can think, in which Turing admits that machines take him by surprise with great frequency.
SystemsSource
Two lanes of assessment
Secure lanes, done under supervision, where students must show what they can do without AI, and open lanes where AI is allowed and taught.
SystemsSource
Unit 42, DNS Tunneling: how DNS can be (ab)used by malicious actors (2019)
A Palo Alto Networks security team's note calling port 53 'nearly always open on systems, firewalls, and clients'.
Systems
US House Committee on Transportation and Infrastructure, final report on the Boeing 737 MAX (2020)
Rather than one culprit, the committee named a chain: faulty technical assumptions, a lack of transparency by management, and insufficient oversight.
SystemsSource
Vann, Of Rats, Rice, and Race: The Great Hanoi Rat Massacre
The historian's account of the 1902 Hanoi rat bounty, found in French colonial archives. The story could not open the article; the rat farming rests on his interviews.
SystemsSource
Waldseemüller, Universalis Cosmographia (1507 world map)
The 1507 world map, held by the Library of Congress, that first printed the name America; his 1513 map dropped it.
PerceptionSource
White House, fact sheet on inaugurating the era of Super Intelligence (2026)
The administration's case for the new name: these systems are not simply imitating or automating human intelligence, so artificial is the wrong word.
Perception
White House, fact sheet on the China state visit (2026)
The administration's statement that the two leaders agreed to use the term super intelligence rather than artificial intelligence.
PerceptionSource
Wiener, Some Moral and Technical Consequences of Automation
Warns that if we use a machine we cannot interfere with once it starts, we had better be sure its purpose is the one we really desire.
SystemsSource
Willingham, Critical Thinking: Why Is It So Hard to Teach?
Argues that you can only think critically about what you know.
SystemsSource
Wizards of the Coast, banned and restricted announcement (13 January 2020)
When it banned a card, the makers of Magic: The Gathering did not accuse anyone of cheating; the combination appeared more often than is fun.
SystemsSource
Working papers and preprints
A study shared publicly before other scientists have reviewed it. It can be careful and still prove wrong, so its numbers are signs, not settled facts.
Systems
Yan et al., Distinguishing performance gains from learning when using generative AI
A paper on telling better performance apart from real learning when students use generative AI; the story takes from it that performance gains are not learning.
SystemsSource
YouGov daily question on the name for AI (28 September 2026)
Of 21,449 US adults in an opt-in online panel, 53% preferred artificial intelligence and 9% preferred super intelligence.
Perception
Nothing matches all of those. Try removing a filter.
This library grows with every story. It holds only what a published story cites, each in our own words, with a link to the original.


