The essays on AI that held up

Most writing about AI is out of date within a year. These 40 pieces are the exception: essays and short papers that still say something true, some of them written decades before the technology they describe. Each has a line on how it has held up: what it got right, what it got wrong, and why it is still worth the time.

The list is a judgement, not a survey. It leans towards judgement, work and learning, because that is what this research is about, and it includes the optimists and the sceptics side by side. For books and studies rather than essays, see essential works, the evidence register and the shorter list of books on AI and human skills.

Judgement and automation#

What happens to human judgement when a machine takes the routine work. The oldest question on this list, and the one the rest keep coming back to.

Man-Computer Symbiosis, J.C.R. Licklider. IRE Transactions on Human Factors in Electronics, 1960. The first serious case for people and machines thinking together rather than one replacing the other. More than sixty years on, it is still the right question.

Some Moral and Technical Consequences of Automation, Norbert Wiener. Science, 1960. Warned that a machine will do what we asked rather than what we meant. That warning is now a whole field.

Ironies of Automation, Lisanne Bainbridge. Automatica, 1983. Automate the routine and people get worse at the hard moments, which are the only ones left to them. Drift, described forty years early.

Humans and Automation: Use, Misuse, Disuse, Abuse, Raja Parasuraman and Victor Riley. Human Factors, 1997. The four ways people get automation wrong. Most AI rollouts manage at least two of them.

Conditions for Intuitive Expertise: A Failure to Disagree, Daniel Kahneman and Gary Klein. American Psychologist, 2009. When expert judgement can be trusted: a regular environment and honest feedback. The test every AI deployment should be put through.

The Human Factor, William Langewiesche. Vanity Fair, 2014. Air France 447 told in full. Still the best account of what automation does to the skill it was meant to support.

Why Doctors Hate Their Computers, Atul Gawande. The New Yorker, 2018. How a system built to serve professionals ended up being served by them.

Moral Crumple Zones, Madeleine Clare Elish. Engaging Science, Technology, and Society, 2019. When the system fails, the nearest human takes the blame. Worth reading before signing off a human in the loop.

Thinking, learning and attention#

What happens to the mind that hands its effort to a tool, from search engines to language models.

Is Google Making Us Stupid?, Nicholas Carr. The Atlantic, 2008. Mocked at the time, and much closer to right than its critics.

Google Effects on Memory, Betsy Sparrow, Jenny Liu and Daniel Wegner. Science, 2011. We remember where to find things rather than the things themselves. A small study with a long afterlife.

The Shallowness of Google Translate, Douglas Hofstadter. The Atlantic, 2018. Fluency without understanding, explained five years before it became everyone's problem.

ChatGPT Is a Blurry JPEG of the Web, Ted Chiang. The New Yorker, 2023. The clearest picture of what these models are, and it has aged better than most expert commentary.

There Is No A.I., Jaron Lanier. The New Yorker, 2023. Treat it as a way for people to work together, not as a creature. Unfashionable then, more useful now.

Why A.I. Isn't Going to Make Art, Ted Chiang. The New Yorker, 2024. Art is a long series of choices, and the tool makes them for you. The same is true of a strategy paper.

Writes and Write-Nots, Paul Graham. paulgraham.com, 2024. Stop writing and you stop thinking. The whole argument in a few hundred words.

Work, jobs and expertise#

What automation does to work: the predictions, the economics, and the expertise underneath.

Economic Possibilities for our Grandchildren, John Maynard Keynes. Essays in Persuasion, 1930. Right about the productivity, wrong about what we would do with it. The best warning against predicting leisure.

On the Phenomenon of Bullshit Jobs, David Graeber. Strike! magazine, 2013. Work that even the people doing it believe is pointless. AI can now produce it at scale.

Why Are There Still So Many Jobs?, David Autor. Journal of Economic Perspectives, 2015. Automation takes tasks, not whole jobs. Still the best answer to the job-loss headline.

Moore's Law for Everything, Sam Altman. moores.samaltman.com, 2021. The builder's case for sharing the gains. Worth reading against what the industry went on to argue for.

The Turing Trap, Erik Brynjolfsson. Daedalus, 2022. Imitating people or extending them. The SuperSkills argument in an economist's language.

Centaurs and Cyborgs on the Jagged Frontier, Ethan Mollick. One Useful Thing, 2023. The plain-English account of the most important workplace AI study so far.

AI Could Actually Help Rebuild the Middle Class, David Autor. Noema, 2024. The optimistic case that expertise becomes cheaper to acquire. It holds only if judgement survives.

The Simple Macroeconomics of AI, Daron Acemoglu. NBER Working Paper, 2024. The sober economist's number, and the useful counterweight in any boardroom.

The technology, and where it is going#

How the machines work and what their builders and critics expect. Read for the reasoning; the dates are where most of them go wrong.

Computing Machinery and Intelligence, Alan Turing. Mind, 1950. Replaced whether machines can think with a test of behaviour. We have lived inside that swap ever since.

The AI Revolution: The Road to Superintelligence, Tim Urban. Wait But Why, 2015. The piece that took the exponential argument mainstream. Read it for how the future felt then.

Software 2.0, Andrej Karpathy. Medium, 2017. Saw that we would stop writing programs and start training them.

The Bitter Lesson, Rich Sutton. incompleteideas.net, 2019. Two pages that anticipated the next five years better than anything longer.

Deep Learning Is Hitting a Wall, Gary Marcus. Nautilus, 2022. Published months before ChatGPT. Wrong about the wall, still right about reliability.

What Is ChatGPT Doing … and Why Does It Work?, Stephen Wolfram. Stephen Wolfram Writings, 2023. The best long explanation of how it works, for anyone prepared to concentrate.

Situational Awareness, Leopold Aschenbrenner. situational-awareness.ai, 2024. The boldest timeline anyone published. Read it for the argument, not the dates.

The Intelligence Age, Sam Altman. ia.samaltman.com, 2024. The industry's optimism in its shortest form.

Machines of Loving Grace, Dario Amodei. darioamodei.com, 2024. The fullest optimistic case from someone building it. Read it next to Narayanan and Kapoor.

AI as Normal Technology, Arvind Narayanan and Sayash Kapoor. Knight First Amendment Institute, 2025. The strongest sceptical case, and the one to read next to the optimists.

Power, governance and risk#

Who decides, who carries the consequence, and what a technology quietly settles before anyone votes on it.

Do Artifacts Have Politics?, Langdon Winner. Daedalus, 1980. Technologies carry decisions inside them. Somebody chose the default.

Technology and History: Kranzberg's Laws, Melvin Kranzberg. Technology and Culture, 1986. Technology is neither good nor bad, nor is it neutral. Still the best single line on the subject.

Five Things We Need to Know About Technological Change, Neil Postman. a talk in Denver, 1998. Every technology is a trade. Ask what it takes away before you ask what it gives.

Why the Future Doesn't Need Us, Bill Joy. Wired, 2000. Wrong on the timing, right that the people building it were asking the wrong question.

On the Dangers of Stochastic Parrots, Emily Bender, Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell. FAccT, 2021. Named the risks of large language models before most people had used one.

Gradual Disempowerment, Jan Kulveit and others. arXiv, 2025. People lose control not through rebellion but by handing over decisions one at a time. Drift, at the scale of a civilisation.

The Urgency of Interpretability, Dario Amodei. darioamodei.com, 2025. The head of a frontier lab saying plainly that nobody yet understands how these systems reach their decisions.

Rahim’s own#

The essays from Box of Amazing that answer the pieces above, and the research pages that develop them.

The Architecture of Drift (2026). Where the drift versus design matrix comes from.

Are You Flying, Or Are You Being Flown? (2026). Algorithmic drift, through Air France 447. The answer to Bainbridge and Langewiesche.

The Missing Rungs (2025). What happens to the route into senior work when the junior work goes.

Synthetic Seniority (2026). Work that looks senior from someone who cannot yet judge it.

Solving Synthetic Seniority (2026). The last thirty per cent, and why it has to be earned.

The Crossing (2026). The SuperSkills ladder.

The Thing That Proves You're Human (2026). Outsourced recognition.

Rules Before Tools (2025). Decision rights and named owners before the next model release.

The New Imposter Syndrome (2026). Phantom authorship: not being sure the work is yours.

This Is Zombie Work (2025). Graeber's question, a decade on and automated.

Show Your Working (2026). Proof of work as a professional problem.

What I Tell Parents About AI (2026). Thinking first, tools second.

On this site: Drift versus design, The missing rungs, The judgement premium, Experience debt, Silent failure, Decision provenance, Human Reserved, The five rungs of AI use, The Altitude Lens, The source rule.

How this list is kept#

Pieces are added when they have had time to be tested by events, which usually means at least a year. A note changes when the verdict does. Suggestions are welcome through the contact page.