Learn AI · Level 4 · Expert view · step 6 of 13
Essays that shaped AI: 10 must-reads, from Turing to today
Some ideas in AI are quoted so often that knowing them is part of being fluent in the subject. These ten essays and papers, from 1950 to 2025, are where those ideas come from. They're all free to read, and we've balanced the optimists with the sceptics so you can make up your own mind.
Where it began
The two documents that founded the field.
The question that started it all
Beginner1. Computing Machinery and Intelligence by Alan Turing (1950)
The paper that started it all. Turing replaces the question "Can machines think?" with a practical test, the imitation game (now the Turing test), then answers nine objections to thinking machines, several of which people still raise today.
- Surprisingly readable, and often witty
- Introduces the Turing test
- Predicts "learning machines" decades early
Price: Free to readOn: PDF
Where "artificial intelligence" got its name
Beginner2. A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon (1955)
The short funding proposal that coined the phrase "artificial intelligence". Its famous claim, that a group of scientists could make significant progress in one summer, shows both the ambition and the over-optimism that have run through AI ever since.
- Where the term "artificial intelligence" was born
- Only a few pages long
- Its topics, from language to self-improvement, are still on the agenda
Price: Free to readOn: PDF
How modern AI works
Three essays that explain why today's AI works the way it does.
How AI changes software
Intermediate3. Software 2.0 by Andrej Karpathy (2017)
A short, influential essay arguing that neural networks are a new kind of software: instead of programmers writing rules, they collect examples and let the computer find the program. It explains why AI is changing how all software gets built.
- A clear mental model for how AI changes programming
- Short: a 10-minute read
- By a founding member of OpenAI and former Tesla AI lead
Price: Free to readOn: Web
Why scale keeps winning
Intermediate4. The Bitter Lesson by Richard Sutton (2019)
A one-page essay by a pioneer of reinforcement learning with a hard-won lesson from 70 years of AI: methods that use more computing power, such as search and learning, have always beaten methods that build in human knowledge. It is the idea behind scaling up.
- One page, endlessly quoted in AI labs
- Explains why "bigger" keeps winning
- Written by a Turing Award winner
Price: Free to readOn: Web
The best deep explanation of ChatGPT
Intermediate5. What Is ChatGPT Doing … and Why Does It Work? by Stephen Wolfram (2023)
A long, illustrated explanation of how ChatGPT produces text one word at a time, from probabilities and neural networks to embeddings and training, written by a scientist for curious readers. The best deep explanation of language models for non-specialists.
- Step-by-step, with many diagrams
- Explains why next-word prediction works so well
- Also published as a short book
Watch out: Long, and some sections get mathematical; skim those.
Price: Free to readOn: Web
The big debates
Read these in pairs. Each side makes its strongest case.
The case that AI could take off fast
Beginner6. The AI Revolution: The Road to Superintelligence by Tim Urban (Wait But Why) (2015)
A long, illustrated, very readable two-part essay that introduced millions of people to the ideas of AGI and superintelligence, and why some thinkers believe AI progress could suddenly accelerate.
- The most accessible introduction to AGI debates
- Funny stick-figure drawings make big ideas stick
- A good way into the expert-level arguments
Watch out: Written before ChatGPT; it presents one side of a debate, so pair it with "AI as Normal Technology".
Price: Free to readOn: Web
The case that it won't
Intermediate7. AI as Normal Technology by Arvind Narayanan and Sayash Kapoor (2025)
Two Princeton researchers argue that AI is a powerful but normal technology, like electricity or the internet, whose effects will unfold over decades as society adapts, rather than a coming superintelligence. The leading counterweight to "AGI soon" essays.
- A careful, evidence-based alternative to AGI predictions
- Practical on policy and safety
- By the authors of the book AI Snake Oil
Price: Free to readOn: Web
The optimist's case
Intermediate8. Machines of Loving Grace by Dario Amodei (2024)
The Anthropic CEO's long essay on what could go right: how powerful AI might speed up biology and medicine, mental health, economic development and governance. A detailed version of the optimistic case from inside a frontier lab.
- The most detailed optimistic case from a lab leader
- Concrete about specific fields such as biology
- Useful for understanding why labs race ahead
Watch out: Written by the head of an AI company; read it alongside sceptical views such as "AI as Normal Technology".
Price: Free to readOn: Web
The critic's case
Intermediate9. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? by Emily M. Bender, Timnit Gebru and colleagues (2021)
The research paper that gave critics of large language models their best-known phrase. It argues that ever-bigger models carry environmental costs, absorb biases from their training data and produce fluent text without understanding, and asks who bears the risks.
- The classic critical view of large language models
- Raised issues of bias and cost that are still debated
- Readable for a research paper
Watch out: Written before ChatGPT; some of its claims about what models can do are contested today.
Price: Free to read (open access)On: Web, PDF
The most memorable metaphor
Beginner10. ChatGPT Is a Blurry JPEG of the Web by Ted Chiang (2023)
The science-fiction writer compares a language model to a compressed photo of the internet: good at reproducing the gist, prone to making up details where the compression lost them. A memorable way to understand hallucinations.
- One of the most-quoted explanations of what LLMs are
- Beautifully written by a celebrated author
- Short and non-technical
Watch out: On The New Yorker, which limits free articles per month.
Price: Free within The New Yorker's monthly limitOn: Web
How to read this list
- Short on time? Start with The Bitter Lesson (one page) and ChatGPT Is a Blurry JPEG of the Web (ten minutes).
- Want to understand language models properly? Wolfram’s What Is ChatGPT Doing explains it step by step; pair it with our guide AI vs machine learning vs deep learning and the glossary entries for transformer and embedding.
- Want the big debate? Read The AI Revolution and AI as Normal Technology back to back, then see where today’s models stand on the AI ladder.
These essays are part of the Expert view level of our Learn AI path. For whole books, see the best AI books.
How we chose
We picked pieces that are widely cited by people working in AI, changed how people think about the field, and are free to read online (one is on The New Yorker, which allows a few free articles a month). Several authors work for, or lead, AI companies, which we note on their cards. We review the list every quarter. Is there an essay you’d add? Tell us.