Karen Hao2025Reported history
Dreams and Nightmares in Sam Altman's OpenAI
A chatbot feels weightless. Hao's reporting gives it a map: whose words it learned from, whose hands cleaned it up, and whose water and power keep it running.
The core idea
Karen Hao is a journalist who has covered artificial intelligence for MIT Technology Review and The Wall Street Journal, and who profiled OpenAI years before ChatGPT made it famous. Empire of AI follows the company from its founding as a nonprofit in 2015, through the launch of ChatGPT, to the November 2023 crisis in which its board fired Sam Altman and then took him back days later. Hao argues that the industry OpenAI leads behaves like an empire: it lays claim to data, labour, land, energy and water that are not its own, and justifies the taking with a mission to benefit all of humanity. Drawing on interviews with around 260 people, she sets scenes inside the company beside reporting from the places that carry the costs, including data workers in Kenya and Venezuela and communities fighting data centres in Chile and Uruguay. The book ends by asking how that kind of power could be checked and spread out.
The word is chosen with care. Hao is not saying AI companies conquer territory with armies. She is saying they follow an old pattern: claim what others made, pay the people at the bottom as little as possible, and explain it all with a mission so grand that objecting sounds like standing in the way of progress.
That is why the book moves between two kinds of scene. One is the boardroom and the research lab in San Francisco, where the stakes are described as the future of the species. The other is a labeller's desk in Nairobi or a dry town in Chile, where the stakes are a wage, a night's sleep, a water supply. Hao asks the reader to hold both at once.
Old empires took land and minerals. This one takes writing, art, personal data, and the land, power and water its data centres need.
Now: training data, sites, supplyThe researchers are paid like stars. The people who label and clean the data are paid by the task, often far away and out of sight.
Now: annotation and moderationThe taking is framed as a gift: AGI that will benefit all of humanity, if only the right empire gets there first.
Now: the race to AGIInteractive · what's in a prompt
Six stages stand between the world and the answer on your screen. Open each one to see what it takes and who carries the cost, as the book reports it. Tick the questions worth asking, and leave with a list you can use on a vendor, an employer or your local council.
Stage 01 of 06Words and art taken from the web
Stage 02 of 06People who label and filter
Stage 03 of 06Chips, and the money to buy them
Stage 04 of 06Power for the data centres
Stage 05 of 06Fresh water to cool the servers
Stage 06 of 06Owned by a few, closed to the rest
Concept anatomy
HourLife's drawing of the loop the book describes. A mission justifies a race, the race demands scale, scale needs resources from somewhere, and the power that collects at the centre makes the mission sound more urgent still.
The outline of the story the book reports, in dates that are on the public record.
Hao describes a field split over what AGI will do. Look at what both sides agree on.
Shared premise: AGI is near and decisive, and those who build it must be trusted with it. Either way, power flows to the labs.
Community insights
Entry 01HourLife takeaway
Hao borrows the frame from the colonial empires of the 1800s. The resources have changed: people's writing and art, cheap labour, land, electricity and fresh water. The pattern has not. The point is not that the people involved are villains, but that the structure rewards taking.
Entry 02HourLife takeaway
OpenAI bet that bigger models, more data and more computing power would win, and once it moved, its rivals followed it into a race. Hao argues that this one bet, rather than anything inevitable about the technology, is what drives the hunger for data centres, chips and workers. Other ways of building AI existed, and still do.
Entry 03HourLife takeaway
OpenAI was founded to make sure artificial general intelligence benefits all of humanity. Hao reports how that mission was later used to explain secrecy, a for-profit arm and an ever faster race. She keeps asking who is counted in that 'humanity', and who is left out of it.
Entry 04HourLife takeaway
Hao reports on Kenyan contract workers, hired through the outsourcing firm Sama, who sorted violent and abusive text so OpenAI could build a content filter. They were paid a few dollars an hour, and she describes the lasting harm the work did to some of them. The safety the user sees was built on someone else's exposure.
Entry 05HourLife takeaway
Hao reports how Venezuela's economic collapse pushed educated people into piece-rate data labelling on online platforms. She argues that the gap between their pay and a researcher's pay reflects how the industry values people, not some objective difference in the worth of the work.
Entry 06HourLife takeaway
Data centres draw electricity and fresh water, and they are often placed where local people have little say. Hao follows residents in Chile and Uruguay who organised against planned Google data centres while their regions faced drought. The costs of AI land on a map, not in the abstract.
Entry 07HourLife takeaway
Hao describes a field split between people who think AGI will save us and people who fear it will end us. Both camps agree it is near and enormous, and both conclusions hand more power to the few labs building it. In her telling, that split helped set up the 2023 board crisis.
Entry 08HourLife takeaway
A company with 'open' in its name came to share less and less about its models and training data. Hao argues that when a few firms fund most research and decide what gets published, they also shape what everyone else believes about AI. OpenAI declined to take part in her book, so she built it from hundreds of interviews and documents.
Entry 09HourLife takeaway
Hao does not ask readers to wait for a wise leader. She points to artists and writers who sue or strike, communities that demand real terms before a data centre is built, and small projects such as Te Hiku Media in New Zealand, which builds language tools with its community's consent. Power spread across the supply chain is harder to concentrate.
Action steps
Write down every AI tool you touched in the last seven days, what you used it for, and whether the result was worth it. Mark the uses you would happily drop.
In each AI tool you use, find the setting for whether your conversations train future models, and choose deliberately. Note what the tool says it keeps and for how long.
Search local news and planning records for data centre projects in your region. Note what power and water they plan to use, and when the public can comment.
Pick one kind of work you will not hand to a model, such as a letter to a friend or your first draft of an idea. Write it down so the choice is on purpose.
Read one long investigation into the people who label data and moderate content for AI companies. Write down one thing that surprised you.
When a product feels like magic, name one group of people whose labour made it possible. Do it out loud once a week, in a conversation or a note.
Before you open a chatbot, ask whether a search, a template or your own memory would do. Save the big model for the jobs that need it.
When you read a big AI headline, rewrite it as one claim you could check, and note who gains if you believe it. Keep a running list for a month.
Once a season, follow one local decision about power, water or land use: read the agenda, and send a comment or a question if you have one.
Every answer a machine gives you was paid for somewhere, by someone. Ask where, and ask who.HourLife note after Karen Hao
Pick setup actions and daily habits from Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI, choose a start date, and download a printable checklist with real calendar pages.
Karen Hao is a journalist who has covered artificial intelligence for MIT Technology Review and The Wall Street Journal, and who profiled OpenAI years before ChatGPT made it famous. Empire of AI follows the company from its founding as a nonprofit in 2015, through the launch of ChatGPT, to the November 2023 crisis in which its board fired Sam Altman and then took him back days later. Hao argues that the industry...
HourLife’s key takeaways include: The AI industry runs like an empire: it takes what is not its own and tells a story about saving the world to make the taking feel right. Scale at all costs was a choice, not a law of nature. A mission big enough can be used to justify almost anything.
It's a strong pick for readers exploring Become Hard To Manipulate, Digital Literacy, and History Worth Knowing. HourLife distills its core idea into key insights and practical actions worth trying.
List the AI you used this week — Write down every AI tool you touched in the last seven days, what you used it for, and whether the result was worth it. Mark the uses you would happily drop.
About five minutes. The HourLife summary distills Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI into its core idea, 9 community insights, and 9 practical actions you can apply right away.
One idea from one book and one small habit from another, every week or once a month, on the day and around the hour you pick, where you are. Nothing else, and one link stops it.
Print it, pin it, post it. Ways to take Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI off the screen and into the world.
The core idea, key takeaways, and how to apply Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI, as one clean, readable page.
Read summaryEvery action from this page as a printable to-do list, with a 7-day tracker.
Download PDFA shareable card with the book and its top-voted insight, for saving or posting.
The summary card and every quote card in six sizes: Instagram feed, Story, Pinterest, YouTube thumbnail, phone wallpaper, and link preview.
Save the insights that land, build the habits that stick, and keep what you learn from Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI working long after you close the page — inside HourLife.
