Atlas of AI: The Book That Made Me Uncomfortable About Technology (in the Best Way)

Book by Kate Crawford · Published 2021 · Yale University Press

Kate Crawford’s landmark work doesn’t ask whether AI is smart. It asks who pays the price for it.

I came across Atlas of AI the same way I discover most things – through a rabbit hole. A LinkedIn post, then a podcast episode with Kate Crawford, then the realization that I needed to actually read the book. Living in New York, surrounded by people who work in tech and talk about AI like it’s the best thing that ever happened to humanity, this book felt like a cold shower. A necessary one.

Crawford is a researcher at USC and Microsoft Research, but she writes like a journalist. The book is not about how AI works technically. It’s about what AI costs – in lithium, in labor, in data, in power. And once you see it that way, you can’t really unsee it.

          AI is not in the cloud. It’s in the ground.

One of Crawford’s most striking moves is to open with a visit to a lithium mine in Nevada. That’s where she starts – not with a server room, not with a chatbot. Because the first thing she wants you to understand is that AI has a physical body. It consumes land, water, minerals, and energy at a massive scale. Every time we use a « smart » feature on our phone, something somewhere is being extracted.

This materialist approach cuts through the usual abstraction around AI. We talk about « the cloud » as if it’s weightless. Crawford shows it’s anything but. It’s a useful correction, especially for those of us in marketing and business who tend to talk about digital transformation without ever thinking about what’s underneath it.

« Every AI system is built on a supply chain of extraction – of data, of labor, of natural resources. »

         The labor no one talks about

The chapter on data labor is what stuck with me most. Behind every « intelligent » AI system are thousands of workers – in Kenya, in the Philippines, in Venezuela – who spend their days labeling images, moderating violent content, flagging misinformation for cents per task. Crawford calls this « ghost work. » It’s invisible by design, because acknowledging it would disrupt the narrative that AI is autonomous and neutral.

Coming from France, where labor rights and worker protections are taken seriously (sometimes to a fault, I know), reading about these conditions hit differently. The contrast between how AI is sold – efficient, clean, empowering – and how it’s actually built is stark. And it raises real questions about what it means to talk about « ethical AI » when the foundation is this kind of invisible exploitation.

        Power is the point

What makes Atlas of AI more than just a critique is Crawford’s insistence that this isn’t accidental. The concentration of AI power in a handful of companies and states is not a bug – it’s a feature. Whoever controls the data, controls the prediction. Whoever controls the prediction, controls an increasing amount of how the world works: what you see online, whether you get a loan, whether you’re flagged by a government system.

The EU is pushing back on this with the AI Act, and France in particular has been vocal about AI sovereignty. But reading Crawford, you start to wonder if regulation alone is enough when the power asymmetry is already this deep.

      Is it perfect?

Honestly, the book can feel heavy at times. Crawford is thorough to the point where some chapters read more like academic papers than narrative nonfiction. If you’re looking for actionable frameworks or business playbooks, this isn’t it. But if you want a book that shifts how you think about what AI actually is – not a technology, but a political and economic project – then it’s hard to find a better starting point.

Sources:

• Crawford, K. — Atlas of AI (2021, Yale University Press)

• EU Artificial Intelligence Act (2024)

• Dhruv Ghulati — Interview with Kate Crawford, Tech Policy Press (2021)

• Gray, M. & Suri, S. — Ghost Work (2019, Houghton Mifflin)