11 hours ago
The new imperialism — AI's true price is exploitation and brutal extraction
An instant New York Times bestseller traces the arc of artificial intelligence not as a story of innovation, but of control: over data, environmental resources, people, and ultimately, the future.
On the surface, isn't this an exhilarating moment?
'Generative AI is thrilling: a creative aid for instantly brainstorming ideas and generating writing; a companion to chat with late into the night to ward off loneliness; a tool that could perhaps one day be so effective at boosting productivity that it will increase top-line economic activity,' writes US tech journalist Karen Hao.
'Under the hood, generative AI models are monstrosities, built from consuming previously unfathomable amounts of data, labor, computing power, and natural resources.'
Hao's description comes early on in her new book 'Empire of AI: Inside the reckless race for total domination', which has caused a sensation in Silicon Valley.
It charts the rise of OpenAI, the company responsible for ChatGPT, and in so doing records how its founder, Sam Altman, and his colleagues have traded early idealism for something much darker.
What started as a non-profit focused on building safe artificial general intelligence (AGI) has rapidly transformed into a profit-chasing, opaque tech behemoth in an arms race against its competitors, which will end — well, somewhere nobody currently is capable of understanding.
The focus of Hao's book: how the AI race is recreating the familiar contours of colonial-era exploitation by constructing a kind of empire in real time, built not on land or oil, but on compute power, data and labour.
And like empires of old, it functions through brutal extraction.
Developing world targeted for dirt-cheap labour
To train large language models like ChatGPT, what is required are humans. Ideally, humans who speak English and are willing to work for a pittance.
In Kenya, Hao reports, OpenAI has outsourced work for its content moderation systems to local workers earning barely more than the minimum wage. Their task: to read and categorise thousands of graphic, disturbing text descriptions so the company can build safety filters for its chatbot.
'Hundreds of thousands of grotesque text-based descriptions,' she writes, have to be sorted into different categories: bestiality; adults raping children.
The job is profoundly psychologically scarring — but what do the tech oligarchs care? It's not Americans doing this work.
'With the many other countries that the tech industry relegates to this role, Kenya shares a common denominator: It is poor, in the Global South, with a government hungry for foreign investment from richer countries,' writes Hao.
Venezuela is another example. Hao explains how the country's economic collapse created a workforce desperate enough to accept almost any wage: 'Venezuela suddenly checked off the perfect mix of conditions for which to find an inexhaustible supply of cheap labour: Its population had a high level of education, good internet connectivity, and, now, a zealous desire to work for whatever wages.'
The book recounts the story of one Venezuelan woman working up to 22 hours a day just to make ends meet. The tasks for which she earned pennies were small, repetitive, and exhausting — labelling datasets, transcribing audio, annotating images. In other words, the invisible labour that makes AI appear magical.
South Africa, of course, has been targeted too, with facial recognition software developers circling the country like vultures in search of valuable data about black faces.
Writes Hao: 'Facial recognition companies from all over the world were jostling to get a foothold in [South Africa] to collect valuable face data, especially after the industry had received significant criticism about their products' failures to accurately detect darker-skinned individuals.'
Environmental toll still unknown
All this extraction requires a physical backbone. Hao devotes a section of the book to the vast data centres that underpin modern AI.
The amount of water, electricity and raw materials required to keep AI systems running at scale is immense, and growing. Altman told a conference in June that a 'significant fraction' of the world's total power should ideally go towards running AI.
'Hyperscalers call their data centres 'campuses' — large tracts of land that rival the largest Ivy League universities, with several massive buildings densely packed with racks on racks of computers. Those computers emanate an unseemly amount of heat, like a labouring laptop a million times over. To keep them from overheating, the buildings also have massive cooling systems — large fans, air conditioners, or systems that evaporate water to cool down the servers,' writes Hao.
These centres require so much water that the tech companies are increasingly looking to developing countries to make a Faustian deal: we'll pay you to host our data centres, and in exchange leach vast quantities of water from your system. (A court stopped a planned Google data centre construction in Chile last year after an outcry from citizens about the water cost.)
Exactly what the environmental toll is is still unknown, because companies like OpenAI refuse to allow close monitoring.
This will all pay off … maybe.
Exactly what is all this harm in aid of?
Productivity and prosperity, we are constantly told.
But Hao writes of a June 2024 global study from the Upwork Research Institute, which found that 77% of workers said AI tools had added to their workload due to the amount of time they now had to spend reviewing AI-generated content while under pressure to do more work.
Hao also cites the Nobel economics laureates Daron Acemoglu and Simon Johnson, who have surveyed transformative technologies throughout history and concluded that they very rarely bring widespread prosperity.
One example: the invention of the cotton gin in the 1790s, which brought farmers untold wealth and established the American South as the largest global exporters of cotton.
Guess who did not benefit in the slightest?
'With the surge in cotton production, enslaved Black people were forced to work longer hours and physically coerced by even harsher means to squeeze out every ounce of their labour.'
This book's warning is not about robots rising up. It is about the humans already in charge.
The real threat, Hao argues, is not future annihilation but present-day exploitation. It is not that AI will become autonomous, but that it is already being wielded by a small elite to consolidate wealth and power.
This is the bargain we are told to accept, writes Hao: 'The staggering price society needs to pay for what it is developing will someday be worth it.'
