AI’s raw nerves
In November 2023, Sam Altman’s sudden sacking and almost immediate reinstatement as head of OpenAI left the world holding its breath. The episode seemed like a Silicon Valley soap opera. In reality, as Karen Hao demonstrates in *Empire of AI*, it was the moment when the failure of governance at one of the world’s most influential technology organisations became impossible to ignore. Behind closed doors, a handful of billionaires and their ideological allies were deciding the fate of a technology destined to redefine work, knowledge and power relations on a global scale – and even the company’s own employees were kept in the dark.
Hao, a journalist at the Wall Street Journal and formerly of MIT Technology Review, has been covering OpenAI since 2019, when the company was still a relatively unknown start-up. The result is an investigation of rare depth that weaves together three narrative strands: the company’s internal history, from its foundation with Elon Musk to its transformation into a commercial behemoth worth tens of billions; a reconstruction of the scientific debate on artificial intelligence, from the origins of connectionism to the current dominance of deep learning; and a field investigation spanning Chile to Kenya, Arizona to South Africa, into the tangible impacts that the race for generative models is having on the most vulnerable communities: accelerated mining, data centres that are depleting water resources, and content moderators exposed to traumatic material for a few dollars an hour.
The central theme of the book is a critique of the very concept of artificial intelligence as a rhetorical device. Hao points out that the term was coined in 1956 by John McCarthy as a marketing ploy: the word ‘intelligence’ lends the technology an aura of inevitability that obscures its true nature as a sophisticated statistical calculator. This confusion between pattern matching and genuine cognition is not harmless: it fuels the anthropomorphisation of models, justifies the replacement of human labour and, above all, legitimises the concentration of unprecedented resources in the hands of those who promise to ‘recreate’ intelligence without there being a shared scientific definition of what it actually is.
It is here that the book strikes a raw nerve that goes far beyond the – albeit compelling – corporate narrative. OpenAI’s dominant vision – namely, to build an artificial general intelligence capable of matching or surpassing human performance in ‘economically relevant’ tasks – presupposes a notion of computational intelligence that is isolated and measurable through benchmarks. But decades of research in cognitive science, neuroscience and the economics of knowledge tell a different story: human intelligence is inherently relational, embodied and culturally situated. It develops and is enhanced through interaction with environments that are complex from an aesthetic, creative and social perspective, which train the brain to manage uncertainty and novelty. This participatory and distributed dimension of cognition is precisely what generative models, by their very design, do not capture and cannot replace.
Hao realises this when, in the epilogue, he contrasts OpenAI’s ‘imperial’ model with the experience of Te Hiku Media, a Māori organisation that has developed a speech recognition model for its own language using just two GPUs and 310 hours of audio recorded with the community’s full consent. It is the image of an artificial intelligence that functions not as a substitute for human intelligence but as its amplifier: small-scale, specialised, and rooted in a living cultural ecosystem. A model that echoes what the most advanced research on skills is discovering: that environments capable of stimulating complex thought, experimentation and creativity produce cognitive abilities that are transferable to any field, and that it is these ecosystems – not machines – that are the true driving force behind human development.
