After NEXUS: Three Books That Reshaped How I Read AI
After NEXUS: Three Books That Reshaped How I Read AI
Have you noticed how we talk about artificial intelligence as though it were a single coherent force, rather than a collision of incompatible ideas, economic interests, and philosophical traditions? I spent years absorbing technical papers and vendor whitepapers, convinced that deeper understanding of neural networks or transformer architectures would clarify the stakes. Then I read Blake Blake's NEXUS, and realised I had been reading the wrong literature entirely.
That book — which traces how connectivity itself rewires human cognition, from writing to the internet — cracked open something I'd been missing: AI isn't a technology problem. It's a social one. The algorithms matter less than the power structures they embed. And once you see that clearly, three other books become essential reading, not as supplements to technical knowledge, but as corrections to it.
The first is Yuval Noah Harari's Sapiens. I know, I know—it's been on every LinkedIn influencer's shelf for a decade. But most people read it for the narrative sweep, the story of how humans conquered the world. What I needed from it, and what I finally understood after NEXUS, was something else: the observation that large-scale cooperation depends on shared myths. Money is a myth. Nations are myths. The idea that humans are inherently rational is a myth. So when we deploy AI systems at scale—when we ask algorithms to make decisions about credit, hiring, criminal justice—we're not implementing neutral logic. We're encoding myths about human worth, economic value, and social order into machines, then hiding behind claims of objectivity. Harari shows that this sleight of hand is ancient. Recognising it in AI requires seeing how myths have always driven human societies, long before we had neural networks to blame for our choices.
The second is Cathy O'Neil's Weapons of Math Destruction. This one is more direct: it's a catalogue of how algorithmic systems, even well-intentioned ones, amplify inequality and lock people out of opportunity. What struck me most was not the examples—they're damning and necessary—but her argument that the power asymmetry is structural. We deploy algorithms without understanding them, without consent from those affected, and without meaningful accountability when they fail. And because they're opaque, people can't fight back the way they could against a human decision-maker. O'Neil is writing about the governance problem, not the technical one. And that's the conversation we're not having loudly enough. We debate whether AI is "safe" or "aligned," as if safety were a property of the code rather than a question of who controls it and for what purpose.
The third is Bruno Latour's We Have Never Been Modern. This is the hardest of the three, and the one I return to most often. Latour argues that our entire intellectual tradition rests on a false divide: nature versus culture, science versus politics, fact versus value. We pretend we can separate them, keep them in different boxes. But we can't. Every scientific fact is entangled with social interests. Every technology embeds politics. And when we build AI systems, we're not separating intelligence from values—we're weaving them together, then claiming we've done something neutral. Latour's insight is that we need to stop pretending. We need to acknowledge that every system we build is a hybrid, a mixture of the technical and the social, and we need to govern it as such. No amount of alignment research will change that fundamental truth. What changes is whether we admit it.
These three books sit uneasily together. Harari is a historian, O'Neil a data scientist turned critic, Latour a philosopher of science. But they converge on a single point: the AI crisis is not about intelligence. It's about power, myth, and the stories we tell ourselves about objectivity. If you're reading only technical literature on AI—papers on scaling laws, constitutional AI, mechanistic interpretability—you're missing the essential context. You're reading as though the problem were solvable by engineering alone.
That doesn't mean the technical work is irrelevant. It means it's insufficient. We need both: people who understand the mathematics deeply enough to ask hard questions about what systems actually do, and people who understand the social consequences well enough to demand accountability. We need engineers who read Latour. We need policymakers who understand neural networks. And we all need to stop pretending that AI is a neutral tool waiting to be "aligned" by the right incentive structure.
The uncomfortable truth these three books converge on is that alignment is not a technical problem we can solve and move on from. It's a permanent tension between what systems can do and what we collectively decide they should do. That tension requires ongoing negotiation, not a final solution. It requires admitting that every choice we make about how to build and deploy AI is a choice about what kind of society we want to live in.
Which is why I keep returning to them. Not because they give me answers, but because they ask the right questions.
Have you read any of these? What's shifted in how you think about AI after encountering work that challenges the technical frame?