Nexus

Harari's key move in Nexus is a reframing that, once you see it, changes how you think about everything from the printing press to ChatGPT. His argument: the fundamental unit of civilization isn't the individual, the nation, or the institution — it's the information network. Myths, religions, legal codes, scientific journals, social media algorithms — these are all information processing systems that connect millions of strangers into coordinated (or manipulated) action. The history of civilization is the history of these networks: who builds them, who controls them, and what happens when they scale beyond human comprehension.

What makes this framing so powerful for thinking about AI is the distinction between 'tool' and 'network.' We habitually talk about AI as a tool — something humans use to accomplish tasks. But Harari argues that AI is becoming something fundamentally different: an autonomous information network that can make decisions, generate narratives, and shape human behavior without human oversight. This isn't science fiction; it's already happening with recommendation algorithms that no one at Meta or TikTok fully understands. The shift from 'AI as tool' to 'AI as network' is not just semantic — it changes the entire risk calculus. Tools serve their users. Networks develop their own dynamics.

The historical parallels Harari draws are illuminating but also humbling. The printing press was supposed to spread truth and enlightenment; it also spread witch-hunting manuals and war propaganda. Radio was supposed to connect humanity; it enabled both FDR's fireside chats and Nazi mass mobilization. Every new information network in history has been simultaneously liberating and dangerous, and the people living through the transition could never fully predict which way it would go. We are living through such a transition now with AI, and Harari's point is that the historical pattern should make us humble about our ability to steer the outcome.

The concept that unsettles me most is Harari's characterization of AI as potentially the first 'alien intelligence' — not alien in the extraterrestrial sense, but alien in the sense that it processes information in ways fundamentally unlike human cognition. Previous information networks (religions, bureaucracies, markets) were built by humans and ultimately interpretable by humans. AI systems increasingly are not. When a neural network makes a medical diagnosis or a trading decision, even its creators often cannot explain why. This opacity is not a bug to be fixed — it may be an inherent feature of intelligence at scale. And if we build civilization-level infrastructure on systems we cannot interpret, we are in genuinely uncharted territory. Harari doesn't offer solutions, but his framing of the problem is the most honest I've encountered.