Book Review of "Nexus: A Brief History of Information Networks from the Stone Age to AI"

"Nexus: A Brief History of Information Networks from the Stone Age to AI" by Yuval Noah Harari

Overall, I found Nexus frustrating and unhelpful. I debated whether writing a harsh review was worth the effort—after all, perhaps I’m the one missing the points—but I decided to proceed. The book's distortion and incompleteness, paired with Harari’s immense popularity, make it important to at least present an alternative view, even if my reach is minuscule.

The book’s subtitle, A Brief History of Information Networks from the Stone Age to AI, does not match its disappointingly narrow focus. Harari, in the epilogue, argues that history can guide us in the age of AI. While he candidly admits he is not an AI scientist, he leans on his expertise in history. However, the “history” here is almost exclusively political and religious—and even then, highly selective. There is nothing about scientific, technological, or economic history. Nor is there an exploration of the various types of information networks critical to modern society. For example, the role of information in economic contexts—such as sharing the availability of knowledge, goods, services, prices, and quality—is completely absent. To assume one can explain the implications of a powerful technological phenomenon like AI purely through political and religious contexts is naive at best, and dangerous at worst. Ironically, naivety is a crime Harari frequently accuses others of throughout the book.

Yes, there are interesting things I learned. Harari’s exploration of self-correcting mechanisms and their role is one such instance. He contrasts these mechanisms in political, religious, and scientific institutions, pointing out how the U.S. Constitution is amendable while religious texts like the Bible are considered infallible. Similarly, his discussion on decentralized information in democracies versus centralized control in dictatorships is good. But they feel like isolated curiosities rather than a cohesive argument.

This is my third Harari book after Sapiens and 21 Lessons for the 21st Century. My frustrations with Sapiens were similar—centered on its handling of the past—while I found 21 Lessons thought-provoking in its forward-looking speculations. Perhaps Harari is better at looking ahead than reflecting on the past. In Nexus, his bold assertions, supported by selective evidence, undermine the book's credibility. But, armchair critics like me are more equipped to critique explanations of the past, since there is evidence, than any speculations of the future.

Take, for instance, his claim that humans triumphed over Neanderthals due to their ability to create stories. Why not language? Why not the ability to pass on skills or coordinate actions? Harari doesn’t even entertain these possibilities. Similarly, his example of Oppenheimer using atomic knowledge to exert power feels cherry-picked—why not mention Darwin’s theory of evolution, which reshaped thought without seeking control? Cherry-picked stories followed by big assertions is a common theme in the book.

Another glaring omission is the lack of any substantial discussion on language as an information tool. For a book supposedly about information networks, this oversight is baffling. Some sections feel unnecessarily long or tangential, such as the discourse on witches and paper’s role in spreading misinformation.

And Harari’s critique of big tech companies, while valid, is presented without sufficient depth. He asks why they can’t be like healthcare in privacy. But, jumping from there to assume that this is an answer for a business model is frivolous as it fails to analyze innovation or cost structures across the sectors. Such surface-level arguments recur throughout the book, relying more on sweeping assertions than rigorous analysis.

Harari does shine in his speculative moments, but alas they are too few and too late. His analysis of the schism between physical and online identities is thought-provoking, as is his argument that AI’s amorphous nature could pose dangers more subtle—and perhaps greater—than nuclear weapons. Unfortunately, these insights are too few and come too late in the book.

Also, his view of what AI is sweeping and at times confuses with just plain computing.

Harari states that he avoids discussing AI’s positive aspects because tech companies already do so. For me, this is unconvincing. If the book were clearly framed as purely a one-sided scenario, I could consider buying that, but it is not. This one-sidedness distorts rather than disseminates knowledge.

In summary, Nexus delivers little. Its narrow lens and overreliance on selective evidence make it more misleading than enlightening. While there are moments of brilliance, they are insufficient to justify the book as a whole.