Nexus — A Brief History of Information Networks from the Stone Age to AI

From Sapiens to Nexus

Yuval Noah Harari’s career has been defined by the ambitious synthetic overview: Sapiens covered the full scope of human history; Homo Deus projected into the future; 21 Lessons for the 21st Century addressed the present moment. Nexus is his return to the large framework, this time applied specifically to the question of information: how have human societies managed information across their history, and what does the emergence of AI mean for that management?

The central argument is organised around a distinction that sounds simple but has significant implications: the difference between intelligence and reliability. Previous debates about AI have focused on whether AI is intelligent — whether it can think, understand, create, reason. Harari argues that this is the wrong question. The right question is whether AI is reliable — whether the information it produces and the decisions it makes can be trusted.

This reframe matters because intelligence and reliability are not the same thing. A system can be extraordinarily intelligent — capable of generating sophisticated, convincing, beautifully constructed information — while being systematically unreliable. It can produce information that is wrong, biased, or deliberately misleading with exactly the same fluency and apparent authority as it produces accurate information. The reliability question is, in Harari’s analysis, the question that determines whether AI is a tool for human flourishing or a threat to it.

The Historical Framework

Nexus earns its subtitle — “a brief history of information networks” — by tracing how human societies have managed information from the Palaeolithic period through the present. Cave paintings were information networks; the first writing systems were information networks; the printing press was an information network; social media is an information network. Each transformation in the dominant information technology has reorganised how societies think, how they make decisions, how they maintain accountability.

Harari is particularly interested in the error-correction mechanisms that different information networks develop. Healthy information networks — the ones that produce reliable information over time — have mechanisms for identifying errors and correcting them: scientific peer review, journalistic fact-checking, judicial processes for establishing facts, democratic deliberation. Information networks that lack these mechanisms tend to produce and amplify misinformation, to the detriment of the societies that depend on them.

The historical examples he uses — the Catholic Church’s control of information in medieval Europe, the role of the printing press in the religious wars, the specific dynamics of early social media — are not original research but are synthesised to support the analytical framework with the accessibility that characterises all of Harari’s work.

AI as a New Kind of Information Network

The book’s central claim about AI is that it represents a qualitative break from previous information technologies rather than a quantitative extension of them. Previous technologies — including social media — amplified human information. AI generates information autonomously. This is the distinction Harari considers most important.

Previous technologies created challenges for information reliability by distributing existing human-generated information more widely and rapidly, enabling the amplification of misinformation that human actors had produced. AI creates challenges for information reliability by producing new information that no human actor generated — text, images, analyses, recommendations — with a quality that can be indistinguishable from human production but without the human accountability that usually attaches to information production.

The accountability gap is the concern. When a human journalist writes a false story, there is a person accountable for the falsehood. When an AI system generates false information — at scale, continuously, across platforms — the accountability question becomes extremely difficult to answer.

Democracy and AI

The most urgent section of Nexus concerns the specific threat AI poses to democratic information ecosystems. Democracy depends on citizens having access to reliable enough information to make meaningful choices. It depends on shared facts — on a common reality that participants in democratic deliberation can argue about. It depends on the possibility of deliberation being in good faith — on the assumption that participants are trying to understand the world rather than construct alternative realities.

AI’s specific capability — generating convincing but false information at scale, cheaply, continuously — is a direct threat to each of these dependencies. Harari argues that democracies that have been resilient to previous misinformation challenges may not be resilient to AI-generated misinformation, because the scale and quality are qualitatively different from anything the resilience mechanisms were designed to handle.

Harari’s Limitations

Nexus is Harari at his best in terms of analytical ambition and accessible synthesis, and at his most characteristic in terms of limitations. The historical analysis selects examples to fit the framework; critics have noted that Harari’s historiography has always been more impressionistic than academic. The specific predictions about AI’s trajectory are made with more confidence than the epistemic situation warrants.

These limitations, familiar from his earlier books, do not undermine the core value of Nexus. The framework it provides for thinking about AI — as an information network with specific properties that challenge existing reliability mechanisms — is genuinely useful for policymakers, technologists, and citizens trying to understand what the AI moment means.

Our rating: 4.3/5 — Harari’s most politically urgent book. The reliability-vs-intelligence distinction is the contribution that will last.