There is a powerful, alternative way to interpret Large Language Models. Rather than viewing them strictly through the lens of artificial general intelligence, they can be seen as a massive, modern attempt to agglomerate all the world’s information into a single place. The dream of a universal library, or a world brain, is not new — it is an ancient human ambition that is finally finding its ultimate form not in paper or microfiche, but in data graphs.
I. The Ancient Ambition: Libraries as Centres of Conquest and Wisdom
The desire to gather all human knowledge began thousands of years ago, though early attempts were rarely neutral. The first large library, the Library of Ashurbanipal, was the result of conquest and the aggressive acquisition of material from defeated kingdoms; people did not voluntarily hand over their clay tablets.
Other ancient hubs sought permanence and total domain. The Ramesseum, the library and archive of Pharaoh Ramses II, was designed to be eternal, housing the medical and literary texts of ancient Egypt.
The most famous of these institutions was the Library of Alexandria, part of the larger Alexandrian Museum. Alexandria was a literal attempt to acquire all the world’s knowledge. The rulers even instituted a book tax, where ships entering the harbour had their books seized and copied before being returned. This aggressive pursuit sparked rivalries, notably with the Library of Pergamon in modern-day Turkey.
The lineage continued. Beyt al-Hikma, the House of Wisdom in Baghdad, collected classical knowledge from the Roman Empire and translated works from Greek, Persian, and Syriac into Arabic — not merely a translation centre but a generative one, producing breakthroughs in medicine, mathematics, and astronomy before its destruction. The Bodleian Library in Oxford systematically gathered the knowledge of the Eastern world and preserved vital classical connections to science and mathematics. And today most nations maintain a national repository; the Library of Congress, the largest physical library in the world, houses over 178 million items, and by 2012 its digital footprint alone was estimated at 15 terabytes.
II. From Microfilm to Google Books — the Twentieth-Century Pivot
As human knowledge expanded exponentially, physical structures could no longer hold it all. The concept shifted from a library of physical books to compact storage devices.
In 1939 H. G. Wells proposed the idea of a World Brain — a universal library stored entirely on microfiche. Wells envisioned a universal encyclopaedia encompassing all human knowledge to foster world peace, a concept many now compare to Wikipedia. This directly influenced Vannevar Bush’s seminal 1945 Atlantic essay, As We May Think, which introduced the Memex — a conceptual machine that could store all the world’s knowledge on microfilm and let users navigate information by following interconnected trails. It was not until the advent of modern computers that Bush’s vision became technologically viable.
Fast-forward to 2002, when tech giants tried to realise the dream. The Google Books Project was a massive effort to scan millions of volumes from university libraries; public libraries were to receive one free terminal to access this digital universe. But copyright and legal battles ultimately gridlocked the project. While it was technically possible to put millions of books onto a single station, the law prevented it from scaling commercially. This complex battle over fair use and copyright is captured brilliantly in the documentary Google and the World Brain, and it directly mirrors the copyright debates we are having today regarding artificial intelligence.
Because data and facts cannot be copyrighted, organisations will inevitably find legal, low-cost ways to extract the underlying facts from copyrightable material under fair use. The universal library will exist online, but its structure will look entirely different from a collection of digital books.
III. Science Fiction Becomes Science Fact
The concept of an all-knowing repository has always captured the imaginations of science-fiction writers. Isaac Asimov famously wrote about the Encyclopedia Galactica in his Foundation series, a concept later echoed by David Brin in his Uplift trilogy. Gordon R. Dickson pondered a similar structure with the Final Encyclopedia. Conversely, Borges’s The Library of Babel — an infinite library containing every possible combination of letters — feels increasingly real today as AI engines generate vast amounts of digital slop.
Yet we are watching science fiction slowly become science fact. While some argue the universal library failed, initiatives like the Million Book Project and the Internet Archive prove that the physical-to-digital transition is very much alive.
The real evolution, however, began in 2012 when Google introduced the Knowledge Graph. Meta, Yext, and Wikipedia followed quickly. Today, knowledge graphs are the hidden backbones for search engines, recommendation systems, and modern AI.
IV. The Convergence — LLMs and the Universal Knowledge Graph
With more than 13 zettabytes of data currently floating across the web, information has outgrown the boundaries of any physical structure. We may never build a universal library in the traditional sense, but we are actively building a Universal Knowledge Graph.
Large Language Models should be viewed as linguistic unifiers of knowledge rather than sparks of general intelligence. They are inherently language-oriented; the more data you feed them, the more they reflect the emergent properties of language itself.
The next frontier lies in the deliberate unification of LLMs and structured data. Researchers are actively mapping this out, in comprehensive studies like Unifying Large Language Models and Knowledge Graphs: A Roadmap and surveys exploring LLM-empowered knowledge-graph construction. We are already seeing this play out in two ways: architectural protocols — frameworks like RAG (Retrieval-Augmented Generation) and the Model Context Protocol that surgically connect AI to high-quality databases — and domain-specific graphs, such as the world’s first open-source Anti-Money-Laundering knowledge graph from Almtrix, or LLM-driven knowledge-graph reasoning for 3D city modelling.
Ultimately, LLMs and knowledge graphs will merge completely, creating a seamless system for universal knowledge. When that happens, we will finally have access to a desktop machine much like Vannevar Bush’s original Memex — except instead of clicking through microfilm, we will interface with an interconnected web of the world’s collective intelligence.