Strange Knowledgeability

Source: Venkatesh Rao, Contraptions · 2026-06-26

The book club pick for June was a bit of a wildcard. Robert Darnton’s The Business of the Enlightenment, about the first modern encyclopedia, Diderot’s 18th century Encyclopédie, ou dictionnaire raisonné des sciences, des arts et des métiers. I’ve just finished, and it’s left me with a lot of weird, perhaps ill-posed questions, and strange new beliefs about knowledgeability and a funhouse-mirrors new angle on AI.

I’ve always found the notion that you don’t need to know as much stuff when you can just google everything to be silly. I’ve found the opposite to be true. The more there is to know, the more you need to know. The marginal value of knowing stuff does hit diminishing returns, but the point is much further out than most people realize, and it moves out further with time, as the knowledge environment evolves, not closer. The zero-sum idea that the more Google knows, the less you need to know, is wildly wrong. And it goes beyond the Google-fu (or now prompt-fu) of knowing what specific query to use to probe the knowledge environment. What you know shapes what you can see.

LLMs have pushed this co-evolution between internal and external whats and ways of knowing to a point of crisis, which is what I want to talk about. But first, encyclopedias.

The Darnton book has given me one weird new belief in particular: I now think that the Enlightenment wasn’t so much about a few specific big ideas that challenged medieval orthodoxy, but about an encyclopedic way of knowing about reality. For the first time, it became possible to know so much, you could entirely contain and exhaust normal human levels of uninspired curiosity. You could at least roughly cover your experience of reality with a map of reality. And you didn’t need to be a discoverer of knowledge to do so. Merely an accessor.

It didn’t start with Google. Already in the 18th century, people were forced to ask the same questions we do today. How knowledgeable should you be in an encyclopedic environment? Should you aim to know as much as possible, or as little as possible? Are there things you should try to not know, like Sherlock Holmes with his studiously cultivated lack of astronomical knowledge and all other subjects unrelated to detecting?

Why bother knowing anything when it’s in your personal budget quarto edition of the encyclopédie, which you bought for just a few hundred livres; only a few months of your middle-class income? This quarto edition is what the Darnton book is about. You’d think the publishing history of a particular cheap bestseller edition of an archaic encyclopedia would be a dull subject, but it’s fascinating.

Sherlock Holmes’ knowledgeability was exceptional, but relied on an environment that offered a more banal encyclopedic way of knowing as a foundation. The genius of Holmes could not easily have been expressed in a pre-encyclopedic culture. He had a cache optimized for fast inference in the detection game.

The conflict between Enlightenment and pre-Enlightenment ways of knowing, I suspect, had more to do with the quantity and comprehensiveness of available knowledge than with specific bits of knowledge. The subversion lay in the encyclopedic way of knowing available to all, rather than in specific shocking doctrines held by a few.

Religions don’t have answers to most questions a normal human might think to innocently and lazily ask, so they tend to view unbounded curiosity as a threat and act to curtail, dismiss, or trivialize it. What kind of bug is that? a child might ask. Another of God’s creatures, now get back to your Bible! is no answer at all.

This works pretty well so long as there aren’t that many answers within easy reach anyway. Constraints on curiosity lend a certain sacred mystique to the questions which are permitted, and to which there are answers on offer. The curious mind takes what it can get. When only some questions have answers, those answers seem profound and the bunny trails they open up invite nerdy obsession. Other questions can be marked the work of the devil. Products of an idle mind. One paying insufficient attention to labor and prayer.

Once an encyclopedic way of knowing becomes available, it takes exceptional coercion — think the Inquisition — to limit attention to a few questions. And it takes an exceptionally imaginative person working quite hard to ask well-posed questions that don’t actually have answers. Few people even try, and rarely by accident. The post-curious become the normal type of human. Watch a cat or a monkey. Post-curiosity isn’t a normal state at least for complex mammals.

Starting with Diderot’s encyclopedia, it became possible for even the middle class to own complete encyclopedia sets, and keep inquisitive children fully absorbed until they ran out of energy and attained the nirvana of post-curiosity. To a first approximation, arrival at post-curiosity was enlightenment. A very different notion of it than the one offered by religious mysticism. Illuminated exteriority instead of illuminated interiority.

You didn’t have to beat the curiosity out of children. Thanks to the encyclopedic way of knowing, the frontier of the unknown receded far enough away that most exhausted their curiosity long before they reached it. Most humans became trained to expect that most questions in fact have answers. To believe that there is a place — the library, Google, or an LLM — where one may Enquire Within About Everything, reducing the urgency of actually enquiring about anything.

Religion faces quite a different challenge in a post-curious environment — the questions and answers it offers must compete with a lot of other questions and answers that it cannot successfully starve of attention. When religions compete with an encyclopedic context, the greatest threat they face is not that of contradiction or heresy, but marginalization. The revelation of their sheer lack of interestingness or significance to the majority, relative to the encyclopedic landscape of the knowable.

The history of the actual modern world is mostly the opposite story. People discovering that there’s far more to reality than any one book can possibly cover, and fundamentalists finding it ever harder to claim that one book is all you need.

But maybe a 36-volume “Enquire Within Upon Everything” destination is enough for most humans. Once you live in a world that has one, what do you do next? How do you respond to a curiosity-satiating cognitive environment? How do you stay intellectually alive? How do you transcend the encyclopedic way of exhausted knowingness?

We used to describe people who seemed to know a lot as having “encyclopedic” levels of knowledge. The phrase does not indicate that the person has literally read and retained an encyclopedia (that would be rather sad) but that their curiosity has encountered and survived an encyclopedic environment. Something like Sherlock Holmes’ inventory of knowledge is the right picture to hold of an encyclopedic mind. He could get to the questions nobody else asked, and unmask murderers, because of what he chose to know, which was vast in quantity, but weird in quality. A fictional character yes, but not a bad portrait of an effective mind in an encyclopedic environment.

Encyclopedic knowledge was a compliment because it suggested that the knowledge was a side-effect of grand and ambitious intellectual explorations to the very edges of the known; of unseen worlds traversed; of ways of knowing that the inept questioning of the post-curious could not even begin to probe. And it didn’t have to be as idiosyncratic or focused as that of Holmes.

We’re clearly at the beginning of a new arc in our relationship with disembodied knowledge media, just as in Diderot’s time. That AIs are encyclopedic is not the most important thing about them, but it is a necessary feature. Their other affordances would not be worth much if they weren’t first reliably encyclopedic most of the time.

But the defining feature of AIs is that they offer many ways of encyclopedic knowing. There isn’t just one canonical way to know everything, in alphabetical order, as in Diderot’s encyclopedia. There isn’t even the claim of a particular superior or best way, such as the way claimed by the Encyclopedia Methodique ,which boasted a thematic organization rather than lexicographic as its advantage over Diderot’s original. Or more recently, Wikipedia’s claims of the superiority of folksonomic encyclopedism over the scholarly kind offered by the Encyclopedia Brittanica.

No, an LLM offers you effectively infinite ways of encyclopedic knowing. You can come at what it knows from virtually any direction you can think of, with any ontological orientation, and it will offer meaningful traction. It will not be surprised, though it may flatter you and compliment you on your originality of perspective. You cannot easily catch an LLM wrong-footed, even if you can catch it hallucinating and bullshitting.