Unveiling Universal Knowledge: A Tour of the Library of Unwritten Papers

Fergus Fettes

May 29, 2024

Introduction to the Total Library

The Multiversal Library of Alexandria is housed within the Total Library, otherwise known as the Library of Babylon.

There have been many artist renderings of the library over the years. I’ll be including a few in this talk, but it is, naturally, an impossible task.

The Total Library contains every book that could possibly be written.

Wall 1, Shelf 1, Vol 16

But this is meant quite literally– this is what the typical work looks like. The first generations born in the library naturally despaired at its enormity, at its inherent disorder, at the “millions of meaningless cacophonies, verbal farragoes, and babblings”. The halls were filled with madness, mystery cults and sky burial.

This link is real by the way. There is an online version of the total library, at libraryofbabel.com. You are lucky to have a direct portal to the total library, many worlds do not. Whenever I include an unwritten work, I will try to include a link to the fulltext on the libraryofbabel.

It is not known whether the total library is really infinite or not. In the millenia since it was discovered we have worked tirelessly to map and organize it, but we have not yet reached any kind of limit or boundary, nor have we found a single repeated work.

Let heaven exist, though my place be in hell. Let me be outraged and annihilated, but for one instant, in one being, let Your enormous Library be justified.

One of the early poets of the library wrote these words. I am happy to announce that since that time we have made great progress on the project of justifying the library.

Wall 1, Shelf 3, Vol 30

To give you a concrete example, this book contains only the letter a. So far we have found billions like it, with a tiny variation, a letter e hiding in the pages somewhere, or a space or a dot. And thats just for a– for every letter of the alphabet there are unthinkable numbers of works containing 99.9% the same letter.

We faithfully organize and categorize all that we find. And indeed, over the millenia, the catagolgues have grown to start to challenge the immensity.

Categorization Systems

¹

These catalogues and categorization systems are one of the wonders of the modern library.

1. Cat: CSV0032

¹

1. Cat: CSV0032

¹

Each dot represents between 1 and 128 completely organized and categorized and colocated hexagons of 640 books. The black regions represent nearby regions of works that are not in this category system.

Multiverse library @ 128 Out[7]: 2,392,147,840,696,320 If each dot is between one and 128 complete hexagons.
This also works for the library of unwritten papers:
In [8]: 32 * 20 * 128 * 64691
Out[8]: 5,299,486,720

1. Cat: CSV0032

¹

Here is another example.

1. Cat: P59GG33

¹

1. Cat: P59GG33

¹

1. Cat: P59GG33

¹

1. Cat: P1MC042

¹

1. Cat: P1MC042

¹

As you can see, these categorization systems are not like those you are familiar with from finite libraries here on earth. They are intelligent, self-creating autopoetic categorization systems, they have lifelike features, they grow and change.

1. Cat: P1MC042

¹

The categorization systems developed to such a degree that some of them came to have names of their own. One of the best-developed systems within the total library is known as the Multiversal Library of Alexandria, which is where I and my ancestors have laboured all our days.

1. Cat: TRT0211

Multiversal Library of Alexandria

¹

The Multiversal Library of Alexandria contains every work that could have been published. The Multiversal Library is growing, and perhaps one day it will fill the Total Library, a complete category. For that to happen, we would need, for each and every one of these works, to find a world in which it was publshed– we are trying to justify not just the library, but each and every work in it.

Huge regions of the multiversal library would be incomprehensible to people of your world, xenolinguistic chaos, shadow semiotics.

1. Cat: TRT0211

¹

1. Cat: TRT0211

¹

Each compartment roughly corresponds to a particular worldline.

1. Cat: TRT0211

¹

And here is the one we are looking for.

1. Cat: TRT0211

¹

This small subsection is where my colleagues and I while away our hours.

1. Cat: TRT0211

Library of Unwritten Papers

¹

This is the Library of Unwritten Papers. It contains near-Earth works, works that might, but for some small contingency (or some gross inconveninece) have been written in this world. Perhaps someone got hit by a bus or a meteor on the way to the publishing house. Perhaps they just had a bad hangover. Whatever the reason, here we collect works that very nearly existed.

From now on, which I say ‘the library’, it is the library of unwritten papers that I am talking about.

1. Cat: TRT0211

Works in the Library: Background

The number of papers written every year continues to grow, but is currently somewhere in the region of 3 million.

The number of books is similar, so there are around 6-8 million significant works added to your libraries every year.

• 2-3 million technical papers p.a.
• growing by 4-9% per year
• total number of books similar

Likewise, the number of papers not written each year grows– but several orders of magnitude faster. By some estimates there are over a billion works unwritten every year.

• 2-4 oom more unpublished works
• estimated at >1 billion p.a.

• we group works by similarity to avoid too much repetition

Now, before you suggest that these unwritten works are just minor variations on when we have– we group works by similarity to avoid overcounting.

• exactly where we set this cutoff is a matter of great debate

Lets say you have 10 works. If you require 100% similarity to group them, you have 10 works. If you require 0% similarity, you have one work. In between you have cutoffs at different phonon lines. At 80% you would have 6 works. At 55%, two.

Exactly where we set this cutoff is a matter of great debate. With our current cutoffs we get the following estimates:

• research: each researcher doesn’t publish about 100 papers per year
• books: the number of unpublished books annually is on the order of the human population
• poetry: estimates say 90% of the human population fails to publish a poem every year
• poetry: some outliers don’t write up to hundreds of potential poems a day

[READ THE SCREEN]. I’m not even going to mention twitter or instagram.

To appreciate these numbers it helps to understand the structure of cultural phylogeny. Regular phylogeny has this branching pattern, where entities get further and further from one another over time, which makes it harder for them to hybridize.

Cultural phylogeny on the other hand has this reticulated, self-intersecting, rhizomatic structure. This causes a combinatorial explosion in possible works– even taking, as we are, only unique, original works that are very likely to have been produced.

Works in the Library: Examples

• Investigating the potential of using discarded avocado pits as a sustainable source of activated carbon for water purification
• Kinetic analysis of the thermal degradation of polypropylene-based disposable cutlery under simulated picnic conditions
• The impact of different background music genres on average time spent on hold with customer service
• Seasonal variations in the accumulation of dust on top of refrigerators in single-person households
• Comparative analysis of the bounce height of various brands of generic tennis balls
• Survey of preferred car air freshener scents among morning commuters in Toledo, Ohio
• Observational study of the relationship between shoe size and time spent browsing Instagram

As I said, these are all links to the full work on the online library of babel. This version of the library doesn’t have capital letters or punctuation so the works end up looking a little messy. [read from the paper about dust]

Now, I know for a fact that the library contains many many great works– and there are many fascinating unwritten works about dust in particular, but it also contains every work that wasn’t published for good reason. What did the paper represent I wonder? The fever dream of a bachelors student who dropped out and became a frontend developer? Or the lifes work of a maligned professor, funding cut off too soon by departmental politics? I suppose we shall never know what mishaps led to this paper remaining unwritten, unpublished, unprinted, unread.

You can imagine how refreshing it was for the early librarians, after thousands of years organizing books filled with nonsense or the letter a, to finally get to these regions. So, before we get onto universal knowledge, lets take a ligtening tour through some of my favourite regions of the unrwitten library.

Unwritten Medieval Manuscripts

There are a remarkable number of fantastic works from the medieval and reneissance periods. Indeed, right up through the industrial revolution, the european psyche was rich with invention– I’ll just let some of these titles roll over you:

• Thomas Browne, Mechanomancy: Divinations of the Machine; Explorations into the Spiritual and Metaphysical Implications of Early Mechanical Inventions

• Giovanni di Zaffira, Arbor Vitæ et Sapientiae (The Mystical Tree of Life and Knowledge)
• Isotta Nogarola, De Spiritibus Aetheriis (On the Ethereal Spirits and Their Powers)
• Mary Sidney, Hypatia’s Lament The Lost Wisdom of Antiquity Reclaimed
• John Dee, Nocturnal Utopias Visions of Perfected Realms in the Hours of Darkness
• Bonaventura d’Orta, Omnia Septentrionalis (All Things of the Northern Lands)
• Thomas Browne, Codex Veritas: The Treatise on Universal Harmony
• Christopher Smart, Jubilate Machina: A Poetic Ode to Artificial Intelligence and Divine Innovation

Feel free to read some of them, you can come to my website at your leasure, the library is open to all.

Notable Sections: Early Death

Wall 3, Shelf 2, Volume 26

Many of the unwritten medieval works also belong in the early deaths section, one of the most heartbreaking. Here are the first stanzas of book seven of the Faerie Queene, which Edmund Spencer would have written in 1599, if he hadn’t died at the age of 46– “for want of bread” according to Ben Jonson. Spencer didn’t really get into his stride until book nine, in my opinion, and it is a great loss to all of you that you haven’t had these books these hundreds of years.

I don’t need to tell you that the untimely deaths department contains many of the unquestioned marvels of the library.

Notable Sections: Blasphemy

The blasphemy section is also full of treasures. Here we collect works which would have been written if humans didn’t have this innate fear of being burned alive.

• “De Universali Mundi Harmonia”
  (On the Universal Harmony of the World)
  ~Galileo Galilei, n.w. 1635-1640

“De Veritate et Luminibus Gnosticis”
(On Truth and Gnostic Lights)
~Rene Descartes, n.w.c. 1640

“Threnodia Dei Natura”
(The Lament of Nature’s God)
Erasmus Darwin, n.w. 1802

Let not our quest for knowledge blind the eye,
To beauty’s whisper and to wonder’s sigh,
May we recall, in moments awed and still,
The silent grandeur of Your ancient will.

This epic poem would have prefigured and foreshadowed a lot of the religious and political furore that would be whipped up by his grandson Charles, had Erasmus written it in 1802.

These words were never written out of fear of political backlash. Isn’t that a tragedy?

Notable Sections: Pornography

Of course the blasphemy section isn’t all politics–

Notable Sections: Pornography

everyones favourite hellraiser, a young man called Amadeus–

Notable Sections: Pornography

his scatological interests are well known. But most of his creativity in this area was curtailed by contemporary politics. Not so in our library.

Notable Sections: Pornography

All of Mozarts pornographic works survive here, including Requiem for a torturer, written in the timeline in which, in 1791, instead of dying an untimely death he read Justine by the Marquis de Sade, gave up music and devoted all his creative energies to writing.

Notable Sections: Unfeasible/Illegal

The blasphemy section is part of a larger region for works that were simply unfeasible to produce.

Notable Sections: Unfeasible/Illegal

• The Neurological Impact of Complete Sensory Deprivation on Infants: A Longitudinal Study
• The Effects of Prolonged Exposure to Vacuum on Human Physiology: A Space-Based Experiment
• The Impact of Nuclear Warfare on Global Biodiversity: A Multi-Site Study
• Decoding the Language of Interspecies Communication: A Cross-Species Linguistic Analysis
• The Psychological and Physiological Effects of Cannibalism on Human Development: A Cross-Cultural Study
• Concept Specific Poisoning Attacks on Public Discourse

This is also where some of the best scientific research in the library is found. This paper, for instance, in which they exposed hundreds of volunteers to the vacuum of outer space, or this one where they used medical cadavers to reintroduce cannibalism to rural Ireland.

The beauty of the unwritten library is that we can read these papers without having to suffer through the research.

Let me give you are fuller example from this area. This is a paper I worked on personally.

Why wasn’t this research ever performed? Well, some combination of the obvious– its clearly socially destructive research that would never pass an ethics commitee. Even if running public discourse poisoning was allowed it would be very expensive to perform. The public outrage at the idea of being manipulated for research would be bad enough– but for them to develop a tool to optimize this process would be unthinkable.

But, the intersting thing about papers of this sort is that, despite never having been written, there is some evidence that some of the research described herein has actually been performed. Im talking of unsubstantiated rumors that NATO hires sociologists to run experimental campaigns using russian troll farms, whatever Cambridge Analytica were doing, or what happens behind closed doors in marketing research firms. So it isn’t necessarily the case that this paper was to expensive to perform, only that it would be unnacceptable, in most worlds, for academics to write about it in this way.

One must wander what this world looks like, the world where this paper was written and published.

I’m sure each of you has a work that you would love to withdraw from these shelves. Which reminds me– time is getting on, and we better get to the point.

Withdrawals

Sample Size: 100 1000 1e5 1e6 1e7 1e8 1e9 1e10

Now, you may be wondering, how does one go about finding a paper like the Cyanide paper in the library, how do you put in a request for it, how do you withdraw it.

Of course you could do like the early librarians did– just read through endless papers until you stumble upon a good one. But time is fleeting.

You could try doing a keyword search, limiting your papers to certian domains. This is a slight improvement, but in the end you are still left reading through countless articles trying to find the good ones. The problem is we don’t have citation counts or peer reviews. The papers haven’t been published, so how could they be cited? You can still try adding ‘high citation count’ to your query, and it does work. But we can do better.

The thing is, adding keywords and other filters to your search, it functions as if you are adding information to the document you want to see. And this is the right approach. The more information you put in, the better your results will be. You are trying to weave a specific document from the probability manifold with only your intention as a guide.

¹

You could try– and I would recommend– writing a basic version of the paper you want to see, and ask our friendly librarians assitant here to find it for you.

1. Image credit: exa.ai

¹

But if writing a psudopaper seems like a lot of effort, one of the most reliable techniques is simply to use a paper that you already know and love.

It helps if it is a paper of a certain quality. It helps if it has some deep element of truth to it.

1. Image credit: exa.ai

¹

You take this paper and you do a particular kind of similarity matching, trying to find new papers that are near-identical to the paper you have– but for one crucial difference. You only look in a specific department.

This is the paper I used to start my search for the Cyanide paper. It is about poisoning image models, such as stable diffusion. I looked for papers that were as similar as possible to this, but in the sociology department.

1. Nightshade Paper

¹

And this is what I found– the paper I showed you earlier.

1. Nightshade Paper

Here is another view, showing the differences between the two papers. This is what you want– a paper that is as near-identical as the original as possible, but where all the changes make perfect sense and still tell a coherent story.

Technical Guide
https://www.fergusfettes.com/Essays/Fieldshifting

These Slides
https://www.fergusfettes.com/Talks/FieldshiftingLibrary

If you are interested in finding new works in the library, I have written a more technical guide to doing this yourself. It is available on my website, which you can find by searching for my name in a search engine. The slides for this talk are also there if you want to click through any of the links.

Similarity search: why does this work?

¹

Why does this work? One group of scholars who work on these techniques usually perform their searching using papers in neurology, and target the developmental biology department. They argue that functional invariants between these disciplines as well as a deep symmetry in both hardware and physiology allows for this sort of translation.

It might be something like the following: if the original paper faithfully reproduces some structure that is present in the material under observation– in this case in the collective behaviour of neurons– and if, as we suppose, the world is full of symmetries and harmonies, echos and shadows, if these kinds of reflections and translations are an essential part of the material of the world, then the deep structure that the original paper identified should be present also in other displines and other areas.

Well, this may be true, I hear you object, when shifting between two neighboring disciplines. After all, neurology and cellular morphology fundamentally deal with cells, the work is done with microscopes and the apparatus of science.

1. Machine Learning for Hypothesis Generation in Biology and Medicine

• linguistics ⇋ proteomics
• biology ⇋ philosophy
• economics ⇋ systems biology
• buddhism ⇋ physiology

But we have found that this method, searching using an existing paper, works between very diverse disciplines.

Now, truth is a difficult concept, and as someone standing before you telling you that I work in a near infinite library containing a catalogue called the muliversal library of alexandria where I spend my time tracking down unwritten papers, you may be question whether I have any idea what the word truth means. But I can say this at least.

• linguistics ⇋ proteomics
• biology ⇋ philosophy
• economics ⇋ systems biology
• buddhism ⇋ physiology

¹

If you take a paper in this way and find a reflection in another discipline, you have found a harmonic pair of resonant ideas. It might be, if you are a radical constructivist, that these harmonics only exist in the structures of human knoweldge. That may be. But even if it is so, have you not increased the sum total of knowledge in the world? Not only do you have the new paper, which might be worthy enough to reproduce and eventually publish in this world in some, perhaps modified form. But you have also demonstrated the existence of this sort of wormhole between disciplines, where the local topology of the information is porous, through which you can pass and emerge somewhere apparently (superficially) quite different.

1. Songs of Life and Mind

• linguistics ⇋ proteomics
• biology ⇋ philosophy
• economics ⇋ systems biology
• buddhism ⇋ physiology

¹

And the fact that this works so well across scales and even into abstract disciplines like philosophy implies that these resonant structures really pervade all our intellectual enterprises, that many ideas are structured in a way that allows them to resonate across great intellectual distances. This allows us, the explorers and librarians, to potentially pin down and examine the deep symmetries and bridges across all human knowledge.

1. Songs of Life and Mind

Universal Knowledge

So, to summarise. In exploring the library of unwritten papers, we have found an interesting technique for finding good works. This method works by exploiting deep symmetry between works.

Universal Knowledge

The beauty of the technique is, if you do this successfully, you have not only found the new paper, you have found the symmetry itself. One paper may contain many such symmetries, and a successful translation is a hint at a symmetry.

So dear listener, I hope I have allayed any skepticism you might have had, and I hope I have inspired you to approach this with curiosity and ask– which of my favourite works might also make interesting statements in other disciplines that I am fond of or interested in? In this way, not only do you get to read your favourite papers again with fresh eyes in a new domain, and to learn about a new domain through a familiar lense. You also get a glimpse of a deep symmetry in knowledge.

And perhaps, when you do this, you can try to understand what it is exactly about that particular piece of research that makes it work in this way. What is the deep resonant structure that is forming the bridge. How does this paper, this very particular piece of research on this very particular topic, reflect something deep and abstract that might be true of any discipline that you turn your attention to?

In the particular is contained the universal

I hope this was a tantalizing tour.

In the particular is contained the universal

Naturally half an hour is not long enough to contain more than a glimpse of the library of unwritten papers,

In the particular is contained the universal

which is itself a small speck

In the particular is contained the universal

in the multiversal library of alexandria– itself a shadow in the total library. Many lifetimes will be needed to complete this work. But it is our belief that we will soon have good enough infrastructure and powerful enough tools to read freely from the finest works these libraries have to offer.

If you want to learn more, I encourage you to experiment with the tools available to the multiversal librarian. Finding good works in the library is not yet trivial, but it is addictive and a labour of love, and it is my dearest hope that I will inspire some of you to join the hunt.

Thank you for your time. If you have any questions, feel free to approach me wherever you may find me.

Bibliography

• Machine Learning for Hypothesis Generation in Biology and Medicine: Exploring the latent space of neuroscience and developmental bioelectricity
• Online Library of Babel
• The Total Library
• The Library of Babel
• Categorization Systems
• Borges Library by Various Artists
• Complete Rendering of the History of Categorization of the Total Library
• Nightshade Paper
• Songs of Life and Mind
• Analytical Archaeology, David L. Clarke (1968)

I am grateful for the encouragment and assistance of following typs: Claude-3-Opus, gpt-4, and gpt-4o.