Vannevar Bush wrote an article 80 years ago that I would like you to read. It serves as a call to the scientists of the time to focus on the vast store of human knowledge and make it more accessible now that the fervor of WW2 has passed. Bush writes mainly about two things:
- A minituarized camera: Walnut-sized, forehead-mounted, and activated by a cord running down your sleeve. The future scientist photographs everything that they find interesting simply by looking at it. Thoughts can be added by voice or note when time and situation allows it, but they are stored separately.
- The memex: An intimate mental scaffolding for associative thinking. It is mechanized, the size of a desk, and holds all information that you need for research work: books, records, notes, all your communications, etc. Its main feature is to forge "trails": a list of records the user considers related.
The article can be found here, but it is not required for this essay.
When I read the article, I took my usual collection of notes, filed them away, and moved on to the next item on my reading list. However, its content has occupied my thoughts more than any other article I've read of recent. It overlaps a lot with the Zettelkasten method1 which I've been using for a little more than a year, so it struck some notes with me. Comparing my current workflow with what Bush proposes, there are similarities, but there is room for improvement, especially so if we play with the possibilities of the future.
I have some thoughts about the future of knowledge work and the overlap between the Zettelkasten method and the mechanized memex. What follows is a non-exhaustively attributed mix of my and Bush's lines of reasonings on the topic.
The funny little camera
Today the article has a humorous angle due to the dated technical assumptions, but the methodology presented for recording what you're currently looking at for future review has undeniable value. This is evident to anyone who has ever used a conventional camera. Cameras are now small enough to be embedded into glasses. They are digital as well, so we don't have to insert a new roll of film after 100 photos, either. While no model is likely to offer a cord to pull on, I presume they are even easier to trigger, if not recording continuously.
So we have arrived at the future, but while the scientist isn't necessarily confined to their desk to take records, some questions remain:
- What do we do with the images after they are taken? Where do they go?
- How do we combine photos and notes in practise?
- Should the audio recordings be transcribed?
- Will the photos even retain their significance over time?
An image taken of a memorable moment can have a half-life longer than a life-time, but what about the numerous photos taken at work? If you usually take a hundred photos a day and look at one from a month ago, is its context immediately clear? The attached notes undoubtedly help, but not exhaustively; there is always a non-zero information loss over time. Consider also selection: how do we reach back in time and find a set of interesting photos that we know were taken three months ago? Are they indexed by date? Project? Both?
Let us extrapolate into the future on the same track as Bush. Instead of only photographing what we're looking at and adding notes later, what if we record the thoughts we associate with the photo as it is taken? If consciousness is defined by electrical activity, the signals can be measured and stored as a snapshot of mental activity. Further still, if consciousness is sufficiently mapped out, the same signals can be replicated. So: we augment the photo (or video) with our transient thoughts at the moment, archive it, and allow ourselves to revisit it later. Now that our thoughts are recorded, the images' half-life should be significantly longer than the order of months. Let us further speculate that the information loss over time has been reduced to the point of being negligible, for sake of argument.
Conventionally, we document thoughts using one of the following methods:
- writing them down;
- illustrating them; or
- by speaking them into a microphone.2
Optionally, some metadata is added to point to a page or figure, rather than copying it. This is subsequently filed away, perhaps given a pass or two later to fill in details. But all of these methods come with an impedance mismatch: transforming a thought into a persistent medium (using your own words) takes effort. This effort is not without intrinsic value, however: applying it forces you to actively understand rather than passively filing something under "Sure, I get it."3
Active application of effort, in order to create something, appears to be is a critical component when it comes to learning.
Off the top of my head I cannot think of something I've learned passively, without any form of feedback loop created by applying effort.
I have attended too many lectures where a passive absorption of information via monotone monologues was the idea, but I don't recall learning anything from them.4
Lectures were effort was required, usually by interaction with the lecturer, are those I remember (and cherish).
Let us return to our thought-reading device. For sake of practicality, let this be all it does: what we are currently looking at is encoded in there somehow also. Further, presume also that the question of "How are these thoughts replayed?" has been answered: we have gained the ability to replay recorded memories with perfect recall. With perfect recall, there is no impedance mismatch when transforming from persistent storage; no information discontinuity. Gone are audio journals, pen and paper, and the need to break the flow while reading something: just capture every new piece of interesting information and proceed with the next.
Jogging while reading a book (listening, perhaps) is a now a valid research method. Performing routine tasks becomes an even more legitimate method of assimilating new information, as spontaneous insights arise at the periphery of consciousness, while connections to existing knowledge are cataloged in real-time.
But where did the conscious engagement go? While creating this incredible device, we seem to have removed the application of creativity otherwise necessary for absorbing new information. Now when we finish a good book, the result is some opaque dataset of countless records in archive that we need to do something with.
Bush writes:
[The researcher's] excursion may be more enjoyable if he can reacquire the privilege of forgetting the manifold things he does not need to have immediately at hand, with some assurance that he can find them again if they prove important.
With our perfect recall of records we have achieved maximum possible enjoyment: with full assurance of recall we can confidently forget everything we learn. Great, but how do we "find them again?" What filter do we construct and apply on our archive to pull out that small piece of information that we know is in there somewhere? In short: how do we select?
Effort is required to read a book; a paragraph, a figure, a whole chapter must be comprehended; there is no such thing as a free lunch. As we engage with a book and internalize its content we effectively create a personal edition of it, tuned to our modes of thinking. This is no different from reading a book today, other than the resultant dataset contains some information that is difficult to capture today (read: a digital representation of our exact thoughts). This remarkable dataset has potential, we just need to discover how best to utilize it.
Let me introduce an analogy: A forest is a collection of knowledge; its full extent difficult to comprehend due to the density of its trees. When you or someone else learns something, a path opens up throughout it; those 800-page textbooks result in notably long paths. Unless we occasionally traverse these paths, nature will reclaim them, impairing our ability to navigate through the knowledge space.
If an individual's path is undisclosed and remains so on their passing, it is quickly obscured. To prevent this, it is essential to revisit each others paths regularly; bringing the original trailblazer along as a guide ensures you avoid any pitfalls along the way.
But haven't we been here before?
Isn't this path a bit boring? Monotonous? Wouldn't it be more fun to walk a different trail every time?
Investing excessive time in revisiting old knowledge may leave little room for new growth. Is it possible to intertwine these trails, allowing them to cross and intersect, so that each venture into the forest is unique, information half-life be damned?
Enter the memex
The memex, as presented, is a piece of furniture that you work at. It is not a computer — it certainly does not have an Internet connection — but it is not very far off. It is purported to have storage (of microfilms) large enough for you not to worry about limitations. A user feeds this machine the books they read, reports they write, communications had with peers, and photos taken with the funny little camera (along with the notes). All entries are indexed with codes accessible in a desk-mounted book for reference.
The furniture features two interactive screens for displaying and composing records. Its main feature, as previously mentioned, is to help the user remember associations. In the course of research they will consult books, converse with colleagues, attend conferences, and spend a lot of time thinking and doing experiments. Those inclined towards engineering may write engineering notebooks instead.
Within the memex, read and written records are sequentially associated with the press of a button; the result is a new record that Bush refers to as a "trail". The trail is manipulated like any other record: pages can be turned to, and notes can be written into margins. It is your train of thought (and work) transformed into a persistent form. Any time a book or page that is included in a trail is revisited, users can instantly re-engage with and extend them. How is that for making a walk throught the forest more interesting?
The human mind works by association and the memex exploits it. Unlike traditional methods where information is indexed by name and number in complex folder structures, the memex indexes by how you associate items, supplemented by unique codes. The system employs a flat data structure; there is never any confusion over where a document is located. While putting all documents in the same box does increase accessibility, it does not ease the selection process. An "index record" (a repurposed trail) can then be used to filter a subclass of information. Ideally, something akin to tags would solve the initial dive into the archive, but their implementation in a mechanical memex can be complex; as Bush presents the machine, subclass information is specified top-down in trails rather than bottom-up in the records themselves.5
So to answer the question, "How do we select?" We associate. By associating relevant records, we effectively expand the area where our search is most likely to yield results, similar to identifying a tree in a forest that has been painted to stand out. In order to start our search inside this area, or somewhere close to it, we add subclass informations to the records (using tags or an index record) as they are fed to the memex.
Further, as Bush's article highlights, if you during a lunch conversation recall having created a relevant trail years ago, you can duplicate it and forward it to your peer; they'll feed it to their own memex, extending it as they see fit. And so the forest becomes easier yet to navigate.
The preceding paragraphs provide a summary of Bush's memex, a concept that was technically achievable 80 years ago using the conventional technology of the time — with the notable exceptions of the forehead camera, dry photography, and highly dense microfilms (but these were not far away). Significant technical advancements have been made in the last 80 years, including the advent of digital photography; what then does the modern memex look like?
The modern memex is no longer a piece of furniture: it is a computer program with Internet access. It can consume records of any type: documents, videos, whole databases, links to external media, among others. Each record is automatically assigned a Universally Unique Identifier (UUID). You don't have to remember or write any UUIDs down: it is there as metadata only, a bookkeeping detail in the background. You instead assign a title and some tags to a record as it is fed to the memex:
- Moon surface data from the Lunar Reconnaissance Orbiter
:space:lunar:dataset: - Course of Theoretical Physics by Landau & Lifshitz
:book:pdf:physics:math: - Notes from NixCon 2024
:conf:nixos:computer-science:@berlin:
The selection process is straightforward: by entering any substring of the title or tags in any sequence, users receive a list of potential matches:
- Forgotten the name of the mission that mapped the moon? Find the dataset by searching "
:space: surface moon". - In need of Laundau's first book on physics but can't remember his name? Search "
theo phys :book:". - Want a compiled list of all conference notes taken in Berlin? Enter "
:conf:@berlin:".
Narrow the search further by adding date constrains.
Sharing a research trail is as easy as forwarding all the relevant files. One of these contains the trail itself, enumerating all associations by simply linking to the right UUIDs. Additionally, modern trails are further augmented by containing executable, possibly interactive, code.6
This is what my current setup behaves like. What it lacks is two monitors side-by-side for display and notetaking, but for the moment I prefer a single monitor and writing down my notes on paper. This is a custom solution: you'll find a plethora of memex implementations online; some that I are aware of, but haven't tried7 are (in order of top-of-my-head recommendation):
A look into the future
[Prophecy] based on extension of the known has substance, while prophecy founded on the unknown is only a doubly involved guess.
In the future, as we achieve the capability to record thoughts without mechanical aids, we will seamlessly associate them with existing knowledge and frequently visit random locations in the forest, keeping nature's entropy at bay.
The creative processes we feared to have removed will still be there: Technology marches faster than evolution; we will be using pen and paper for another hundred years,8 but the notebooks are now truly ephemeral; the act of composing our thoughts remains irreplaceable, but the laborious task of formatting them has been rendered obsolete. Like the use of complex numbers of toggling between time and frequency domains, pen and paper serve as an intermediate tool. But just like today, the most significant problems will still find resolution in moments of introspection, often during a shower.
Footnotes
1 Which I have previously covered in my previous essay.
2 This list has not changed by a lot over the last 80 years, other than the digitalized aspects.
3 This can be compared to using a marker pen in a textbook to highlight an interesting section. This is a passive act: nothing new is created.
4 Only after my time in the receiving end of lecture halls have I find a method I'd like to try: reading up on (and studying) the material before the lecture, and utilize it for review instead (and there ask all pertinent questions).
5 Compare this to finding a specific vacation photo: bottom-up the image is tagged with whomever were along on the trip, and when/where it was taken; top-down the photo is somewhere in a folder called Photos.
6 See Jupyter and Babel in Emacs' Org Mode.
7 That would require me to leave Emacs.
8 Or something akin to it: like seamless holographic writing using our fingers, implemented in the most basic of brain implants.