Breaking the Word Lines: Bill Burroughs vs. AI textuality
Back in the hip 1960s, the renegade US writer William Burroughs came out with some sharp commentary on language as a virus and writing as a source of mind control. Contrast these with LLMs.
The US “beat” novelist William Burroughs proposed literally to break the word lines — interrupting conventional word combinations that acted as thought viruses to insinuate fixed ideas into our reading practices. By cutting up or arbitrarily disjointing written phrases and then piecing them together again in typewritten texts, he attempted to neutralize these viruses.
The idea was to create new meanings from bits and pieces of language torn from well-oiled syntagms and recombined in arbitrary ways. Writing became an experiment in manipulating paranoid linguistic textures, almost certainly influenced by the use of psychotropic drugs.
For him, obvious similes, typical phrasing, well-known idioms, and conventional expressions — all thought “viruses” — influence our traditional thought patterns.
If you want to abolish the way these viruses automatically control our typical thoughts and imaginations, Burroughs reckoned you have to literally break them up and repurpose them. Naïve psychology, perhaps, but much inspired by the powers of the typewriter and tape recorder — pre-digital technologies that can record and reorganize speech/text content.
This artistic venture was similar to the way painting and sculpture’s lines had been “broken” in the post-1900 Western art world. Realistic portrayal of nature and humans through form and coloring rapidly took a back seat. Traditional visual “lines” were broken, materials, colors, and media were mixed more radically, and we began to “read” artworks in new ways.
Operating in pre-digital days, Burroughs literally chopped up pages from physical books or sheets of typed paper with a pair of shears and remixed the bits (via operations such as cutups or fold-ins) into pages of new, arbitrary word lines (he never used the term “sentence” of course).
Chosen word lines would then be retyped as a fair copy, either with their phrases running onto each other to form new sentences, or broken into phrases… separated by dots… or dashes to create… a scatter of linguistic images… and corresponding ideas and sensations.
When read as a text, the whole sequence would either congeal into meanings/images in the reader’s mind or simply exist as a weird illegible word soup mixed into the implicit flow of the text.
For example, we might pick out from the previous paragraph the phrase “dashes to create” as a complete word line that opens up new images and meaning fields. Silence to say goodbye, a famous Burroughs leitmotif, began life as a cutup.
Today, the recent burst of AI-driven neural textuality (using LLMs), now available as a digital toolset, casts an interesting new light on how we use language to create or influence ideas in people’s minds.
We now have a machine that starts with a vast, wild mix of word data (in fact patterns of letters or characters across millions of parameters) and then digitally synthesizes this word/character mass into smooth, grammatical word lines. These will be exactly geared to standardized prose conventions, inherited from the mainly public discourse scraped from textual information sources on the Web.
This output is an exact reversal of the Burroughs cutup/fold-in method of creating new word lines. He started from conventional order and laboriously broke it up; LLMs start from a chaos of digitally coded characters/letters and manage to synthesize standard, often bland prose styles in the blink of an eye. Text as a service.
The question now is: will digital viruses be designed and launched to break at scale the lines that compose these multi-billion word sets? Will they be able to generate brand new, original expressions of ideas from the mix? Would this ability require excessive compute power to scale? Could it transform a database of all the world’s text into a new site for experimentation, as different linguistic viruses explore different pathways through the word lines? And if so, would this represent a threat or a promising new pathway?
The key problem with text media today is the time and concentration it takes to read: humans can listen to background music when reading, and you can now hear text as speech, but these make it difficult to scan and peck at vital information. We will surely invent ways (visual, linguistic, multimedia?) to further streamline the reading of textual knowledge as summaries or précis while augmenting understanding. Or maybe switch to visual wizardry to translate some word lines into multi-dimensional imagery.
When a new writing app comes along (a NakedLunchGPT?), we could benefit from renegade machines that attack those hidden viruses in some new way. Or will we be just content with reading automated industrial prose at scale and speed, viruses and all?
