A Parallax of Tensor Sharding: The Finite Self in a Culture of Infinite Variation, or Library of Babel as Honeymoon Suite.

I went for a walk recently with my closest confidant outside the family unit, A. We have been friends since high school, although the word friend is a term of limited use when describing our bond and our influence on each other’s lives.

On this recent walk, it rained the entire time. It was just wonderful. We traipsed flat marshland and farming outposts, industrial declivities of dirt and rail, as we came across old radar stations and the ruins of stone schoolhouses.

I talked to A about something that had been on my mind that converged with some other things we’d been discussing: what I referred to as the tragedy of the single self.

It occurred to me some months ago, when I was finishing a long piece of prose I’d been working on called ‘Fellow Disjecta, Oh Sunny Danger Time’. Towards the end of the work, I say this:

We become locked into one possible version of who we are, socially expected to stay consistent with this model, made manifest via the seeds of our past for the rest of our days.

In other words, the person we become is the person, for the most part, we are expected to remain as. We can be slightly different versions of ourselves at work compared to home; we might employ parts of ourselves with friends and not with family. But overall, our lifetime is spent mastering the version of ourselves that was set in motion from our earliest attempts at self-composition.

There are those amongst us who do create quite seismic shifts in the configuration of self: more than just a change of aesthetics and personal presentation, we might quite drastically alter the way we speak, how we socialise, who we love, the beliefs that guide our days, and more. When these changes occur, it can create a dislocating experience for everyone involved - family don’t quite know how to respond to this new person, and friends and colleagues are trepidacious. Often, these changes coincide when the opportunity to move to a new city or country presents itself.

However, for most of us, when we commit to who we are, our growth and development across the years is focused on refinement rather than reevaluation. In fact, we try to stay away from large shifts if for no other reason than we have been socialised to avoid doing so: to have multiple personalities is a diagnosable mental health condition. We have many ways of describing people like this, often in terms of the perceived lack of authenticity they have: they’re two-faced, they wear masks, they’re shifty, you never know who you’re going to get when you talk to them.

My consideration of this idea is that, at forty, I will likely be the same version of myself, with little variation, for the rest of my days (for better or worse). This is who my wife married, this is who I am to my children, and this is the self that integrates with work, neighbours and friends. And, more than all of that, this is who I know myself to be - I don’t work on becoming someone else, I just try to smooth out some of my rough edges and learn a few new skills over time.

But there feels to me a particular tragedy in this model: is this really all there is? How many times have I looked at others across the years and thought, I’d like to be more like them - more socially confident, more relaxed, more of this, less of that. Why couldn’t I retain a version of myself while also becoming someone entirely different, a whole new variation of self that I switch to when desired? Is it too much to expect to master being many different types of people when being one is hard enough?

I raise this consideration to draw a line in the sand that we might use as a location marker for what I’m going to say about infinity. If we are a finite self, it feels to me like it now resides in a culture of infinity that will increasingly confront whatever value is found in scarcity.

While the neurodiversity movement was establishing itself and giving language to the many types of minds we have in society, comparing the idea favourably to biodiversity and how rich an environment becomes when it is heterogeneous, so too were online spaces like Tumblr giving a new online language to a paralingua of terms that extended how gender, sexuality, attraction and self could be understood.

It felt like the more that the self was transferred to the abstraction of an online topography, the more that one’s avatar could be composed through fluid and evolving definitions that did away with the tragedy of the finite self; your cyberself could be anything you imagined, and when the offline world rapidly disintegrated to become Chronically Online, this became more mainstream.

Rather than this simply being a thought experiment, let’s make it concrete: When I teach university students about inclusive teaching practices, I no longer talk about core characteristics of autism and ADHD and so on; I talk about neurodiverse and neurodivergent student experiences, for which we become, as educators, interlocuters of observation and consideration. We get to know each student for who they are, not for a rigid schema of characteristics on a continuum of less or more.

This presents benefits and difficulties that are both sourced from the same action: as the old joke goes, stereotypes are a real time-saver. If you have a model of the educational needs of an autistic student in your mind, then it can help you get up to speed quickly when you’re approaching a new autistic student for the first time. However, it can also distort your understanding of who this new student is as you apply templates from past experience to a student for who these may not be suitable.

In this way, we’re confronted with a pedagogical challenge: how to educate students in a neurodiverse paradigm when the focus is on getting to know each student on their own terms and not according to a diagnostic checklist. We are confronted with a version of infinity: if you’ve met one neurodivergent student, you’ve met one neurodivergent student, and as a result, schemas and systems are rendered moot.

In ‘The Twilight of the Idols’, Friedrich Nietzsche said:

I mistrust all systematisers and avoid them.

The will to a system is a lack of integrity.

What do we make of this? In Nietzsche’s eyes, any attempt at putting a structure to the world is an act of fiction and should be acknowledged as such. But is never is acknowledged - systematisers consider their model of reality to be true, that’s the whole point of their endeavour: it’s the whole scientific and mathematical enterprise, be it zoological, economic, sociological, what have you. Whenever somebody says, ‘This process will yield these results, and we have the data to prove it’ or ‘This hierarchy shows the relationship between these elements’, and they try to insinuate that their system has a reality outside of itself, Nietzsche walks the other way.

In the world of artificial intelligence and Large Language Models, two research papers have been released - one from Anthropic, and one followed closely behind by OpenAI - on the topic of systematising, from a reverse-engineering perspective, the way that chat bots think and speak.

Look at the subheading of OpenAI’s paper:

We used new scalable methods to decompose GPT-4’s internal representations into 16 million oft-interpretable patterns.

And, the opening lines of Anthropic’s paper, titled ‘Mapping the Mind of a Large Language Model’:

We mostly treat AI models as a black box: something goes in and a response comes out, and it's not clear why the model gave that particular response instead of another.

and,

Opening the black box doesn't necessarily help: the internal state of the model—what the model is "thinking" before writing its response—consists of a long list of numbers ("neuron activations") without a clear meaning.

So, the papers then go into discussing how they are attempting to identify and categorise ‘features’ - defined by OpenAI as ‘patterns of activity that we hope are human interpretable’ (what a definition!) - of which both papers agree there are millions of these that comprise the way these Large Language Models think.

It isn’t quite an infinity - OpenAI puts the number of ‘features’ of their GPT4 model at around 16 million. So, 16 million identifiable, discernible patterns of activity that can be labelled and give a sense of how the chat bot communicates with us. What will that number be for GPT5, and so on - and, as with all large numbers, when will the numbers become irrelevant, as we simply bask in the immensity of it all: the same immensity that makes the activity of our own brains, like the grains of sand on a beach, feel contained, feel approachably finite.

I’ve been getting this same feeling when I walk into my old university library, recently.

As part of a new weekly habit, I’ve been going early on a weekend to the university library where, two decades earlier, I studied to be a teacher. Sometimes I just walk between shelves and stacks, taking in the immensity of all the books I’ll never read, and I absorb the ambience of the solitude there - I’m not the only one not reading all the books there, I rarely come across another soul in the reading rooms I go to, away from the computer labs and group work settings.

There is, of course, much that has already been said about the difference between the seemingly infinite information superhighway abstractions of the internet compared with the finite being and time that accompanies the physicality of the offline world. For my part, I’ve been relishing both, as I think more about the emerging world of generative cultural infinity compared with the dusty old books I still cling to.

Just the other month I found, at my local bookshop, a beautiful, old and rare set of journals that, in the 1920s, published a pirated version of James Joyce’s ‘Ulysses’ in New York. They were called ‘Two Worlds Monthly’, operated by the polarising figure Samuel Roth. The copies I found, and took home with me, were brought to Australia from New York by an actor now in his autumn years.

Reading through these journals, thinking not only about Joyce but also the other featured acolytes of the literary canon - D H Lawrence and so on - brings into focus the nature of artistic tomes that have provided the evolving soils our culture continues to spring from. It also brings me to my final consideration of finitude and infinity in this essay: the value of scarcity and potential pathways beyond this.

When you consider the work of James Joyce, you consider the four books he wrote: Dubliners, Portrait of the Artist as a Young Man, Ulysses and Finnegans Wake. These four books constitute his literary output. He wrote other things, of course - the play Exiles, poems, letters, assorted prose - but these are by the by. When you read Joyce, you read his four books. You can dedicate a lifetime to reading and interpreting them. In my university library, there is a thick doctoral thesis dedicated to just three pages of Finnegans Wake. We know this; it is par for the course of cultural study.

The same goes for, say, Beethoven - he wrote nine symphonies, so if you want to study a Beethoven symphony, you know the boundaries of what he produced. And, there is an inherent value to scarcity here - the fact that Beethoven wrote nine symphonies instead of nine thousand symphonies means something. It makes a different to how you approach them. You can know the first movement of Beethoven’s sixth symphony very well by listening to recordings of it, by however many different orchestras, many times over.

Again, you can spend a lifetime listening to a finite number of works by Beethoven. In fact, this is how most of us consume culture. We listen to a limited number of tunes, most of which we were introduced to during our aesthetically formative teenage years, over and over again. The radio does the same. In a way, you could say that a limited number of notes in a scale provide a limited number of chords; these are set into a limited number of musical structures with a limited number of instruments; these are then recorded in a limited number of ways, and set into a limited number of playlists which we consume during the limited time we have to consume them.

But what if Beethoven did create nine thousand fantastic symphonies? If an orchestra was putting on a performance of Symphony Number Six thousand, three hundred forty-seven by Beethoven, even if you didn’t know it would you go along because you know Beethoven was a master? Would you just be okay with enjoying something you’d never heard before, because you trust the artist and the typical results?

And what if Beethoven created the same number of symphonies that OpenAI claims its GPT4 model has as language ‘features’ - 16 million. If your music streaming software offered Beethoven Symphony Number Fourteen million, five hundred eighty-two thousand, six hundred thirty-nine, what would you think? How would the lack of scarcity impact your capacity to enjoy the music?

This is the question presented to us by generative art fuelled by artificial intelligence. I heard an educator talking the other day about ChatGPT and how it could write an episode of Paw Patrol but couldn’t write an episode of Bluey. The comment received much applause and laughter from the audience - the point was clear, a children’s television show like Paw Patrol is perceived as simple and formulaic. It wouldn’t be difficult to get ChatGPT to write an episode of it that would be consumable by its audience. However, a show with the soul and sensitivity of Bluey could never be written by a piece of cold, robotic, pattern-based software like ChatGPT.

It doesn’t take much thought to consider that, well intentioned, this is a bit of a limited comment - is it well intentioned because it’s a comment trying to retain the humanity found in flesh and blood people writing stories for other likebodied folks, but is it limited because of the timelines we’re working in: ChatGPT is barely eighteen months old in its public release, and we are already acknowledging it can write a television show, by itself, that consider to be a bit formulaic. But how many months until ChatGPT can readily create a Bluey episode, with all the funny, quirky, human (and dog) moments the audience loves?

For many, it is about the question of whether ChatGPT and other systems will be able to write something with the sensitivity and smarts of Bluey (which, if we consider the pace of its learning and how it is already passing Theory of Mind tests with ease, I’d be hesitatant to bet against it), and also whether we even want this to happen.

On the question of whether we want this to happen, while it seems that all signs would point to No, I suggest that we may already be embracing a Yes.

Take any of those always on, twenty-four seven YouTube videos that play endless live sessions of chillhop study music:

They are very popular. 14 million users subscribe to LoFi Girl, always with tens of thousands of live viewers at any given time across their videos. And what do the videos contain? Chill lofi beats by amateur producers across the internet, songs of around three or four minutes that always contain the same elements: a simple trip-hop drum beat, smooth jazz bassline, gentle acoustic guitar and synth melodies, the occasional vocal sample. Try it by clicking the play button on the embedded video above, and then try it again ten minute later - it’ll be a different song, but the elements are identical, and the variance between tunes is intentionally limited so they will always be enjoyable and relaxing in a muzak, wallpaper kind of way.

At the moment, these channels are still curated based on tunes composed by individual folks across the internet. But they won’t be for long - how many months until there are AI generative channels that take a handful of drum beats, baselines, guitar and synth samples, and put them into a blender that churns out an endless variety of infintely morphing songs. And, I can imagine, they’ll be enjoyable and successful. A small uproar might be present at first, as the individual character of tunes and the composers that made them decry the souless generation of pattern-based songs, but when you think about it, the samples themselves that these tunes employ are pre-recorded and available for just such a remixing enterprise. I think the acceptance of such a music channel will be swift and without much fuss.

So, let’s now take this back to Bluey - while ChatGPT can’t currently write a perfect Bluey episode, the next iteration might. And, when it can, it’ll be able to write an inifite number of Bluey episodes: we won’t be limited to just three seasons anymore, we’ll have millions of Bluey episodes, as many as the computational power can generate and upload to the increasing weight of harddrives and server farms located across global pastoral landscapes (if we think the world is in ecological trouble now, wait until the mass AI generation of high definition video and interactive media calls for the felling of rainforests or the artificial irrigation of deserts or the low-orbital hosting of physical media to hold the gravity of the heavy clouds being formed).

And, just to really drive the point home in case this all seems a bit far fetched, just imagine a Bluey interface where you can upload your child’s face, type in a plot (it could be used as a virtual social story, say your family is moving house and you want an episode that reflects this, with your child and family members mapped onto the characters of Bluey) and enjoy the generated episode that results. This won’t be a far fetched, repulsive cultural product: it will be relished and shared online with abandon (for whatever the monthly subscription model charges).

Listening to Beethoven’s Symphony Number Fourteen million, five hundred eighty-two thousand, six hundred thirty-nine seems a bit absurd, just like if we were browsing for Joyce’s novel number sixty-three thousand, five hundred twenty-four. There is value in scarcity - I don’t want Joyce to have written seventy thousand novels, I just want the four that he wrote. Why is this the case - it is a product of our biology? How many offspring do we produce across a lifetime - somewhere between zero and a half dozen, say. Most folks will have two kids, give or take, and that’s enough: we may be unlimited in how much love we have to give, but when your child is sick in the hospital you dedicate all your finite resources towards their care. This would be a different conversation for a frog laying thirty thousand eggs in shallow water.

Questionable assertions about biology aside, I do feel like we’re entering a slippery slope moment where the repulsion we feel towards an infinite generation of Beethoven symphonies, Chillhop music channels or Bluey episodes might be curtailed by our interest in just putting something on in the background or personalising the experience to our own desires.

And, in time, how much will be remembered of Joyce’s novels and Beethoven’s symphonies - for that matter, what will be remembered of the source material for all those chillhop videos, the J Dilla and DJ Shadow tracks that fuel their inspiration. Will future generations only know the generative outputs that fed on those cultural tomes of the past, and will this matter? It feels like there has ben a distaste for the heavy cultural canon of the past for quite some time - maybe this is how we move on.

I wrote this essay sitting in a cabin on the outskirts of the university grounds, after tiring of the library. I tired of the abundance, the weight of it all. After taking a walk outside, beneath the song of bellbirds, I found a little wooden room on stilts that used to be used for training and events many years ago. Now, it sits empty and, to my pleasure, the doors were unlocked and it was mostly clean inside. I sat and wrote in isolation for the morning, offline.

Kafka has a line about not needing to go out into the world to get inspiration or ideas, that you just have to sit quietly at a desk and let the world come to you. Okay, but now I’m cycling back through my thoughts - the social expectations attached to the finitude of our singular selves; the infinitely divisible ways we can perceive and project ourselves when we change landscapes or interface with the cyberpastoral; the trajectory of neurodiverse pedagogy; the manner in which we engage with four Joyce novels instead of four million; the pathway towards an infinite number of Blueys, all of them wearing our faces and telling our stories, the limited stories of our day to day lives, with the unlimited quantities of love we give to those few people we let in.

As I sit here and consider all this, I think of Nietzsche reading Anthropic’s research paper on ‘Mapping the Mind of a Large Language Model’. “I mistrust all systematisers and avoid them” - I’m with you, Friedrich, but when the frames of their systems encase the boundaries of our culture and consciousness, you have to wonder where the limits of infinity might still be found.