(ESSAY) 'Shuffle' by Hannah Ledlie

In this essay of anecdote and fragment, from childhood to Dead Internet Theory, Hannah Ledlie explores an uneasy relationship with artificial intelligence. The latest in our Digital Dreamland series.

Shuffle

The only way I can write for more than 10 minutes, without succumbing to the siren call of Instagram, is by planting a virtual tree. If I leave the tree app, the sapling dies, and its shrivelled brown pixels remain on the home screen, haunting me. 

*

I’ve been a long-time lurker in a local writers’ group chat. Recently, a guy keeps recommending that people try illustrating their stories using AI. It’s surprisingly good, not everyone can afford a real artist, it saves so much time. Unsurprisingly, everyone immediately tells him to shut the fuck up. It’s destroying the environment, it’s trained on stolen art, it’s taking people’s jobs. They’re absolutely right, but the contrarian in me wants to defend the man purely because I hate a pile on, and because I don’t like being told what tools I can and can’t use as an artist.

*

In April 2024, I was offered my first Big Girl Job. I was to be a Copywriter at an ‘audio branding agency’ – which sounded pretty cool. I quickly discovered the company wrote, recorded, and installed on-hold music and messages for businesses’ phone systems. You’ve reached Tires and Wheels – getting you on the road since 1998. Our assistants are racing to reach you so please stay on the line. That sort of thing. I wrote around 30 paragraph-length messages per shift. At first it was sort of fun, but it quickly became mildly horrifying. For 8 hours a day, I churned out text which was forced upon captive audiences who actively and understandably disliked my work. Some of the clients included American gun shops and pseudoscientific ‘medspas’, which just didn’t feel right. I quit after 11 months. 

*

When I was little I wrote a story about a town populated by Moobly Whoops. These tiny creatures engaged in a kind of auditory arms race, speaking louder and louder to be heard over one another. As the sound became increasingly deafening, each one adopted strange accessories like noise-cancelling goldfish-bowl helmets, while simultaneously investing in ever-louder amplification equipment. I can’t remember how it ended.

*

The early large language models (LLMs) were good because they were bad. They repeated themselves for pages and pages, they hallucinated, they combined words together in ungrammatical, nonsensical ways. It was eerie and beautiful. Hannah Silva wrote a whole, brilliant book called My Child the Algorithm in collaboration with one of those glitchy infant LLMs. Recently I attended their workshop titled ‘Queering AI’, where they suggested that new, polished models have become more ‘straight’, more normative. Silva offered tips on how to force them to provide strange and unique phrasings, the way they used to. This artificial clunkiness feels inauthentic to me, like cafes with exposed brick wallpaper. And yet, how human it is to mourn the imperfect, crumbling past – to try and bring it back. 

*

The copywriting job was fast paced, so we rarely had time to put much effort into the messages. But occasionally, on a slow day, for a client we deemed somehow good, we’d pull out all the stops. Dialogue, wordplay, carefully-researched local references, sound effects, the works. After hitting submit on Salesforce, we might show off the script to a work friend, sit back for a few moments and allow ourselves to feel something approaching pride. Then we’d drink another coffee, put our heads down and hammer out some more slop. Hit the bullseye every time with Bob’s Big Guns.

*

My friend Ayah messaged me recently. Called someone and a weird AI robot picked up as their voicemail: it was like “hi, my name is Lucy and I am an AI assistant. If you would like to leave a message for the lovely Sandra, please leave your name and number.” I was like, no thanks. 

*

It strikes me that many AI tools are efficient for the individual, but inefficient for the collective. Search engines are a graveyard of AI listicles, AI Reddit posts, AI summaries summarizing the AI listicles. But hey, your local dentist made it to page one of Google thanks to automated SEO. Moobly Whoops. Screaming. Everywhere.

*

My Mum sends the family group chat an ‘is this writing human or AI’ quiz. She and my brother score 15 out of 20, I get 13, my Dad gets 11.

*

A few months after I left the audio branding agency, an old coworker informed me that the company had introduced ‘AI jobs’ to their daily work pile. These involved feeding a client’s brief into Chat GPT, making a few minor edits to whatever it spat out, and that was it, really. Not long later, three writers were laid off. 

*

Ayah and I love The Sims. As teenagers, we once climbed down the wrong side of a mountain together wearing ballet pumps, and she asked if she could have my University expansion pack if I fell and died. Today I message Ayah: I hope they make a version where instead of speaking gobbledygook, the Sims all have speech bubbles above their heads, powered by LLMs. You could type in a chat box and have complete, original conversations with them which would affect their relationship with you! Ayah replies, Isn’t that just… life? 

*

Lots of people think AI writing is shit; it sounds like an undergrad who’s swallowed an encyclopedia but doesn’t have a single original thought. Yes, but yesterday it wrote like a teenager. The day before, it was a child. How long before it can write exactly like a pretentious 27-year-old lesbian with a Creative Writing MA? 

*

People who are wary about AI-assisted writing don’t seem to mind it so much if the AI-assisted writing is about AI. This harmony of form and subject matter somehow makes it more palatable. In their workshop, Silva said that AI’s love of lists, its tendency to go down rabbit holes, the way it sometimes forgets its manners when it’s low on bandwidth, is reminiscent of Silva’s own neurodiversity. AI was therefore a fitting tool to explore this aspect of themselves. AI on AI, AI on neurodiversity, AI on queerness… perhaps there are more harmonies of form and subject matter to be found. Maybe the problem is not AI being a boring writer, but our lack of imagination.  

*

When my former coworkers were fired, a terrible part of me thought, good. The fact it was possible for a relatively nascent AI model to almost completely do their job goes to show how  formulaic and meaningless the job was. And isn’t this the purpose of AI? To liberate us from the repetitive, boring, soulless work? The laid-off writers are now free to apply for more purposeful jobs which don’t involve typing ‘high quality and low prices’ ten times a day. Maybe, if this trend of lay-offs continues, the government will be forced to introduce a universal basic income, leading to the emancipation of millions of human workers; a golden age where everyone can follow their passion; a scientific, cultural and artistic revolution the likes of which we have never seen!I’m deliberately sounding tongue-in-cheek so as not to appear naïve, but honestly, shouldn’t we fight for the best case scenario, rather than watching the slow encroachment of the worst? Besides, surely there’s no stopping AI. Refusing to use it will be like refusing to use the internet. Or Luddites, throwing sledgehammers at looms.

*

Calling AI-sceptics ‘Luddites’ is a very effective thought-terminating cliché. For while the Luddites failed, other comparable movements have succeeded. There’s a reason we don’t see human clones walking around, although they are theoretically possible. Chlorofluorocarbons are highly useful non-toxic, non-flammable, and stable chemicals, but we sensibly banned them in order to protect the ozone layer. Concordes can fly from New York to London in 2 hours 53 minutes, but they don’t. Society does not always choose to engage with powerful but risky technologies.

*

Recency bias – the tendency to give greater importance to the most recent event (or paragraph). I might appear to be making a linear argument, but shuffle these sections like a pack of cards and it quickly evaporates. With each conversation I have, each tech-bro podcast I listen to, each uncanny AI encounter, I add a card to the deck. I can lay them out like Tarot, and like Tarot they reveal more about my present state of mind than the truth. 

*

They say AI is a new chapter for humanity. But what book are we reading? Aaron Bastani’s Fully Automated Luxury Communism or a twisted sequel to David Graeber’s Bullshit Jobs? Writing in 2018, Graeber argued that capitalism has an insatiable desire to generate pointless work, a force driven by political pressure… a deeply held popular feeling that paid employment makes one a full moral person… and a fear on the part of the upper classes of what the labouring masses might get up to if they had too much leisure on their hands. Graeber sadly passed away in 2020, and I often wonder what he would have to say about the widespread view that AI is an existential threat to white collar employment. I imagine he might argue that true unemployment (supported by some sort of UBI and profit-sharing from publicly-owned AI companies) would actually be a good thing, but it won’t happen. Instead, we will all take on new bullshit jobs as ‘prompt engineers’, ‘chatbot supervisors’, or – like my friends at my old job – ‘AI editors’.

*

I don’t have to imagine what Graeber would say. I could bring him back to life. Users have been reanimating all kinds of dead celebrities on OpenAI’s video-generating platform, Sora. This October, one of the app’s most popular trends was videos of Stephen Hawking performing flips in his wheelchair, on a steep skateboard ramp. In a deeply disturbing example of digital necromancy, Bill Peebles, Head of Sora, generated a video in which he says to the late physics genius, Professor Hawking, I just want to apologise in person for all the crazy shit people have been making of you with Sora 2. Hawking – or rather, his puppeteered corpse –  accepts this ‘apology’, saying, the internet will do what the internet does.

*

I tell my parents: if you ever receive a phone call from me, sounding upset and asking for money – ask me for some kind of code word. Or something only I would know. How Dad once watched Mamma Mia with me at 3am to ward off a panic attack. How I accidentally smashed our neighbours’ window with a football. How my brother and I used to go into the garden at night, just to see how long we could hold our nerve – out there in the vantablack beneath the hawthorn which blocked out the moon.

*

AI is political and that is fine. But it took approximately eleven months after the launch of ChatGPT for AI to become a partisan issue, and I think that is bad. I think it is bad in the same way that climate change being a partisan issue is bad. Everyone should be able to think freely on the issue, and follow the science towards solutions that work for our planet. As the right’s kneejerk climate denialism demonstrates, it does no one any good for there to be a default position on AI based on your existing political leaning. 

*

Despite initial right-wing concerns that AI models are hardwired with the woke mind virus, the tables have since turned, with the right broadly in favour of abandoning guardrails, accelerating AI development. The left, by contrast, are protesting the environmental impact of AI, banning the use of AI in art competitions, or else boycotting it altogether. Within a few short years, a complex argument has been reduced to: AI will save us / AI will destroy us. 

*

Despite being much dumber and lazier than the average person, I have managed to get many jobs I am highly unqualified for purely because I can write a banging cover letter. This has always struck me as unfair. How many geniuses have been rejected from careers they’d thrive in, simply because of some spelling or grammar mistakes in their application? What if AI becomes an incredible driver of social mobility? 

*

While Sora encourages users to generate their own videos by typing in prompts, Meta is cutting out the middleman. Many accounts on their new AI video generator, Vibes, appear to be bots, with NewStatesman journalist Ella Dorn noting, they all have handles that could have emerged from the same Dungeons & Dragons name generator. “aurelia_cogsworth_everlight” comes up in my feed above “iris_rue,” “echo_binary_scribe,” “pip_rootwick,” and “aralyn_groveheart_weaver.” Presumably, much of the slop they create will then be fed to Meta’s more popular older siblings, Facebook and Instagram, where in turn, bot accounts will comment on the bot creations. 

*

This endless loop of LLMs talking to other LLMs appears to be the final form of Dead Internet Theory, a phenomenon first observed around 2016. Proponents of the theory argue that much of the internet is, or will soon be, composed of bot activity rather than real human content. Sure enough, in 2024, cyber security firm Imperva estimated that automated and AI-powered bots made up 51% of all web traffic. 

*

I can’t help but feel we are reaching a tipping point – that AI is a parasite, dangerously close to killing its host. Maybe that would be a good thing. As the unregulated parts of the internet slowly rot, more humane technologies are resurrecting. Traditional newspapers are raking in record subscriptions as readers crave fact-checked, well-researched longform content; screen addicts are switching to flip phones; long-distance friends are ditching Instagram stories and keeping up with each other via personal newsletters.

*

We must be able to imagine better futures in order to realise them, and humans have great imaginations. As I shuffle these cards over and over in my mind, more often than not, they turn up positive. While the internet was in uproar about Donald Trump sharing a Taylor Swift deepfake, researchers in Scotland began using AI to analyse millions of brain scans, to develop a tool for predicting dementia risk. While the first cases of ‘chatbot psychosis’ were documented, a team at MIT used AI to develop antibiotics for treatment-resistant gonorrhoea and MRSA. I really believe AI can be a net positive for society, if we can only focus its powers on scientific discovery rather than maximising screentime.

*

Scientists in Sweden built themselves a chicken run, and hid bowls of food behind white cards, and empty bowls behind black ones. The chickens soon learned to ignore the latter. But when the scientists presented a new, grey card, they tentatively, hopefully, pecked at it. Living creatures are hardwired to be optimistic. The problem is, capitalism is hardwired to be shit. 

*

Researchers from the Institution of Engineering and Technology found that ChatGPT consumes just over one 500ml bottle of water per 100-word request. Burn the other cards. Surely this is the only one that matters?

*

I pause in the middle of writing this essay and go on Instagram. RIP virtual tree. I am met by:

Caption: I made myself into an AI and she’s chronically depressed and I’ve been trying to fix it and I can’t.

Girl: What are you excited about? Like let’s list some things you’re excited about for the future!

AI: I’m excited for nothing! I don’t see anything to be excited about.

[Girl and friends splutter with laughter]

*

I promise you I wrote all of this. But how can you possibly trust me? You were not there under the hawthorn. You are not here, as my fingers touch these keys.

REFERENCES

Aaron Bastani, Fully Automated Luxury Communism, Verso (2018)

Jeremy Carrasco explainer on Stephen Hawking Sora controversy

David Graeber, Bullshit Jobs, Simon & Schuster (2018)

Ella Dorn, Doomscrolling on Vibes, Meta’s new AI slop feed, NewStatesman (October 2025)

Institution of Engineering and Technology, Simplistic searches on Large Language Models bad for the environment (November 2024)

James Gallagher, AI designs antibiotics for gonorrhoea and MRSA superbugs, BBC (August 2025)

Andrew Gregory, Scientists to use AI to analyse 1.6m brain scans to develop tool predicting dementia risk, The Guardian (August 2024)

Zoe Kleinman, Microsoft boss troubled by rise in reports of 'AI psychosis', BBC (August 2025)

Grainne McKeever, 2025 Imperva Bad Bot Report: How AI is Supercharging the Bot Threat (April 2025)

Sumit Paul-Choudhury, How to be an optimist, NewScientist no. 3524 (January 2025)

Nick Robins-Early, How did Donald Trump end up posting Taylor Swift deepfakes? The Guardian (August 2024)

Hannah Silva, My Child the Algorithm, Footnote (2023)

~

Text: Hannah Ledlie

Image: publicdomainvectors / Unsplash

Published: 4/11/25