On 13 March 2026, an anonymous TikTok account called @ai.cinema021 posted the first episode of a dating show starring fruit. A loud, aggressive lemon. A snooty banana. A strawberry with an ego problem. The whole thing was generated: visuals, voices, script, the lot. It was called Fruit Love Island.
Nine days later it had more than three million followers, which made it, by several accounts, the fastest-growing account in the platform's history. Episodes were averaging over ten million views each. By the time it collapsed the series had passed 300 million views in total.
People argued about the fruit. They picked favourites. They speculated about couplings. Amaya Espinal, who won the seventh season of Love Island USA, said she would never watch it and called the characters problematic, which is a remarkable sentence to have to write about a banana. Joe Jonas and Zara Larsson both engaged with it publicly, and Larsson deleted her post after the backlash arrived.
Then, on 28 March, it stopped. Twelve of the twenty-two episodes had been removed by TikTok, reportedly classified as low-quality AI content. The creator posted a run of furious videos complaining that “like half of my videos got removed” and that people had spam-reported the account. The stated reasons for quitting were hate and no revenue. Fifteen days from launch to abandonment, with a bigger audience than most commissioned television will ever see.
That is the whole argument of this piece in miniature. Something that nobody wrote, nobody performed in and nobody directed captured hundreds of millions of hours of human attention, generated genuine parasocial argument about characters made of citrus, made its creator no money, and was deleted by the platform that distributed it. Every actor in that sequence behaved rationally. The result was still incoherent.
On 29 August 2026, the Daily Guardian, the Iloilo-based Philippine daily, published a piece arguing that this sort of thing is not a passing internet novelty at all. It is the industrialisation of serialised storytelling: the collapse of the distance between having an idea and publishing an episode, and the conversion of the cliffhanger from a narrative technique into an engagement mechanism. The paper's contention was that the machine does not have to feel anything in order for the audience to feel something.
That is correct. It is also much older, much better evidenced, and much more uncomfortable than it first appears.
What The Fruit Knew
Start with the objection everyone reaches for first. Surely we can tell. Surely something made without intention, without a person on the other end who meant it, reads as hollow.
The evidence says no.
In 2024, Brian Porter and Edouard Machery of the Department of History and Philosophy of Science at the University of Pittsburgh published a study in Scientific Reports that ought to be better known than it is. They gave 1,634 participants ten poems each: five by well-known poets including Shakespeare, Byron, Dickinson and Eliot, and five generated by ChatGPT 3.5 in the style of those poets. Participants identified the AI poems at below chance, roughly 46.6 per cent accuracy. They were more likely to guess that the machine poems were human. The five poems judged least likely to be human-written were all by actual poets.
A second group of 696 participants rated the poems on fourteen characteristics without knowing the source. The AI poems scored higher. On quality, on beauty, on emotion, on rhythm.
Then comes the part that matters. When participants were told which poems were machine-made, the ratings fell. Same poems. Same words. Lower scores.
Porter and Machery's explanation is deflationary and convincing: non-expert readers prefer accessible verse that communicates a theme directly, they expect AI poetry to be bad, and so they misread their own enjoyment as evidence of human authorship. The preference is real. The attribution is a story people tell themselves afterwards.
This is not an isolated finding. A body of work on what researchers call the AI disclosure penalty has now documented the same asymmetry across images, paintings, songs and prose. One 2025 study of reader perception shifts, drawing on 990 responses from 261 participants across six kinds of writing, found that disclosing AI involvement erodes perceived trustworthiness, competence, caring and likability, with the steepest falls in social and interpersonal contexts. Participants attributed the drop to a perceived loss of sincerity and diminished effort. Notably, higher AI literacy softened the penalty rather than sharpening it.
Put the two findings together and you get something genuinely strange. The response to the work and the response to the label are separate systems. The first does not consult the second. We feel it, then we decide whether we were entitled to.
The Structures That Do The Work
There is a much older tradition that predicted exactly this, and it did not need a single experiment.
In 1928, Vladimir Propp published Morphology of the Folktale, in which he took a corpus of Russian wonder tales and reduced them to thirty-one narrative functions and seven character types. Villainy. Lack. Guidance. Acquisition of a magical agent. Liquidation of lack. The content varied wildly. The skeleton did not. Propp's argument was that the effect of these stories was produced by their structure, and that the structure was portable, recombinable and finite.
In 1946, W. K. Wimsatt and Monroe Beardsley published “The Intentional Fallacy” in the Sewanee Review, arguing that the design or intention of an author is neither available nor desirable as a standard for judging a work. The poem, they wrote, belongs to the public. It is to be evaluated on its own operation, not on what its maker meant.
Neither of these was written with reference to machines. Both describe the conditions under which machine-made narrative works.
Then there is the mechanism by which a story gets inside you. Melanie Green and Timothy Brock's 2000 paper in the Journal of Personality and Social Psychology introduced transportation, the state of absorption into a narrative involving imagery, affect and attentional focus, and demonstrated that transported readers show belief change consistent with the story. The finding relevant here is almost throwaway in the original: transportation, and the belief change that came with it, were generally unaffected by whether the story was labelled as fact or fiction.
Read that again in the context of a fruit dating show. Your capacity to be moved does not first check the ontological status of what is moving you. It was never checking. Fiction has always been a machine for producing feelings about people who do not exist. The novelty is not that the characters are fake. It is that the author is.
Dolf Zillmann and Joanne Cantor's affective disposition theory, first set out in 1977, supplies the rest. Enjoyment of narrative, on this account, runs through moral judgement: we form dispositions towards characters, we want good outcomes for those we like and bad ones for those we do not, and our pleasure tracks whether the story delivers. That process requires a character legible enough to judge. It does not require the character to have been imagined by anyone. An egotistical strawberry is a perfectly adequate object of moral judgement.
And Horton and Wohl were there in 1956, in Psychiatry, describing what they called para-social interaction: intimacy at a distance, the one-sided bond an audience forms with a media persona who cannot know they exist. They were writing about radio hosts and television compères. The asymmetry they identified was already total. There was never a reciprocal relationship to lose.
The Duanju Machine That Got There First
None of this would matter much if AI serial video were a curiosity. It is not. It is the tail end of a commercial format that has already been industrialised, and the numbers are not small.
The vertical microdrama, the duanju, is the direct precursor. One to two minute episodes, shot vertically, sold by the coin, structured so that every instalment ends on a hook. In 2024, China's micro-drama market generated around 50.4 billion yuan, roughly seven billion dollars, and in doing so overtook the country's entire cinema box office for the first time. Domestic viewership hit 662 million by the end of that year. By May 2026, according to Sixth Tone's reporting, the ultrashort-drama sector in China had reached 851 million users and a market value above 100 billion yuan, and the format had overtaken long-form video in average daily use.
The export version is equally striking. Media Partners Asia research reported in August 2026 put ReelShort's revenue at 97 million dollars in 2023, 400 million in 2024 and 785 million in 2025, on track for 1.05 billion dollars in 2026 with its first profit at scale, somewhere near 40 million dollars net after an estimated 12 million dollar loss the year before. ReelShort holds roughly 29 per cent of the market, DramaBox about 21 per cent, with the rest spread across a long tail of some three hundred apps. Deloitte's technology, media and telecommunications predictions for 2026, published in November 2025, forecast global in-app micro-series revenue rising from 3.8 billion dollars in 2025 to 7.8 billion in 2026.
This was all built by humans, filmed fast and cheap, before generative video was good enough to matter. The format was optimised for retention long before the machines arrived. What AI did was remove the last cost floor.
One Yuan Per Second
Here is the number that should stop you.
In the first quarter of 2026, roughly 128,000 ultrashort dramas were released in China. More than 95 per cent of them were AI-generated.
That figure comes from Sixth Tone's 24 August 2026 report by He Qitong, and it is corroborated in outline by CNBC's 26 August account of Chinese producers flooding the market with cheap bets and letting audiences pick the winners. The logic CNBC describes is a straightforward inversion of how entertainment has traditionally worked. Rather than committing capital to production and then buying an audience, short-drama firms produce at volume, test demand at negligible risk, and pour distribution spend only into whatever the feed has already validated.
That capability arrived quickly and from several directions at once. The generative video field in 2026 is not one tool but a stack of competing ones, and the competition is what drove the price down. Google's Veo line, OpenAI's Sora, Kuaishou's Kling, Runway, MiniMax's Hailuo and the open-weight Wan family have converged on a similar band of capability: clips measured in tens of seconds, native audio generated in the same pass as the picture, character appearance held roughly consistent from shot to shot, and output at resolutions that survive a phone screen. Vertical framing, which cinema treats as a mistake, is native to all of them. Sora's own trajectory illustrates how unstable the ground is. OpenAI notified developers on 24 March 2026 that the Sora 2 models were being removed from the API, shut the consumer app and web experience on 26 April, and set the API's final date for 24 September 2026. A company that had built the most talked-about consumer video toy in the world closed it inside a year and redirected the compute. That is worth holding on to when we get to the argument that entertainment is where the infrastructure money comes back from, because the largest player in the field has just walked away from the most obvious version of that bet.
Character consistency is the technical detail that turned clips into serials. A generator that produces a beautiful ten-second shot and then a different-looking protagonist in the next one can make adverts. A generator that holds a face across fifty episodes can make a soap opera. That threshold was crossed somewhere in the past eighteen months, and the microdrama industry was standing directly on the other side of it, already holding the format, the distribution and the audience.
The cost collapse is documented in unusual detail. ByteDance released its Seedance 2.0 video model in February 2026 at a generation cost of roughly one yuan per second of video. Sixth Tone reported a director whose typical production budget fell from about 300,000 yuan requiring seven or eight days of shooting to about 200,000 yuan over three days, with around ten roles tied to stunts and effects simply deleted. He turned down AI work despite a thirty per cent pay cut. Eight of his ten colleagues took it.
The human cost is not abstract. Wang Huan, founder of the production company Zhoutu Culture, sold three cars in July 2026, a Mercedes-Benz G63, an Aston Martin DB11 and a Bentley Continental GT, worth some 4.5 million yuan between them, as his monthly output fell from around 200 dramas to about thirty and his workforce shrank from more than 500 people to roughly 200. His question, quoted by Sixth Tone, is the one every creator in this economy eventually asks: if the platform stopped giving you money tomorrow, what would you do?
The platforms noticed the hollowing out. By May 2026 ByteDance had allocated 1.5 billion yuan and six major platforms had announced around 6 billion yuan in planned investment specifically for live-action production. That is a subsidy for humanness, which tells you something about where the unsubsidised equilibrium sits.
Note the irony. The industry that proved AI video could work at scale is now paying a premium to keep people in front of the camera.
The Cliffhanger Became A Metric
The Daily Guardian's sharpest observation is that “to be continued” has stopped being purely a narrative device and become an engagement mechanism. This is right, and it is worth being precise about why, because the imprecise version of the claim is false.
The cliffhanger has always been commercial. Serial fiction in the nineteenth century was sold by the instalment, and the instalment ended where it did in order to sell the next one. What has changed is not the existence of a commercial motive but the resolution and speed of the feedback.
A Victorian novelist learned about audience response through sales figures arriving weeks later, letters, and the talk of the town. A TikTok creator learns within hours, at the granularity of individual seconds of retention, from a system that is simultaneously the measuring instrument and the distribution channel.
That combination is the actual novelty. The platform does not merely report on what worked. It decides what gets seen, using the same signals, in the same loop, continuously. The first three seconds are not a craft convention that emerged from taste. They are a response to a ranking function. Hold attention past the threshold or the video does not circulate, and if it does not circulate it does not exist.
Under those conditions “unresolved” acquires a specific technical meaning. Resolution is a terminal state. A viewer whose question has been answered is a viewer with permission to leave. So the answer is deferred, not because deferral serves the story, but because the deferral is the unit of production. Another view, another comment, another share, another pass through the ranking system.
None of which makes the resulting stories bad. It makes them selected. There is a difference, and conflating the two is where most commentary on this subject goes wrong. Formulaic serial fiction produced under commercial constraint has given us a great deal that survived: Dickens, radio drama, soap opera, the entire structure of episodic television. Constraint is not the enemy of quality. The question is what the constraint is optimising for, and here it is optimising for a measurable interaction rather than a satisfied reader.
Dickens Sent Martin To America Because Sales Were Falling
The claim that most needs deflating is the participatory one: that the audience has become an informal writers' room, and that this is new.
It is not new. It is one of the oldest facts about serialised fiction there is.
When sales of Martin Chuzzlewit fell in 1843, Charles Dickens did not hold his nerve and trust the work. In the last chapter of the sixth monthly number, published in June 1843, he shipped his protagonist and his manservant off to America, a plot swerve introduced explicitly to arrest a commercial decline. The resulting satire outraged a great many American readers. The Old Curiosity Shop had already demonstrated the other side of the relationship, with readers in New York reportedly crowding the docks to meet the ship carrying the final instalment.
Radio and television serials formalised the practice. Soap operas ran for decades on audience feedback, recasting, resurrecting and rewriting according to what the post and later the ratings said. Fan fiction turned the audience into an unpaid parallel writers' room operating without permission. And on 12 February 2014, an anonymous Australian programmer launched Twitch Plays Pokémon, in which a crowd controlled a single game through chat commands. Alberto Aleta and Yamir Moreno, who later modelled the event formally, put participation at nearly a million players over more than two weeks. The crowd finished the game, and along the way it spontaneously generated a mythology, complete with deities improvised out of inventory items, that had nothing to do with anything Nintendo wrote. Aleta and Moreno's finding is worth noting on its own terms: the players who deviated from the collective behaviour were essential to the group succeeding at all.
Fruit Love Island ran the same playbook with the machinery made explicit. According to DesignRush's account of the series, viewers voted on which fruit should couple up via Google Forms, and the host read viewer comments aloud inside the episodes. That is a nineteenth-century feedback loop with the postal delay removed, and it is the clearest illustration available of what the Daily Guardian means by an informal writers' room.
So the participatory writers' room is not a product of AI. What AI changes is latency and cost.
Dickens could redirect a plot at monthly intervals, at the price of writing the thing. A soap opera could recast over a season. A generative creator can read the comments on episode four in the evening and publish episode five, incorporating them, before breakfast, at a marginal cost approaching the price of the compute. When the loop tightens to that degree, “responding to the audience” and “being written by the metric” become difficult to tell apart, because the audience's expressed preferences reach the creator pre-filtered by the ranking system that decided which reactions were visible in the first place.
That is the genuinely new thing, and it deserves a better description than participation. The audience is not in the writers' room. The audience is in the training signal.
Entertainment As The Return On Seven Hundred Billion
Why does any of this warrant serious attention rather than amused dismissal? Because of where the money is.
In January 2026, Cody Kommers and Ari Holtzman posted a paper to arXiv titled “AI as Entertainment”. Their argument runs against the industry's own marketing. AI is sold as an intelligence technology, a productivity instrument, a tool for augmenting human capability. Kommers and Holtzman contend that entertainment is an emerging and badly underexamined use case, that AI is already widely adopted for entertainment purposes and especially by young people, and that entertainment will become a major revenue driver for the largest AI companies as they seek returns on enormous infrastructure investment.
They also argue that the field's evaluation frameworks are lopsided, built to measure harms while ignoring cultural benefit, and they propose what they call thick entertainment as an alternative: an approach that asks how AI-generated content contributes to meaning-making, identity formation and social bonding rather than merely how much damage it avoids. Their governing analogy is social media, which was sold as connection technology and became an attention business.
The infrastructure numbers give the thesis its force. Alphabet, Amazon, Meta and Microsoft are collectively expected to spend somewhere around 725 billion dollars in capital expenditure in 2026, up sharply from roughly 410 billion the year before, with the great majority directed at AI infrastructure. Add Oracle and the committed figure across the five largest providers sits in the region of 660 to 690 billion dollars. Analysts are already discussing a trillion-dollar year in 2027.
Capital on that scale requires consumer revenue at a scale that enterprise software subscriptions are unlikely to supply on their own. Entertainment is the only consumer category with a history of absorbing that much attention and converting it into money. Kommers and Holtzman are making a prediction, not reporting a fact, and it should be read as such. But the microdrama numbers suggest the prediction is being tested in the market right now, and the early returns are not ambiguous.
There is an honest caveat to enter here. Whether AI serial video specifically will produce durable revenue remains unproven. Fruit Love Island had 300 million views and its creator said it earned nothing. The most valuable AI-native entertainment company in the world may turn out to be a microdrama app that uses generative tools to cut production costs rather than a studio of autonomous machine storytellers. Attention has been demonstrated. Monetisation has not.
Labels Are Not Enough And Everybody Knows It
The regulatory response has arrived, and it is aimed at exactly the wrong thing.
Article 50 of the EU AI Act, whose transparency obligations came into application on 2 August 2026, requires that AI-generated or manipulated content be marked in machine-readable form and that deepfakes be disclosed. The Commission adopted guidelines on 20 July 2026 and has been developing an accompanying code of practice. The deepfake rules apply regardless of intent to deceive, and even where no real person is depicted. Content generated and published before 2 August 2026 does not have to be retroactively labelled. Non-compliance can attract fines up to 15 million euros or three per cent of worldwide annual turnover.
Platforms have moved in parallel. TikTok says it has now labelled more than three billion videos as AI-generated, using a combination of C2PA Content Credentials, creator self-labelling and invisible watermarking. It was the first video platform to implement Content Credentials and has joined the C2PA steering committee. In the first quarter of 2026 alone it removed over 86 million fake accounts. YouTube renamed its repetitious content policy to inauthentic content in July 2025, targeting mass-produced and templated material while explicitly permitting AI tools where the finished video carries genuine human-added value.
Every one of these interventions addresses provenance. None addresses structure.
That is the gap the Porter and Machery result opens up. If disclosure reduces appreciation after the fact but does not prevent transportation during it, then a label is a receipt, not a shield. It tells you what you consumed. It does not alter the consuming. Green and Brock's finding that transportation is largely indifferent to fact-versus-fiction labelling suggests the same thing from the other direction. The apparatus we are building is designed to inform a deliberative faculty that was never in charge of the response in the first place.
There is a second problem, which is that labelling regimes and monetisation regimes are pulling against each other in ways that are genuinely confused. Reporting on TikTok's monetisation rules for AI content in 2026 is inconsistent, with different programmes described as banning AI content outright, permitting it if labelled, or permitting it subject to originality requirements. I could not establish a single authoritative answer, and I am not going to pretend otherwise. What is documented is the outcome in the one high-profile case: the biggest AI serial on the platform had twelve of its twenty-two episodes removed and its creator quit citing an absence of revenue.
If the effect of labelling is that compliant AI content is algorithmically suppressed and demonetised, the rational response is not to stop making it. It is to stop labelling it. Transparency rules that impose a distribution penalty create an incentive to evade them, and enforcement against millions of accounts producing hundreds of thousands of items a quarter is not a solved problem.
The Word For Content Nobody Asked For
Meanwhile the culture has issued its verdict, and it is contemptuous.
Merriam-Webster made “slop” its word of the year for 2025, defining it as digital content of low quality produced usually in quantity by artificial intelligence. Macquarie Dictionary in Australia had already chosen “AI slop”, defining it as low-quality content created by generative AI, often containing errors, and not requested by the user. Fruit Love Island was described by critics, more or less universally, as a perfect example of the genre.
The slop framing is useful and also self-serving. Useful, because volume is a real harm: 128,000 dramas a quarter is not a cultural flourishing, it is a denial-of-service attack on discovery, and it does displace people who cannot compete on cost. The Fruit Love Island case carries that charge too. The creator Joy Ofodu, who had been producing original sketches voicing inanimate objects including fruit characters since 2020 at a rate of two or three hundred a year, said publicly that the AI series appeared to have been inspired by her work without credit. Whatever the provenance, that is the structural complaint: the format is cheap to copy and the copy scales faster than the original.
Self-serving, because “slop” lets the critic locate the deficiency in the object rather than in the response. It says the problem is that this stuff is bad. The evidence says the problem is that it works. Three million followers in nine days is not a story about people being fooled. It is a story about narrative structure doing what narrative structure does, in the absence of anyone having meant it.
And the aesthetic objection has an awkward history. Every industrialised narrative form was called slop by someone. Penny dreadfuls, dime novels, pulp magazines, radio serials, soap opera, and the word “soap opera” itself was not a compliment. Some of that material was rubbish and some of it turned out to be the popular art of its century. Contempt is not a prediction.
What Is Actually Lost
So return to the viewer at two in the morning, arguing about characters no one wrote.
The honest answer to what it means that stories which move us can be made by something that feels nothing is: less than we would like, and not nothing.
Less than we would like, because the emotional response was never underwritten by the author's sincerity. Propp showed the effects were structural. Wimsatt and Beardsley argued the intention was neither available nor desirable as a standard. Green and Brock showed transportation runs regardless of whether the thing is true. Horton and Wohl showed the intimacy was one-sided from the start. Porter and Machery showed we cannot tell, and that knowing changes our rating rather than our reaction. A person who cries at an AI-generated episode has not made an error. They have discovered something about how narrative always worked, and it is not flattering.
Not nothing, because two things are genuinely at stake, and neither is authenticity.
The first is what the story is for. A human serial optimises for a reader who returns tomorrow. An algorithmic serial optimises for a measurable interaction now. Those objectives overlap substantially, which is why the output is often watchable, but they diverge precisely at the point of resolution. A story written for a person eventually ends, because endings are what make the middle mean anything. A story written for a retention curve has no reason to end, because the ending is the moment the metric goes to zero. What is threatened is not emotional truth. It is closure, and closure is where narrative meaning actually lives.
The second is the disappearance of a counterparty. Serial fiction has always been a negotiation, and the audience has always had a hand in it, from Dickens rerouting Martin Chuzzlewit to save his sales to a Twitch chat improvising a religion. But there was somebody on the other side, with intentions of their own, who could refuse. Dickens sent Martin to America and then wrote the America he wanted to write, at the cost of enraging half a continent. The audience pushed, and something pushed back.
Optimisation does not push back. It converges. Feed a system your reactions and it will give you more of what produced them, and the conversation the Daily Guardian describes between creator, audience, platform and algorithm has, at its far end, no participant capable of saying no. That is the loss, and it is a structural one rather than a spiritual one: not that the machine cannot feel, but that it cannot disagree.
Which suggests the interesting question is not whether we should be moved by machine-made stories. We already are, we always could have been, and the label arrives too late to stop it. The question is whether anything in the loop is still capable of wanting the story to go somewhere other than where the numbers point.
At the moment, the only candidate is the person watching at two in the morning. That is a thinner safeguard than it sounds, and it is the only one on offer.
Sources and References
- Kommers, Cody and Holtzman, Ari (2026) “AI as Entertainment,” arXiv, arXiv:2601.08768, 13 January 2026. Available at: https://arxiv.org/abs/2601.08768
- Cripps, Coral (2026) “TikTok Series 'Fruit Love Island' Reveals AI Risk After 300M Views,” DesignRush News, 8 April 2026. Available at: https://news.designrush.com/tiktok-series-fruit-love-island-ai-format-risk
- Fast Company (2026) “'Fruit Love Island' is TikTok's most popular AI-generated series. Now it's facing trouble in paradise,” Fast Company, March 2026. Available at: https://www.fastcompany.com/91519147/fruit-love-island-tiktok-most-popular-ai-generated-series-now-facing-trouble-in-paradise
- Murray, Conor (2026) “Bizarre AI-Generated Fruit 'Love Island' TikTok Videos Drive Massive Engagement,” Forbes, 23 March 2026. Available at: https://www.forbes.com/sites/conormurray/2026/03/23/bizarre-ai-generated-fruit-love-island-tiktok-videos-drive-massive-engagement/
- The Wrap / Yahoo Entertainment (2026) “'Bullying works': AI series 'Fruit Love Island' creator rage quits after mass video takedowns.” Available at: https://www.yahoo.com/entertainment/tv/articles/bullying-works-ai-series-fruit-200000743.html
- “Black Creator Says AI 'Fruit Love Island' Series Was Likely Inspired by Her Content” (2026). Available at: https://www.aol.com/articles/black-creator-says-ai-fruit-164743347.html
- He, Qitong (2026) “China's Short-Drama Boom Was Built on Speed. Then AI Got Faster,” Sixth Tone, 24 August 2026. Available at: https://www.sixthtone.com/news/1018902
- CNBC (2026) “China's short-drama producers flood the market with cheap bets, and let audiences pick the winners,” CNBC, 26 August 2026. Available at: https://www.cnbc.com/2026/08/26/short-drama-china-production-ai-entertainment-economics.html
- Variety (2026) “ReelShort on Track for $1.05 Billion Revenue, First Profit at Scale in 2026, Media Partners Asia Report Finds,” Variety, August 2026. Available at: https://variety.com/2026/tv/news/reelshort-1-billion-revenue-profit-2026-mpa-1236828885/ (see also Deadline, “ReelShort On Course To Reach $1BN Revenue & First Profit In 2026, MPA Report,” August 2026: https://deadline.com/2026/08/microdrama-reelshort-revenue-profit-1237027492/)
- OpenAI (2026) “What to know about the Sora discontinuation,” OpenAI Help Center, and “Deprecations,” OpenAI API documentation. Available at: https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation and https://developers.openai.com/api/docs/deprecations
- Arkenberg, Chris, Dhameja, Ankit, Bottke, Tim and Crossan, Gillian (2025) “Tiny episodes, massive appeal: Short-form serials are gaining viewers and empowering independent studios,” Deloitte Insights, TMT Predictions 2026, 18 November 2025. Available at: https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2026/short-form-video-series.html
- CNBC (2025) “How China's $7 billion micro drama industry is taking on the U.S. entertainment industry,” CNBC, 22 July 2025. Available at: https://www.cnbc.com/2025/07/22/why-chinas-7b-micro-drama-industry-is-taking-over-social-feeds.html
- Porter, Brian and Machery, Edouard (2024) “AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably,” Scientific Reports, vol. 14, art. 26133, November 2024. Available at: https://www.nature.com/articles/s41598-024-76900-1
- “Understanding Reader Perception Shifts upon Disclosure of AI Authorship” (2025), arXiv, arXiv:2510.24011, and Proceedings of the 31st International Conference on Intelligent User Interfaces. Available at: https://arxiv.org/abs/2510.24011
- Green, Melanie C. and Brock, Timothy C. (2000) “The role of transportation in the persuasiveness of public narratives,” Journal of Personality and Social Psychology, 79(5), pp. 701-721. Available at: https://pubmed.ncbi.nlm.nih.gov/11079236/
- Horton, Donald and Wohl, R. Richard (1956) “Mass Communication and Para-Social Interaction: Observations on Intimacy at a Distance,” Psychiatry, 19(3), pp. 215-229. Available at: https://www.tandfonline.com/doi/abs/10.1080/00332747.1956.11023049
- Wimsatt, W. K. and Beardsley, Monroe C. (1946) “The Intentional Fallacy,” The Sewanee Review, 54(3), pp. 468-488. Available at: https://www.sas.upenn.edu/~cavitch/pdf-library/WimsattBeardsley_Intentional.pdf
- Propp, Vladimir (1968) Morphology of the Folktale, 2nd edn, trans. Laurence Scott, University of Texas Press (originally published in Russian, 1928). Overview available at: https://rupkatha.com/V9/n2/v9n241.pdf
- Zillmann, Dolf and Cantor, Joanne R. (1977) “Affective responses to the emotions of a protagonist,” Journal of Experimental Social Psychology, 13(2), pp. 155-165. Overview of subsequent affective disposition research available at: https://www.sciencedirect.com/science/article/pii/S0001691826003070
- European Commission (2026) “Transparency obligations under Article 50 of the AI Act,” Shaping Europe's Digital Future. Available at: https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act
- Greenberg Traurig LLP (2026) “Deepfakes, Chatbots, AI-Generated Text: European Commission Details Transparency Obligations Under the AI Act,” June 2026. Available at: https://www.gtlaw.com/en/insights/2026/6/deepfakes-chatbots-ai-generated-text-european-commission-details-transparency-obligations-under-the-ai-act
- TikTok Newsroom (2026) “Helping people spot and understand AIGC on TikTok,” 10 July 2026. Available at: https://newsroom.tiktok.com/helping-people-spot-and-understand-aigc-on-tiktok
- Social Media Today (2025) “YouTube Clarifies Changes to Monetization Rules Around Inauthentic Content,” July 2025. Available at: https://www.socialmediatoday.com/news/youtube-clarifies-monetization-update-inauthentic-repeated-content/752892/
- Merriam-Webster (2025) “Word of the Year 2025: Slop,” Merriam-Webster, December 2025. Available at: https://www.merriam-webster.com/wordplay/word-of-the-year (see also Macquarie Dictionary, “Macquarie Dictionary Word of the Year for 2025,” 24 November 2025: https://www.macquariedictionary.com.au/macquarie-dictionary-word-of-the-year-for-2025/)
- Aleta, Alberto and Moreno, Yamir (2018) “Collective social behavior in a crowd controlled game,” arXiv, arXiv:1811.09730, 24 November 2018. Available at: https://arxiv.org/abs/1811.09730
Tim Green UK-based Systems Theorist & Independent Technology Writer
Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.
His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.
ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk
Listen to the free weekly SmarterArticles Podcast
