This month’s internet obsession was Willy’s Chocolate Experience,[1] a thinly-veiled rip-off of the ‘Wonkaverse’, the loosely-connected cinematic universe based on Roald Dahl’s 1964 novel Charlie and the Chocolate Factory. The Glasgow event’s marketing bore all the hallmarks of imagery produced by artificial intelligence — colours with a bright unreality; shapes rounded, flowing like liquid. Savvy parents and potential customers might have recognised these as prompt-generated images, but could hardly have been blamed for their shock when confronted with an event so catastrophically underwhelming that it launched a thousand memes, and ended up in the headlines of every paper from British tabloid The Sun to the New York Times.
One parent, quoted in The Independent, described the almost-empty warehouse standing in for Willy’s magical factory as “…[a] disorganised mini-maze of randomly placed oversized props, a lacklustre candy station that dispersed one jelly bean per child, and a terrifying chrome-masked character that scared many of the kids to tears.”[2] Gizmodo later sourced the script for the event,[3] also generated by artificial intelligence. Paul Connell, one of the actors playing ‘Willy Macduff’ told The Independent the script was “… gibberish… In some ways, it was a world of imagination. Like, imagine that there is a whole chocolate factory here.”
The photographs of the dismal factory space (which have now been memed around the world) speak for themselves.[4] The glossy, hallucinatory images or nonsequiteur-laden “dialogue” of an AI creation can seem to offer the possibility of limitless creation, but are woefully inefficient once translated into the real world. Willy’s “Experience” carried unintentional echoes of Banksy’s ‘Dismaland’ project, with its broken down ice-cream vans and haunted fairground attractions, a critique of hypercapitalism and David Cameron’s ‘Broken Britain’. This was less of a joke, more of a con.
Jean Baudrillard coined the term ‘hyperreality’ in 1981, in Simulacra and Simulation. One of the ‘realities’ he analysed was that of Disneyland, the original “world of pure imagination”. He wrote:
Disneyland is presented as imaginary in order to make us believe that the rest is real, whereas all of Los Angeles and the America that surrounds it are no longer real, but belong to the hyperreal order and to the order of simulation.[5]
Baudrillard’s cities are ‘hyperreal’ because they are composed not of things, but copies of things, and indeed people, that are indistinguishable from, if not more real than the original. These ‘simulacra’ are uncanny — not quite human. Naomi Klein’s Doppelganger touched on literary examples from Robert Louis Stevenson’s Jekyll and Hyde to Dostoevsky’s The Double. Philip K. Dick’s science fiction novels return again and again to the figure of the doppelganger, or robotic automata. In each of these literary examples, the experience of meeting the doubled self is monstrous; a perfect embodiment of Freud’s unheimlich.
When we use the weird doppelganger of our artificial intelligences to create the imaginary spaces where we go to be entertained, we do not find ourselves in Baudrillard’s conception of a fantasy that structures hyperreality. Instead we are in a liminal space, haunted by strange and unfamiliar ghosts. As Paul Connell told The Independent, describing the menacing, metal-masked villain of Willy’s Factory (an “evil chocolate maker who lives in the walls” called The Unknown): “It was terrifying for the kids. Is he an evil man who makes chocolate or is the chocolate itself evil?”
Connell is poking fun here at the weird ambiguities of the AI-generated script, but he raises a valid point. AI-generated art fails the reality test because of these kinds of nonsequiturs. ChatGPT and other large language models are prone to ‘hallucinations.’ If there is a gap in what they generate, they simulate something to fill the ellipsis. This is a particular problem if generating scientific papers or news reportage. Michael J. Black, a director of the Max Planck Institute for Intelligent Systems, wrote of his experience using Meta’s Galactica AI model: “In all cases, it was wrong or biassed but sounded right and authoritative.”[6]
This is a problem that one of the creators of Willy’s Chocolate Experience is prepared to weaponise. His substandard attraction went so viral that it was even covered by Rolling Stone, who found out this was not the first time Billy Coull had used AI to make a quick buck. He has masqueraded as a life coach and business guru, but more importantly a publisher: “The 16 books on Coull’s Amazon author page were all published in the summer of 2023 — some of them on the very same day. The synopsis for each is AI-generated.”[7] Rolling Stone’s reporter read some of these books, and found themes relating to anti-vaccine conspiracy theories, Q-Anon narratives and other disinformation.
Coull’s ability to publish that many books in a single day calls to mind the software development term ‘minimum viable product’ — to ship a product in its earliest usable state, with the intent to improve it in later iterations. In reality, MVP is often all that is delivered. The internet is now awash with advice for young authors on how to write the ‘minimum viable book’ for online markets. Many are turning to AI tools like Sudowrite to help them meet demand. Josh Dzieza, writing for The Verge, describes how Sudowrite works:
Authors paste what they’ve written into a soothing sunset-colored interface, select some words, and have the AI rewrite them in an ominous tone, or with more inner conflict, or propose a plot twist, or generate descriptions in every sense plus metaphor.[8]
These tools may solve a problem and fill a market need, but they also risk producing hallucinations, inaccuracies and uncanny effects. Willy Macduff’s enemy The Unknown is frightening precisely because it was generated by the unconscious mind of an AI; because it is a vague outline, a dark shape glimpsed in a dream. It has this in common with the self-replicating ‘supercomposite’ AI creature known as Loab,[9] who went viral in 2022. Whether Loab was a genuine horror from the depths of the machine unconscious or a clever art project that manipulated image prompts, it is undeniable that an encounter with her was horrifying.
Unlike Loab’s creator Steph Maj Swanson, Billy Coull and his colleagues at House of Illuminati did not set out to create something unheimlich. Besides the dubious scares of The Unknown’s evil chocolate (or chocolatier?), their AI creation manifested nothing but disappointment. Machine intelligence augmented the promo for the event, but the actual entertainment fell to hired actors like Connell, most of them students or gig economy workers. They did their best to entertain the kids whose parents paid a ridiculous sum to attend, but even this was stymied by the greed of Coull and his cronies. Like the worst AI-assisted books, the finished product was cheap and shoddy, much like the promoters themselves.
Coull is hardly the first person to use AI to create sub-optimal content. As Ben Child writes for The Guardian: “It is already possible to find material in the darker corners of Netflix or Amazon Prime that feels like it was churned out by lifeless, soulless entities using repetitive formulas and algorithms.”[10] Other journalists have experimented with AI-generated scripts in order to unpack the phenomena.[11] While this remains a topic for viral jokes and clickbait articles, the Writers Guild of America and the Screen Actors Guild were concerned enough about AI to shut down the film and television industries for several weeks with strikes in 2023.
Google’s attempts to counteract racist bias in its Gemini image-generation tool ended up producing ridiculous caricatures of black Vikings and Nazis. As Google’s Prabkahar Raghavan explained:
… we tuned it to ensure it doesn’t fall into some of the traps we’ve seen in the past with image generation technology — such as creating violent or sexually explicit images, or depictions of real people. And because our users come from all over the world, we want it to work well for everyone.[12]
These weird, unanticipated effects are the consequence of asking a machine to produce art from a synthesis of existing images or texts, but within certain parameters. Images of black people, Vikings and Nazis are just data to Gemini. With the statistical category of ‘equal representation’ applied to them, and no category for ‘appropriate’ or ‘accurate’, the synthetic images they produce are not ‘mistakes’ but instead the direct result of the ‘tuning’ Raghavan describes.
Duane Rousselle wrote for Sublation about the AI-generated animated show Nothing, Forever, a parody of Seinfeld that went viral after the script auto-generated a transphobic joke. Just like Gemini, the algorithmic prompts used to generate the show’s scripts were tweaked to avoid causing offence:
… the program was permitted to air again after 14 days, continuing along its old track with a modest adjustment…Thus the scientific solution is always to implement another false twist, often in the form of moral censorship within the ChatGPT code.[13]
Only the elimination of all transphobic language from the model itself would prevent a ChatGPT script from producing transphobic jokes. The AI cannot be made to behave in a way that is moral or otherwise humane; it can only be censored, and in that way tailored to a modern audience’s needs.
Another industry that mines ‘new’ content from established imagery and audio is Hollywood, and particularly Disney itself. Disney is the original master of the accumulation, manipulation and exploitation of intellectual property. Owners of the world’s biggest franchises, they sit on an IP mountain of stories from the world’s commons, such as the fairy tales of the Brothers Grimm. Anyone attempting to produce a new story from the source material must be cautious not to infringe Disney trademarks from their beloved ‘classic’ versions. When most of us think of Cinderella or Sleeping Beauty, it is Disney imagery we see in our heads.
Much of Disney’s material since the 1990s has taken the form of sequels, live-action remakes, reboots and prequels of established tentpoles. Another recent smash-hit was the film that inspired Coull and his gang of hucksters to ‘create’ Willy Macdfuff. Wonka was a prequel to the beloved 1971 film, or perhaps its 2003 reboot (perhaps both). In place of the mercurial, psychedelic shenanigans of the Gene Wilder-led original and the hyper-gothic surrealism of Tim Burton’s Johnny Depp-starring 2003 misfire, Wonka plays like a low-rent East End jukebox musical. The songs have platitudinous, positivistic lyrics and anodyne, technically flawless vocals smoothed out by machine effects. The set design is perfunctory, a strange mish-mash of the forced perspective of a theatre set with the unreal scale of the Disney-patented augmented reality special effects platform, the Volume. Like any technology, the Volume and other digital FX trickery can be used creatively, or as a shortcut. Wonka’s half-hearted VFX produces an ‘uncanny valley’ effect that flattens rather than heightens the film’s reality.
Flatter still is Wonka himself. Timothee Chalamet plays him with wide-eyed innocence, but a strange lack of affect; otherworldly, but unthreatening and sexless. It’s a sharp contrast to Johnny Depp’s (admittedly very strange) choice to play Wonka as a repressed eccentric with visual echoes of Michael Jackson. A sharper contrast still with the brooding, troubled mania of Gene Wilder’s unforgettable performance, shot through as it was with a rich vein of darkness and latent violence. Chalamet’s Wonka, like the film he inhabits, is strangely neutered. Perhaps this is a consequence of tailoring the film for modern audiences, tastes and morality, a fate that has lately befallen other adaptations of Roald Dahl material — and the author’s original novels. The result is colourful but figuratively empty, with bowdlerised dialogue and forgettable songs that recall the bland approximations of an AI-generated script. Wonka is designed to appeal to so broad a majority of tastes that it ends up a bland disappointment, much like its Glaswegian imitator.
In the future, it might be possible for a huckster like Coull and his gang to use technologies like the Volume to create an augmented physical location for a new rip-off Wonkaverse. It is likely that some version of Disney’s Volume technology can be scaled down and made affordable. We laugh at Wonkagate because the con is unsophisticated. Soon, the script, the stage and the characters themselves could all be generated simply and relatively cheaply. This problem is even more pressing when we consider the weaponization of AI text, image and video generation in the game of politics. Deepfake video and audio is likely to increase in sophistication year on year. The upcoming US elections have already seen a form of this, with ‘Robocalls’ placed to voters using a fake recording of Joe Biden.[14] In this instance, the Biden impression was a performance created by (ironically) a New Orleans magician, Paul David Carpenter. In the very near future, that performance could be as easy to deepfake as the calls were to automate.
The recent SAG and WGA strikes ended successfully, with concessions for writers and actors, but the slow, inevitable march of artificial intelligence on the creative industries seems unlikely to be significantly slowed. No matter how fast technology accelerates, art must accelerate faster. If technology copies our abilities, we must hope that our own creations continue to channel the weird alchemy of a Gene Wilder performance, and not the easily-simulated blankness of a Timothee Chalamet clone. As Baudrillard wrote, this is a horizon from which we might struggle to return:
“And so art is everywhere, since artifice is at the very heart of reality. And so art is dead, not only because its critical transcendence is gone, but because reality itself, entirely impregnated by an aesthetic which is inseparable from its own structure, has been confused with its own image. Reality no longer has the time to take on the appearance of reality. It no longer even surpasses fiction: it captures every dream even before it takes on the appearance of a dream.”[15]
Like Baudrillard’s Disneyland, Willy Macduff’s Chocolate Experience exists to conceal a reality itself already populated by simulacra. Just like Disney’s later cultural products, its facsimile of reality is shoddy enough to suggest the fakeness that saturates our culture. Baudrillard argued: “What every society looks for in continuing to produce, and to overproduce, is to restore the real that escapes it.”[16] We are very far from attempting any project to restore the real, but we can think very carefully about how we produce culture — or its simulacra. If we fail, perhaps our future artistic output will at best resemble Giacomo Miceli’s chatbot ‘The Infinite Conversation’,[17] which simulates a never-ending dialogue between the philosopher Slavoj Žižek and the filmmaker Werner Herzog. The simulation of both men’s cadences, interests and peculiar predilections is eerily accurate. And yet, what is actually generated is a torrent of meaninglessness. It might last forever, but it means nothing.
[1] https://willyschocolateexperience.com
[2] Barney Davis, “Angry Oompa Loompas and no chocolate: Wonka actor breaks silence on Willy’s Chocolate Experience chaos.” Independent, Mar. 1, 2024, https://www.independent.co.uk/news/uk/home-news/willy-wonka-experience-glasgow-oompa-loompa-b2505176.html
[3] Thomas Germain, “Here’s the Full AI-Generated Script From the Willy Wonka Disaster”, Gizmodo, Feb. 29, 2024, https://gizmodo.com/heres-the-full-ai-generated-script-from-the-willy-wonka-1851295448
[4] Photos of the event that went viral on X, shared by user @BeardedGenius: https://twitter.com/BeardedGenius/status/1762444050093760761
[5] Jean Baudrillard, trans. Sheila Faria Glaser, Simulacra and Simulation (University of Michigan Press, 1994), p12.
[6] https://twitter.com/Michael_J_Black/status/1593133722316189696
[7] Miles Klee, “Huckster Behind ‘Willy Wonka’ Event Also Sells AI-Written Vaccine Conspiracy Books.” Rolling Stone, Feb. 27, 2024, https://www.rollingstone.com/culture/culture-features/willy-wonka-event-glasgow-billy-coull-ai-vaccine-conspiracy-books-1234976876/
[8] Josh Dzieza, “The Great Fiction of AI.” The Verge, Jul. 20, 2002, https://www.theverge.com/c/23194235/ai-fiction-writing-amazon-kindle-sudowrite-jasper
[9] Dorian Batycka, “Meet ‘Loab,’ the Latest Example of A.I.-Generated Art Creeping Out the Internet (Sorry in Advance for the Nightmares).” Artnet, Sep. 12, 2022, https://news.artnet.com/art-world/meet-loab-the-latest-example-of-a-i-generated-art-creeping-out-the-internet-sorry-in-advance-for-the-nightmares-2173737
[10] Ben Childs, “AI is coming for Hollywood scriptwriters – this is how they are going to do it.” The Guardian, May 12, 2023, https://www.theguardian.com/film/2023/may/12/ai-artificial-intelligence-generating-screenplays
[11] Critics and journalists experiment with AI scripts:
https://www.theguardian.com/film/2023/mar/24/chapgpt-movie-script-ai
https://www.latimes.com/projects/writers-strike-deal-ai/
https://www.nytimes.com/2023/05/27/opinion/ai-screenplays-scripts.html
[12] Prabkahar Raghavan, Gemini image generation got it wrong. We’ll do better.” The Keyword, Feb. 23, 2024, https://blog.google/products/gemini/gemini-image-generation-issue/
[13] Duane Roussel, “Escaping the Metaverse.” Sublation Magazine, Mar. 8, 2023, https://www.sublationmag.com/post/escaping-the-meta-verse
[14] Ramon Antonio Vargas, “New Orleans magician says he made AI Biden robocall for aide to challenger.” The Guardian, Feb. 23, 2024, https://www.theguardian.com/us-news/2024/feb/23/ai-biden-robocall-magician-new-orleans
[15] Jean Baudrillard, Simulations (Semiotext[e], 1983), p151-152 (PDF accessed at https://archive.org/details/Simulations1983/page/n1/mode/2up).
[16] Baudrillard, Simulacra and Simulation, p23.