Art and entertainment in an AI-shaped world
In 2023, I watched an AI-generated trailer for a film adaptation of Heidi. Perhaps a third of it could have passed, at least momentarily, for conventional cinema. The rest was a procession of beautiful errors: unstable faces, impossible movements and images that carried the surface of a film without quite understanding how a film behaves.
It was ridiculous, impressive and faintly unnerving.
At the time, I began arranging the future into decades. Design would be transformed first, followed by art, books, film, television, games, music and eventually technology itself. Each medium seemed to be waiting for its appointed encounter with artificial intelligence.
I no longer think the future will arrive so politely.
Film, television, games, music, books, art, design and software are not separate destinations along a technological timeline. They are increasingly composed from the same underlying materials: language, images, sound, code and systems of rules. Once machines became capable of working across those materials, every medium began changing at once.
The relevant question is no longer when AI will “reach” a particular art form. It is what happens when the cost of producing plausible cultural material falls towards zero.
The Heidi trailer was compelling because it revealed both sides of the transition. It could produce the suggestion of cinema without the discipline traditionally required to make cinema. It knew what an Alpine landscape, period costume and sentimental close-up were supposed to resemble. What it lacked was continuity, physical judgement and a durable understanding of character. It offered fragments of recognisable intention without a mind responsible for the whole.
Since then, the fragments have become more coherent. Contemporary video systems can use reference images to maintain characters and visual styles, extend scenes and generate dialogue, ambient sound and music alongside the image. These capabilities remain imperfect, but they have advanced far enough that the old defects can no longer be treated as permanent limitations. Google DeepMind’s current Veo materials provide one indication of how quickly synthetic video is developing.
The immediate future of entertainment will not necessarily be a button that produces a flawless feature film. It will be the gradual decomposition of production into hundreds of tasks that machines can perform, accelerate or cheaply repeat. Storyboarding, concept art, background design, temporary dialogue, visual effects, translation, editing, music sketches and promotional material can all be generated or revised before a final work reaches an audience.
This will give individual creators powers once reserved for institutions. A filmmaker without a studio may be able to visualise an entire world. A musician may orchestrate an idea without hiring an orchestra. A game designer may populate a landscape with characters who remember what players have done. A novelist may adapt a story into several languages, formats and visual interpretations without surrendering the project to a large production company. (A solo writer may create a lifelong work of his own — Ed)
That possibility is genuinely exciting. It may also produce a culture of overwhelming abundance.
When almost anyone can generate an acceptable image, song, story or sequence of film, the ability to make something will no longer distinguish it. The scarce resource will be attention. More specifically, it will be the ability to decide what deserves attention.
This changes the creative problem. For much of history, artists struggled against the difficulty of execution. Materials were expensive, skills took years to acquire and distribution was controlled by institutions. AI weakens some of those barriers, but it does not eliminate difficulty. It moves difficulty elsewhere: into conception, selection, coherence and judgement.
A system may generate a thousand images in an afternoon. The significant act becomes recognising the one image worth keeping—and understanding why. It may propose fifty melodies, twenty endings or ten versions of a scene. Someone must still determine which possibility belongs to the work and which merely resembles competence.
This is why the familiar reassurance that “AI will never replace human creativity” is too simple. Machines do not need to reproduce the whole mystery of human creativity to disrupt creative employment. They need only perform enough valuable tasks, cheaply and reliably enough, for organisations to employ fewer people. The effect will not be distributed evenly. Some artists will gain extraordinary leverage; others will discover that the work through which they learned their craft has been automated or devalued.
The language of “partnership” can obscure this conflict. A partnership presumes that both parties possess some power over its terms. Many workers will instead encounter AI as a decision already made by an employer: a tool used to increase output, reduce budgets or replace junior positions. The consequences will depend as much on contracts, unions, law and ownership as on the intelligence of the models themselves.
The entry-level problem may prove especially serious. Creative industries have traditionally trained people through imperfect, repetitive and subordinate work. Assistants prepare drafts. Junior designers produce variations. New writers cover routine assignments. Young developers repair minor problems before being trusted with consequential systems. If machines absorb those tasks, industries may become more productive while quietly destroying the paths through which expertise is formed.
Authorship will become similarly difficult to locate. A future film might begin with a human premise, use synthetic actors, contain machine-generated environments, be edited through natural-language instructions and adapt itself to different viewers. Calling it either “human-made” or “AI-made” would conceal more than it revealed.
Law has begun confronting this ambiguity. In the United States, the Copyright Office has concluded that AI-assisted work may still receive protection where a person contributes sufficient expressive authorship through selection, arrangement or modification, but that prompting alone does not automatically make someone the author of everything a system produces. The distinction places human judgement, rather than mere initiation, near the centre of authorship. Its 2025 report is an early attempt to draw boundaries that technology will continue to test.
Audiences may develop their own boundaries. As synthetic material becomes ordinary, provenance could become part of a work’s meaning. People may want to know who made something, whose experiences shaped it, whether the voices and faces were used with consent, and whether anyone accepts responsibility for what it says.
This does not mean audiences will reject artificial entertainment. Most will happily consume it when it is pleasurable, convenient or free. But a parallel appetite for authenticated human experience is likely to grow. A live performance matters partly because it occurs between particular people in an irretrievable moment. A handmade object carries traces of effort that cannot be separated from its value. A novel can matter because another consciousness spent years finding the language for something the reader had felt but never expressed.
In a world of infinite competent material, evidence of finite human commitment may become more valuable.
Personalisation will create another fracture. AI could generate entertainment around an individual’s preferences: a detective series calibrated to their favourite setting, a game that changes according to their temperament, or music composed for the precise pace and emotional texture of their day. Stories may cease to be fixed objects and become services, continuously revised in response to the audience.
There is obvious pleasure in this. There is also a cultural cost. Shared works give strangers something in common. Millions of people can argue about the same ending, remember the same song or recognise the same line. If entertainment becomes perfectly personalised, culture may lose some of its ability to create a public.
Art has never existed solely to satisfy preference. It also confronts us with preferences we did not know we had. It asks for patience, introduces unfamiliar forms and sometimes refuses to please us. A system trained to maximise engagement may become extraordinarily good at giving people more of what they already enjoy. Whether it can reliably lead them towards what they do not yet understand is a different question.
The same tension applies to technology itself. AI will increasingly assist in writing software, testing hypotheses, modelling systems and designing new tools. But the fantasy of a machine that can build everything in the “fastest, highest-quality and overall best way” contains a hidden problem: there is no neutral definition of best.
Best for whom? At what environmental cost? According to which values? A transport system optimised for speed may damage neighbourhoods. A platform optimised for engagement may corrode attention. A workplace optimised for output may become intolerable to inhabit. Intelligence can improve the pursuit of an objective, but it cannot make the objective politically or morally innocent.
The most important human role may therefore be neither making every component nor issuing clever prompts. It may be assuming responsibility for direction. Someone must decide what should exist, whose interests it serves, what compromises it contains and what consequences are unacceptable. These are not residual tasks left behind because machines have failed to master them. They are the central questions from which every meaningful creative and technological decision begins.
Preparing for an AI-shaped future means more than teaching people how to operate the latest tools. It means protecting the conditions under which they can refuse them. It means establishing consent and compensation for the use of creative work, preserving routes into skilled professions, requiring transparency where synthetic media can deceive, and ensuring that cultural life is not entirely governed by whichever systems capture attention most efficiently.
It also means becoming more demanding audiences. We will need to distinguish polish from insight, novelty from importance and emotional simulation from emotional truth. The ease with which something can be generated should neither disqualify it nor excuse it from judgement.
In 2023, that strange Heidi trailer made the future appear distant enough to organise into decades. It now seems more accurate to say that the transition has already begun, unevenly and without a final form. We still do not know what is coming. But we can see the shape of the choice before us.
AI can give us more images, more music, more stories and more worlds than any person could experience in a lifetime. The question is whether abundance will deepen culture or merely fill every available silence.
That answer will not be generated for us.
