AI is having its 1895
In 1895, Auguste and Louis Lumière, the French inventors behind the Cinématographe camera-projector, helped bring moving pictures to paying audiences. When Russian writer Maxim Gorky saw a Lumière programme the following year, he described “a world without sound, without colour”, where everything was “dipped in monotonous grey.” His review was perceptive. It also caught a medium before artists had discovered what belonged to it.
In 2026, we ask a large language model, software that predicts text from patterns in training data, to write a poem. We ask image-generating software to imitate ink, film grain or a human hand, then treat the results as a verdict on AI-generated creative work. I think we are repeating the 1895 error: judging a medium by its first attempts to copy an older one.
Jack Conte, a musician in the duo Pomplamoose and co-founder of Patreon, a membership platform for creators, gives this phase a useful name. He calls slop “that lazy phase after the introduction of a new medium”, when people reproduce things humans already make better. My take is that slop is neither harmless nor final. It is a symptom of imitation, and the right response is to search for the form that imitation cannot reach.
AI slop shows that imitation is easy. It cannot tell us what artists will discover once they stop using AI to mimic older media.
Every medium begins by borrowing
Conte describes early filmmakers placing the camera where a theatre audience member would sit. Georges Méliès, a French stage magician turned filmmaker, broke that frame with a stop-and-restart trick: pause the camera, change the scene, then resume recording. In A Trip to the Moon, his 1902 film about a lunar expedition, cuts between separately filmed shots made bodies vanish and locations change. Editing no longer reproduced theatre. It created events that belonged to film.
Recorded music began with the same borrowed job: preserve a performance. On Revolver, the 1966 album by the British rock group the Beatles, the musicians and their producers used tape loops, short pieces of magnetic recording joined so they repeat, along with sounds played backwards. Recording equipment stopped being only a container and became part of the composition.
Charlie Chaplin, the actor-director famous for silent films, shows that exploration does not require blind enthusiasm. He resisted synchronized dialogue for years, then made speech central to The Great Dictator, his 1940 film ending with a humanist appeal against dictatorship. I think the lesson is subtler than “new tools always win.” A medium matures when artists discover where a new capacity strengthens their purpose and where restraint serves it better.
Cuts, tape manipulation and deliberate speech gave artists choices that older forms could not offer. A tool becomes a medium when its distinct capacity starts carrying the artist’s intention.
The transition sends a bill
The comforting version of media history skips the people who paid for change. When talkies, films with synchronized sound, arrived in 1927, the American Federation of Musicians, a United States musicians’ union, says about 22,000 theatre musicians lost their jobs. Cinemas no longer needed the live accompaniment used for silent films. The new form expanded cinema while removing an existing livelihood.
The same anxiety appeared around electronic instruments. In 1939, the union ruled that adding a Novachord, an early electronic keyboard able to play several notes at once, or a Hammond electric organ could not reduce the number of musicians in an ensemble. This was a workforce-protection rule, not a blanket ban. It addressed the fear that one player and a new machine could replace several musicians.
Economists William Baumol and William Bowen named part of the underlying pressure in 1966: performing-arts “cost disease”. A live symphony still requires roughly the same human time while other industries become more productive, making performed work relatively expensive beside cheap copies. AI increases that pressure by making AI-generated output abundant. That abundance can lower a market price without reducing the human meaning of the work.
I think honest exploration has to hold both outcomes in view. We should search for AI’s own creative form without making creators whose work is used by AI systems carry the whole cost. My standard is that commercial users should be able to show why they may use that work, deliver any payment they promise, and name a person who can resolve a dispute or remove material when usage rights cannot be shown. History gives us courage to experiment. It does not give us permission to send someone else the bill.
A new medium can open artistic territory while destroying paid work. Exploration deserves support only when creators can understand the terms, receive what was promised, and challenge misuse through a real decision-maker.
AI has to prove it is more than a shortcut
Conte pushes beyond history with a thought experiment. Could a music-generating model trained only on recordings available through 1959 have discovered the Beatles’ later studio experiments or a comparable break from convention? I don’t know. The question does not prove that models cannot surprise us. It exposes the difference between producing an unusual pattern and releasing a choice into culture under your own name.
A model can output something rare, but it does not suffer when the work is ridiculed or when an audience walks away. The artist who chooses and releases the result does. That responsibility is part of why human connection matters. Conte uses American rapper Kendrick Lamar’s good kid, m.A.A.d city as his example. The 2012 album tells a story of adolescence in Compton, a city in Los Angeles County, and Conte says it matters to him because of Lamar’s story and pain, not only its sound. Output alone cannot supply that history.
This is where optimism needs a test. Call it the substitution test. Remove live AI generation, the system’s ability to create or alter part of a work for a particular audience in real time, and replace it with the best noninteractive human-made alternative, a finished work that cannot respond. If the central experience survives, AI was a production shortcut. If it collapses, we have evidence that responsiveness belongs to the medium rather than merely speeding up an older form.
I think the 1895 analogy should expire if it never clears that bar. By 2030, we should be able to point to at least one creative experience that sustains a willing audience after its novelty fades and whose defining effect disappears under the substitution test. If a song, film, game or interactive work with every possible response written in advance can replace AI generation without losing what the audience values, then AI has not yet become a distinct creative medium. It is a powerful production system, and we should call it that.
Using AI only to fake a drawing is like entering the holodeck in Star Trek, the long-running science-fiction franchise, and hanging a poster on its wall. The fictional room creates a world that can respond to its visitor; the poster cannot. I think the frontier lies in that response, shaped by an artist who decides the boundaries and answers for what happens inside them.
Bad imitations deserve criticism, but they do not deserve the final word. The next cultural leap begins when an artist uses AI’s strange capacities to make an experience that could not exist without them, then stands behind the choice. The machine may open the possibility, but a human being still has to make it matter.
AI earns medium status when its responsiveness creates a valued effect that no noninteractive human-made work can preserve. Until then, slop is evidence of imitation, while creator rights and human accountability determine whether exploration deserves support.
