AI, Creativity, and the Infinite Cycle

Source: The future of creativity on the internet in a world with AI, Jack Conte, 58:33, uploaded 2026-07-16. The talk was recorded at SXSW 2026.

Jack Conte is angry that AI companies trained models on creative work without consent, credit, or pay. He is also drawn to the expressive potential of the technology. The talk refuses to reduce that conflict to a clean pro- or anti-AI position.

His main model is an infinite cycle. Artists learn a medium, build a practice and an income around it, and sometimes find a stable-looking model. A new tool or platform then changes the medium or its economics. The model breaks. Artists rebuild, and the cycle starts again.

Main points

• A new medium usually begins by copying the medium before it. Early film copied theatre, early records copied live performance, and current generators copy drawing, photography, film, writing, and recorded music. Conte calls this imitative stage “slop”.
• Artists learn through pattern recognition, imitation, and a culture built by earlier artists. The work that moves culture forward then breaks some of those patterns without losing meaning.
• New tools can produce new art while destroying jobs and business models. Film sound, recorded music, and synthesis brought both. Any account that keeps only the gain or the loss misses the history.
• Current AI models are strong pattern machines. Conte has not seen evidence that they can make the cultural leaps, take the social risks, or form the human relations that give much art its value.
• Change does not make theft acceptable. AI companies should pay the artists whose work creates model value, just as they pay large rights holders.

Conflict without a camp (00:00-17:00)

Conte opens with a montage of AI arguments: synthetic slop, work taken from artists, and jobs under threat. He shares the anger. Creators’ work has trained models without their consent, without pay, and without credit.

Anger is only half of his response. New tools have always attracted him because they might offer a new way to express something. He feels lucky to be present at a moment that may resemble the arrival of cinema in 1895 or recorded music in the early twentieth century. The anger and excitement remain fully present at the same time.

Social feeds make this hard to say. Their ranking systems sort people into camps, reward pure positions, and hide the private uncertainty that Conte hears in conversations with creators. He says he has spoken with thousands of creators and finds much more conflict in private than public posts suggest.

His request to creators is simple: examine the whole response before choosing a camp. The current moment needs a better discussion than “AI is good” against “AI is bad”.

Learning film by copying it (01:00-08:00)

Conte’s first camera story starts when he is about ten. A friend brings an analogue 8 mm video camera to his house, and they begin making films. Conte begs his parents for a camera, then records school projects, family trips, and action films with friends. He accumulates hundreds of hours of tape.

A double press of the record button opens stop-motion animation to him. He moves a piece of clay, records a frame, moves it again, and discovers that a camera can create movement that never occurred in front of it.

He studies Nick Park’s Wallace & Gromit films one frame at a time with the frame-forward button on a VHS remote. A 25-minute film at 24 frames per second can occupy most of a Saturday. He copies the films closely, including Creature Comforts, then copies the lighting and staging of Indiana Jones and the opening of Pulp Fiction. At the time he thinks he is stealing from his heroes. In retrospect, he is learning a century of film language through imitation.

The story gives him the first half of his model: culture supplies patterns. An artist notices them, copies them, and gains fluency. The harder step is to add a sentence that was absent from the earlier conversation.

Film stops imitating theatre (03:40-08:00)

The Lumière brothers introduced the Cinématographe in 1895. Early filmmakers had no film tradition to copy, so they borrowed from theatre. Elaborate sets filled one wide, fixed shot. Actors used large gestures as if an audience watched from the stalls. The camera reproduced the position of a theatre spectator.

Conte cites Maxim Gorky’s response to a Lumière screening in 1896: the silent, colourless world appeared to him as monotonous grey. Early film could imitate a play, but its flat, silent, monochrome image made it a poor substitute.

Georges Méliès found a language that belonged to the camera. As a stage magician, he understood how a cut could change a box from empty to occupied, make a person appear, or replace an actor with an explosive effect. The audience did not yet know the trick.

A Trip to the Moon (1902) joins those devices into a 14-minute story. Astronomers in top hats build a rocket, fire it from a cannon, travel towards the Moon, land in its eye, explore a mushroom landscape, fight lunar inhabitants, escape over a cliff, splash into the sea, and return as heroes. Cuts change place and time. A point-of-view shot creates travel through space. Substitution cuts turn bodies into explosions. The film still carries painted theatrical sets, yet its geography, effects, and editing can only exist as cinema.

This is Conte’s test for a mature new medium: it stops apologising for what it cannot reproduce and begins using what only it can do.

A recording studio in the garage (08:00-10:00)

Conte’s father plays jazz piano and teaches him the blues scale when he is six. Conte writes songs, joins a cappella groups and jazz combos, and performs at school. Later, his friend Ryan Lerman shows him how to connect a microphone to a computer and record with Pro Tools.

Conte turns his parents’ garage into a studio. He hangs cheap foam from the Container Store on the walls. He doubts that it changes the sound, but it makes the room feel like a booth. He spends the summer recording.

He has recently met Natalie Dawn, who plays bass, writes melodies, and sings. They are about twenty and cannot afford a commercial studio, so they learn the whole process. Conte plays guitar, drums, and piano, while Natalie performs anything her ear can find. They record, mix, and master their own songs.

Recorded music and “canned music” (10:00-14:00)

Conte frames early commercial recording as another medium that first copied its predecessor. A band enters a room, performs once, and leaves behind a record that aims to sound like the live event. Recording has not yet become a compositional space.

James C. Petrillo, president of the American Federation of Musicians, treats records as a direct threat to live employment. Conte reports Petrillo’s estimate that recorded music took 60 per cent of live music jobs. The union bans recording sessions for years, calls session work participation in one’s own funeral, and runs “canned music” ads that show robots feeding instruments into a grinder for profit.

The next decades produce a different idea of recording. On Revolver, the Beatles and producer George Martin use linked tape machines, short drum loops, reversed sound, and vocals sent through Leslie speakers. These are studio acts, not attempts to preserve a concert. Martin describes the studio as a new canvas.

Film took a few years to find cuts, moving viewpoints, and screen geography. Recording took longer to find overdubbing, editing, looping, reversal, and sound that no live band had played in that form. Both histories fit the same sequence: imitation, technical discovery, and a new artistic vocabulary.

“Slop” as the imitative phase (13:00-16:00)

Conte’s account of AI slop is historical rather than permanent. Image models make pictures that resemble drawings. Video models make scenes that resemble camera footage. Music models reproduce styles established by recorded musicians. The result often feels cheap because people already make the originals better and with a life behind them.

He expects the systems to improve. The open question is whether artists will find uses native to the new medium, as Méliès did with film and the Beatles did with recording. That possibility excites him. The way companies acquired the training material still angers him.

Pomplamoose and the first internet model (17:00-19:30)

Conte and Natalie form Pomplamoose as YouTube appears. The site lets them publish music without a record deal or a lucky discovery by an industry gatekeeper.

Their videos show the real recording process. Each part appears as it enters the track: voices, electric and acoustic guitars, piano, plucked strings, Wurlitzer, synthesizers, and background vocals. Natalie edits the sessions into three-minute clips and leaves some of their messing around at the end.

This seems ordinary in 2026, but Conte says YouTube then held music videos and filmed performances, not detailed home-studio construction. Pomplamoose’s process video becomes part of the work.

The videos reach millions of views. The pair sell the songs as MP3s on iTunes, move tens of thousands of units, and become full-time musicians within months. Conte does not compare this with Méliès inventing cinema. He values the smaller advance: for the first time, he feels that he is adding something rather than copying Wallace & Gromit.

Sound film and its human cost (19:30-26:30)

Warner Bros. releases The Jazz Singer in 1927 as the first mainstream feature with synchronised speech. The Vitaphone system synchronises film with recorded sound. Conte gives its theatre price as $9,000 to$15,000, a large investment that audiences quickly reward.

The new films become “talkies”. Conte reports weekly US cinema attendance rising from about 50 million in 1926 to 90 million in 1930. Soviet filmmakers warn in a 1928 statement that synchronised sound could hinder cinema as an art and destroy its formal gains.

The labour fear is concrete. Silent cinemas employ tens of thousands of musicians to perform scores each day. Conte estimates that a fifteen-piece band can cost a theatre about $50,000 per year. A Vitaphone pays for itself within months, runs all day, and removes overtime and union rules. Musicians strike and revive the language of “canned music”. Conte gives the union estimate of 22,000 lost cinema jobs by 1930.

Some screen actors also lose their place. John Gilbert’s delivery draws unintended laughter. Clara Bow hates talkies, struggles with the pressure around her Brooklyn accent, and leaves film after a severe mental health crisis. Charlie Chaplin says he loathes talkies because they damage pantomime. He makes City Lights without synchronised dialogue in 1931 and gives Modern Times sound effects but almost no speech in 1936.

Conte does not claim to know Chaplin’s private reason. He imagines the conflict: accept irrelevance or alter an art form one has spent a life mastering. Chaplin finally makes a full talkie with The Great Dictator in 1940. With Nazi Germany occupying much of Europe and the United States still outside the war, Chaplin ends the satire by speaking straight into the camera against fascism and mechanised cruelty. The technology he resisted lets his voice reach millions. The machine becomes useful when the artist finds the speech that requires it.

A model breaks for the first time (26:30-30:00)

By 2013, magazines present Pomplamoose as part of a new class of internet musician. MP3 income pays for a house and studio. The band tours the United States, plays thousand-capacity rooms, and approaches 100 million catalogue views. Conte and Natalie reinvest the revenue in larger videos, spaces, musicians, and other collaborators.

Then three changes arrive together:

1. YouTube launches the Cosmic Panda redesign, and Pomplamoose’s growth becomes flat.
2. Spotify replaces iTunes as the main listening model, with different economics from paid downloads.
3. Conte starts Patreon, whose early years consume most of his time.

He works at Patreon from about 08:00 to 22:00, records Pomplamoose with Natalie from 23:00 to 03:00, sleeps around 04:00, and returns to work at 08:00. This continues for roughly eighteen months to two years. Patreon operations lead Tyler Palmer visits his apartment and tells him the arrangement cannot continue.

They put Pomplamoose on hold. Conte talks about the band as if it has died. At 27 or 28, he reads a broken business model as a personal artistic failure. He assumes that a better strategist would have found a permanent model able to withstand the world. Later experience changes that belief.

The synthesizer and the orchestra (30:00-35:50)

Laurens Hammond’s Novachord appears at the 1939 World’s Fair and is often described as the first commercial polyphonic synthesizer. Conte describes 163 vacuum tubes, more than a thousand hand-wired capacitors, a weight near 500 pounds, and sounds that can stand in for strings, piano, or trumpet. Popular Mechanics calls it an orchestra in itself.

The musicians’ union bans the instrument because it may displace more orchestral work. Conte argues that some of this fear came true. He says every Billboard number one in 1941 used an orchestra, while almost none do now. The synthesizer is one part of the change, alongside a hard economic limit.

William Baumol’s cost disease names that limit. A string quartet needs the same number of players and roughly the same performance time after productivity rises elsewhere. An orchestra cannot cut the labour in a Schubert quartet without changing the piece. Conte cites a League of American Orchestras analysis in which 46 of the 63 largest US orchestras run deficits.

Synthesis also expands music. Conte lists the Beatles, Stevie Wonder, the Bee Gees, Whitney Houston, James Brown, Prince, Madonna, Radiohead, Björk, and Daft Punk among its users. He brings modular synthesist Bryan Noll, who performs as Lightbath, on stage and asks the audience to hold the instrument’s beauty, new expressive power, labour history, and fear at once. A verdict of simply good or bad would remove most of what happened.

Pomplamoose rebuilds around frequency (36:00-38:30)

After three dormant years, YouTube rewards daily and weekly upload schedules. Conte has seen weekly releases work with his other band, Scary Pockets. He and Natalie adapt Pomplamoose to the same platform condition.

They fly to Los Angeles once a month, record four songs in one day, film the session, and edit four videos for weekly release. One production day creates a month of uploads and a planned 52 videos per year.

The model works. Pomplamoose passes one million monthly Spotify listeners, and its videos receive more than 80 million YouTube views in the first two years. Spotify, memberships, and YouTube ads bring in more money than iTunes downloads had. The revenue pays a large team of musicians, camera operators, mixing engineers, and producers.

This apparent happy ending lasts five or six years. Its temporary success teaches Conte the point he could not see when the first model failed: change does not mean death. Artists can rebuild.

The infinite cycle (38:30-41:00)

Conte rejects the conference story in which someone struggles, finds the answer, succeeds, and ends the talk. Artistic life repeats the disruption after the breakthrough.

An artist spends years finding a form and an income model. Technology changes, the model breaks, and the artist begins again. Another platform, format, crisis, or business change breaks the replacement. Conte pictures the process as a sine wave with no final plateau.

Pomplamoose’s later cycles make the model concrete:

• COVID requires masks and distance. Hidden faces hurt the videos, so the band builds clear isolation boxes that let musicians play without masks.
• TikTok shifts viewing towards vertical video, then other platforms copy it. The band turns cameras sideways before camera bodies support that use well and designs new recording and edit systems.
• A session can yield 50 hours of footage. Editor Athena cuts it into 30-second studio clips, and short-form views rise sharply.
• Those views turn out to be “empty calories”. They convert poorly to Spotify listeners and bring little ad revenue.

The lesson is neither optimism nor surrender. Stability is temporary, and the next rebuild remains part of the work.

The demand on AI companies (41:00-44:20)

Conte tells AI companies that the current change repeats a familiar injury. Artists’ life work enters a new industrial system without consent, credit, or compensation.

He rejects their fair-use defence as inconsistent. The same companies sign large licensing deals with Disney, Condé Nast, Vox, Warner Music, and other rights holders. If training use requires no licence, he asks, why pay those companies? If their rights merit payment, why exclude the individual illustrators, writers, musicians, and other creators whose work also built the models?

This is not a rejection of technology. Conte runs a technology company and says Patreon would die within three years if it did not adopt new technology quickly. He does not expect the world to give artists consistency, and part of the effort to find a new path excites him.

The boundary is direct: acceptance of change does not cancel creators’ rights. AI companies should pay creators for training use. Copyright and other artist rights exist because societies benefit when they support creative work. Building a payment system is difficult, but it is a tractable design and policy problem beside the technical projects these companies already pursue.

This connects to distributed authorship and musical labour, where tools and reused material can widen authorship while hiding who supplied the labour and who controls the result.

The demand on creators (44:20-45:50)

Conte expects the next few years to break many creative income models. The economic damage will be severe, and the personal damage may reach deeper. Creators already ask whether their own art has value when a model can produce almost anything on request.

He feels that hopelessness on his worst days. Experience lets him recognise it as the low point of the same cycle rather than proof that human creativity has ended.

His confidence rests on four features of human art: cultural invention, personal risk, hard-won scarcity, and connection with another conscious life.

Cultural invention beyond the dataset (45:50-49:00)

Conte remembers early UK dubstep through Skream and later US work by Skrillex. These artists do more than produce a new sound. Their fearlessness and experiments change the accepted form of a song. Timbre, physical vibration, and sub-bass can carry a track without the melodies, words, and choruses expected from pop.

Every artist uses earlier culture. The important move is the break that changes what the medium includes.

Conte proposes a thought experiment: train a music model on nothing after 1959. Could it reach the Beatles without John, Paul, George, and Ringo, then Carole King and Stevie Wonder, Prince, Nirvana and Kurt Cobain, Biggie Smalls and hip-hop, Beyoncé, Daft Punk, Frank Ocean, or Jacob Collier? Each name stands for a cultural turn that the past alone did not specify.

He leaves open the possibility that a future system could make autonomous cultural leaps. He has not seen a current model do it. Current generative systems learn statistical patterns across large datasets, while human creation also involves emotion, consciousness, disobedience, and the decision to ignore the apparent consensus.

Pattern recognition explains part of creativity. Conte doubts that it explains the whole act. This sits beside AI mediation, delegation, and emergence, which asks who supplies judgment, selection, data, and direction when a model participates in music.

Risk and ridicule (49:00-51:30)

Hasan Minhaj tells Colin and Samir that good comedy involves risk. A comedian works outside the normal range of socially safe statements, rejects the popular consensus, and offers an unusual view that still feels true. The result can widen what a culture can say.

Conte extends the point to art. A strange model output risks nothing because the model has no standing, body, career, or social place to lose. A person who makes the same move can face ridicule and failure.

He uses three examples:

• The older brother in Sing Street says rock and roll means risking ridicule.
• The hot-dog-finger sequence in Everything Everywhere All at Once makes Conte more interested in the Daniels because its absurdity exposes their willingness to put a strange idea on screen.
• Taylor Swift’s decision to rerecord her catalogue as Taylor’s Version interests him as the act of a person willing to fight the industry system around her work.

The audience responds to the decision and the person behind it, not only the resulting arrangement of sound or image.

Difficulty, scarcity, and skill (51:30-54:30)

Creator Karen X. Cheng gives Conte the next claim: “Art is scarcity. Art is not what’s easy to make, it’s what’s hard.”

He recalls early invisible-selfie-stick and tiny-planet videos made with Insta360 cameras. The effect initially earns millions of views because viewers have not seen it and assume that it takes unusual work. Once the camera setting becomes common, the same image earns little attention. The tool has made the effect abundant and easy.

Conte expects people to remain interested in humans who push the limits of a tool, body, or system through practice and resourcefulness. More compute or money alone does not provide the same interest.

Bryan Noll performs on modular synthesis while student dancer Braylon Browner joins him. The long wordless section demonstrates the point rather than describing it. Conte then notes that Browner may miss only four class days per semester but has missed 28 because so many people want to work with him. The story treats visible demand as a result of trained ability, though it also leaves the cost to the student sitting in the room.

Information, art, and human connection (54:30-58:30)

Conte dislikes putting every creative work under the word “content” because works serve different human needs. A Monty Booker album and a video about plunging a toilet are both media, yet they ask for different kinds of attention.

For a blocked toilet, a fast and accurate answer is enough. The identity and inner life of the person who gives the instruction add little. AI may replace much work in this lane because the information is the purpose.

An Elliott Smith record works through relation with the person who lived the experience in the song. good kid, m.A.A.d city matters through Kendrick Lamar’s life as a Black child in Compton, his account of that place, his pain, and the listener’s relation to him. The kick and chord changes matter, but the record cannot be reduced to its sound.

Conte expects cheap AI dinner jazz in restaurants, AI instructions for household tasks, and ten-second AI news summaries. He also expects people to seek artists who state who they are, take risks, attempt hard things, and form a relation with an audience.

Some of those artists will make new works. Others will make new media and expressive vehicles that become the vocabulary of artists fifty or a hundred years later. A person will feel something that culture has not yet described, search for the exact form that can convey it, and show other people something they have not seen in themselves.

The closing claim is that anything matters because something alive once made it matter. The tool can change, the income can collapse, and the form can be rebuilt. Human relation remains the reason to care about at least some art.

What the argument joins

Conte keeps four questions together:

1. Aesthetic: What can this medium do that an older medium cannot?
2. Economic: Which jobs, prices, and business models does the tool remove or create?
3. Political: Who owns the training material, gets paid, and has enough power to make a licensing deal?
4. Human: Does the work matter as useful output, skilled difficulty, personal risk, cultural invention, or contact with another life?

This makes the talk stronger than a simple historical analogy. A new medium can produce major art and still arrive through an unjust labour settlement. Adaptation can be necessary without making the terms of change fair. This belongs near platformisation and functional listening, human-centred AI and explainability, and the revolt against technology.

People, works, and references named in the talk

• Film: the Lumière brothers and the Cinématographe, Maxim Gorky, Georges Méliès and A Trip to the Moon, Nick Park’s Wallace & Gromit and Creature Comforts, Indiana Jones, Pulp Fiction, The Jazz Singer, City Lights, Modern Times, The Great Dictator, Sing Street, and Everything Everywhere All at Once.
• Music and recording: James C. Petrillo and the American Federation of Musicians, George Martin and the Beatles’ Revolver, Stevie Wonder, the Bee Gees, Whitney Houston, James Brown, Prince, Madonna, Radiohead, Björk, Daft Punk, Skream, Skrillex, Carole King, Nirvana, Biggie Smalls, Beyoncé, Frank Ocean, Jacob Collier, Elliott Smith, Kendrick Lamar’s good kid, m.A.A.d city, and Monte Booker.
• Instruments and economics: Laurens Hammond’s Novachord, William Baumol’s cost disease, the League of American Orchestras, Pro Tools, and Insta360.
• Creators and collaborators: Natalie Dawn, Ryan Lerman, Tyler Palmer, Athena, Bryan Noll / Lightbath, Braylon Browner, Hasan Minhaj, Colin and Samir, Karen X. Cheng, the Daniels, and Taylor Swift.
• Platforms and companies: YouTube, iTunes, Spotify, TikTok, Patreon, Disney, Condé Nast, Vox, and Warner Music.