AI & the future of media and entertainment

World models are AI systems that generate a moving environment frame by frame as you look around it, rather than building it in advance. Early versions keep such a world coherent for a few minutes, and in games, characters driven by a language model already hold conversations without a fixed script. The scenario for 2040: a series generated while it is being watched. The viewer follows a different character, puts a question to a minor figure or picks a gentler turn of events, and the story stays within the world, tone and ending the writers have laid down. The writers' room then no longer designs episodes but a story world with rules. The same title becomes a different evening's viewing for every person watching.

Big games now take years to develop, with teams combining handcrafted worlds and procedurally generated environments. World models can already produce short playable environments from a single image or sentence, and developers use generative AI for sound, faces and dialogue. The scenario for 2040: a game whose world only comes into being where the player looks, with characters who remember what the player did earlier. Every player gets their own map, their own supporting cast and their own course of events. The studio shifts from building to directing: it sets the physics of the world, the style, the difficulty and the story that has to hold together. A small team could then make a world that now takes hundreds of people.

A digital double is an AI model of an artist's or actor's voice, face and way of moving, made with consent and under licence. Labels now sign licensing deals with AI companies for voices, actors have set out in their collective agreements when a digital replica may perform, and a concert series with digital avatars of a pop group has run for years in a venue of its own. The scenario for 2040: a singer's double sings in ten languages, appears in a game, sends a fan a personal birthday song and performs at a venue on the other side of the world. The artist chooses the roles, the brands and the limits. Heirs will manage a voice the way they now manage a music catalogue.

A personal AI agent searches, weighs and sorts information on its own for a single user. A growing share of traffic to news sites already comes from AI systems, roughly one in ten internet users surveyed uses a chatbot for news every week, and publishers are trialling payment per request and standards that set the terms for AI use. The scenario for 2040: much of the news reaches people through agents that put together a morning briefing for each reader. The news organisation supplies what an agent cannot make itself: verified facts, its own scoops, witnesses and context, in a form machines can read and pay for. Revenue comes in per use. A news organisation that is rarely wrong can charge more for every story.

Adaptive music already responds to activity, time of day and heart rate, and artists supply material for these listening experiences. Generative models now produce complete songs with vocals, and major labels have signed licences with the makers of such models. The scenario for 2040: an artist releases an album that goes on producing new songs for years. Melodies and lyrics grow with the listener, with recurring themes that develop over time, whether on a morning run or at a party. The artist sets the musical identity and decides which stories this living repertoire can tell. Fans compare their versions the way they now compare live recordings. The business model moves towards ongoing access to a licensed style, and a fixed recording becomes a special release.

At major tournaments, AI already produces commentary for highlights, and automated cameras bring amateur matches to a live audience. The scenario for 2040: AI rebuilds a match as a three-dimensional model, and from that model each viewer gets a broadcast of their own. One fan follows the game from a virtual camera position next to their favourite player, with analysis that ties in with earlier matches they watched. Another sees the same decisive move from three positions in a row, with commentary in the voice of a former international who agreed to it. Directors design the storytelling formats these broadcasts grow from. Broadcast rights gain an extra dimension: a match that can be watched in countless ways.

A synthetic audience is a group of AI models that behave like viewers, readers or listeners of a given age, taste and background. Market researchers are already trialling these simulated respondents; studies show that on some questions they come close to real answers and on others they are far off. The scenario for 2032: a studio or broadcaster puts a script, a trailer or a format to a million simulated viewers in twenty countries, sees where attention drops and which joke falls flat in which culture, and rewrites before a single scene is shot. Test screenings become a final check. And a production that succeeds against every prediction could then open up a new genre.