Desire paths teach a machine to want in Aggregate

In urban planning, desire lines are the dirt paths people wear into ground the plan never provided. It's such a great phrase! Aggregate, a game from the French real-time collective Distraction.fun, makes them a curriculum: players take turns steering an artificial intelligence through reinforcement learning—punishment and reward—across a wasteland modeled on rural cruising grounds, where each player’s scraped-out route becomes the map the next one inherits.

The game runs on a concept by Vincent Moulinet, in dialogue with Théo Casciani’s novel Models, and treats what landscape architect Gilles Clément calls the third landscape—the abandoned lots and margins where unplanned life takes hold—as the shared habitat of queer desire and machine learning, two systems left to find their own patterns. The controller, 3D-printed with Pierre Moulin, is deliberately thorny to hold: the avatar’s punishment, made physical. Moulinet resurfaced the project this June alongside “À quoi jouait Asha ?,” an essay written for MAGMAH, the journal of Geneva’s Museum of Art and History, on how machines learn to see—and how humans, in turn, learn to play like machines.

Aggregate has worked the circuit—the Grand Palais’s Fun Palace show, A MAZE in Berlin—and joined this June’s Nuit Blanche in Paris. It lands on a question this year keeps sharpening: who is a game for, once the player might be synthetic? Playfool’s working bot farm, which we covered two weeks ago, comes at the same question from the factory floor—machines that play so you can watch. Moulinet’s answer is more intimate: if a machine can be taught to want, the teaching itself is the game.