Proactive Emergent Text Adventures

Today, artificial intelligence has reached a level of power where a generic prompt like "I want to play an RPG, you be the Game Master" or "Be the GM for the adventure Tomb of the Serpent Kings" is enough to be led on an adventure, with the AI taking on the role of GM in a seemingly effective way.

However, relying on an unstructured prompt does not solve several of the tool's intrinsic problems. Issues remain regarding the lack of true randomness, the AI's accommodating nature that seems to grant the player a sort of "plot armor" to advance the story, a latent feeling of "railroading," and a narrative development that tends to be somewhat flat and uninspired.

The real challenge we set for ourselves was to figure out how to formulate prompts that go beyond this "superficial layer" and overcome the limitations we've highlighted. Creating an effective prompt system that fully harnesses the AI's potential for play is far from simple and requires careful thought, technical expertise, and extensive testing.

The "Rubber Wall"

Let’s imagine throwing a ball against a wall. If the wall is made of rubber, it will always adapt to the impact, bouncing the ball back in a functional, compliant way. Whatever you throw at it, the response “works,” but it lacks real consistency. This metaphor perfectly illustrates the underlying problem: a game world based purely on reactive improvisation.

When a player interacts with an artificial intelligence, their actions are met by a system that generates a plausible response in the moment. If the player decides to go right instead of left, the system will invent something interesting to the right. It feels like freedom, but in fact it’s the exact opposite. Since there is no solid pre-existing reality to “push against,” the player’s choices lose their weight. No matter what the player does, the system’s “rubber wall” will always bounce back a narrative that works, but one that is not the result of a genuine interaction with a coherent world. Their freedom is an illusion, because their decisions do not alter an existing world, but merely trigger the instant creation of a scene. This becomes more or less evident with an approach that does not involve playing through a predetermined adventure.

From a Reactive Narrative to a Proactive World

This new project aims to overcome a central challenge in AI-generated text adventures: the lack of real impact from player choices. Too often, interaction is reduced to an "illusion of freedom," where every action is absorbed by reactive improvisation that, like a "rubber wall," adapts without offering lasting consequences. Our mission is to build "living worlds for meaningful stories" by replacing this model with a solid and coherent game world.

The solution we experimented with was to structure a system of “Fronts,” inspired by the TTRPG Dungeon World, to implement a kind of “preparation” and try shifting from a reactive narrative model to a proactive simulation model. Through this system, we create a dynamic ecosystem of characters with their own agendas and evolving threats. This way, player decisions don't just generate text; they interact with a world that reacts according to its own internal logic, creating consistent yet often unexpected consequences and making every choice truly meaningful.

The “Fronts” system we propose rests on two pillars:

• A Cast of Characters with Goals: Key non-player characters (NPCs) are defined, each with their own motivations, plans, and objectives—characters who act actively in the world to achieve their aims.
• Fronts as Evolving Situations: A “Front” is a threat or complex situation that progresses over time (for example, an impending invasion, the spread of a curse, a war between factions). These threats advance according to their own internal logic, driven by the actions of the involved NPCs and the actions of the players.

In practice, the goal was to create an ecosystem of causes and effects that gives the world its own coherence and vitality.