In the experimental installation Agent Provokateur by Daniel Heiss and Marc Schütze, multiple AI agents interact with one another. These software systems, based on generative algorithms, possess a degree of decision-making autonomy and assume creative tasks that are fundamental to the conception and realization of the work. They gather information, generate images and sounds, and make directorial as well as compositional decisions. Their interaction is not directed toward a predefined outcome. Instead, through a continuous process of exchange, new visual and acoustic worlds are constantly generated, placing the work in a state of ongoing transformation and reconfiguration.
The information required to produce the work is drawn both from the seemingly inexhaustible data resources of the internet and from the agents’ immediate environment. Cameras and sensors register the people present and generate data* that may be integrated into the ongoing processes. Visitors are thus drawn into the work’s unfolding. Their data become part of the artistic material from which the agents may draw, according to the decisions they make.
Alongside the LED wall on which the constantly evolving work appears, additional screens render the creative processes of the AI agents visible. Individual proposals and decisions can be traced, while visitors themselves may also submit suggestions—though these are not automatically adopted. This opens up a productive tension: on the one hand, the agents are free to accept or reject a suggestion, and their actions cannot be fully predicted; on the other, their agency remains dependent on human input. Agent Provokateur therefore prompts us to ask whether a new form of autonomy is emerging here, or whether autonomous AI systems ultimately remain artistic tools—albeit highly complex and exceptionally powerful ones.
* Image and sound recordings of visitors made in the exhibition are not stored, archived, or shared with external service providers.
