Conceptual artworks are conventionally approached as objects of interpretation - visual representations whose meaning is accessed through historical or theoretical reading. Our paper argues that some conceptual artworks function not merely as images of ideas but as epistemic structures that organize and constrain thought, and therefore require activation rather than just interpretation. Focusing on Ágnes Dénes’s Matrix of Knowledge, specifically its Politics and Systems diagram, we treat the artwork as a bounded epistemic environment mapping a topology of political possibility. Concepts function as nodes linked by adjacency and constraint, forming a structured conceptual space rather than a classificatory scheme of ideologies. We formalize the diagram as a conceptual graph and activate it through AI-driven agent-based exploration using reinforcement learning and a structurally oriented topological escape learning. Agents do not represent political actors; they operate analytically within conceptual space, revealing patterns of circulation, stabilization, and conceptual trapping that remain latent in static visual analysis. Our results show that the artwork functions as a learning-capable epistemic system. Reinforcement learning produces path-dependent ideological corridors, while structurally oriented learning reorganizes exploration around connective nodes and reduces conceptual trapping. Methodologically, our study introduces AI-driven visual activation as a new analytical approach in visual studies.
Márton Gosztonyi (Tue,) studied this question.