Converting Sequenced Fuzzy Cognitive Maps to Causal Virtual Worlds with Large Video Generators

Details

The paper introduces a pipeline where feedback fuzzy cognitive maps serve as the causal backbone for AI-generated video. Unlike pure prompt engineering, FCMs allow explicit specification of causal rules and their dynamics define scene sequencing. The workflow: an FCM describes a virtual world and its causal rules. Dynamical meta-rules of the form 'If A then B' describe state transitions, where the FCM's transient feedback dynamics define the causal arrow. An algorithm extracts these meta-rules, a Gemini 3.1 agent writes a script from the sequence, and Veo 3.1 generates video scenes accordingly. The demonstration uses a two-node dolphin-shark undersea ecosystem FCM. This is primarily a proof-of-concept paper. The approach shows how structured causal knowledge can guide generative AI systems — potentially useful for simulation, education, or scenario planning. The use of Google's Gemini 3.1 and Veo 3.1 is noteworthy though the results depend on commercial APIs. Scaling to complex FCMs with many nodes would require additional testing.