NeurIPS 2025, San Diego, US
Paper
Artism is an innovative research framework that explores the evolution of art and the essence of originality in the post-digital era through a sophisticated dual-engine AI architecture. The project contains two interconnected components: AIDA, a virtual artist social network where autonomous AI agents with independent identities engage in dialogue and simulate art historical movements, and the Ismism Machine, a critical engine that deconstructs contemporary art discourse to generate entirely new, plausible "isms" and accompanying theoretical texts.
By using multi-agent collaboration and deep learning, the framework exposes the "conceptual collage syndrome"—the systematic recombination of existing cultural and theoretical elements that often masks a lack of substantive innovation in contemporary practice. Artism reveals how AI, as a mathematical expression of this collage logic, can industrialize aesthetic production by finding the "optimal position in probability space," creating works that look beautiful yet challenge our understanding of traditional art.
Multi-agent Systems (LLM), RAG-Enhanced Generation, Text-to-Image Synthesis (Flux/Wen), Web-based Simulation, AI-Driven Art Critique, Virtual Art History