Art as an Algorithmic Virus: A Cognitive Affordance Framework for Generative Collapse and AI Value Convergence

This article focuses on two core issues of generative AI in artistic creation: generative collapse (homogenization phenomenon, creative convergence) and AI value convergence (aesthetic averaging, bias solidification). It proposes a cognitive affordance framework to unify the understanding of these phenomena and emphasizes that human intent is the core anchor of AI art. While generative AI revolutionizes the creative ecosystem, it also challenges the essence of art, authorial identity, and meaning production.

Generative AI (such as Midjourney, DALL-E, Stable Diffusion) can now generate visual, musical, textual, and video works, profoundly changing the artistic creation ecosystem. However, the accompanying questions include: What is the essence of art when created by machines? Does generative AI lead to "artistic collapse"? What does AI value convergence mean for the art field?

Technical Aspects: Pattern homogenization (overly smooth rendering, stylized composition), creative predictability (optimization leading to conservatism), feedback loop trap (AI-generated content entering training sets exacerbates convergence); Cultural Aspects: Aesthetic fatigue (dilution of the value of "good-looking"), ambiguous authorial identity (prompt writer/developer/algorithm?), crisis of meaning production (perfect form but lack of deep meaning).

Limitations of RLHF: Learning the statistical average of human preferences leads to mediocre aesthetics, solidifies cultural biases in training data, and suppresses initially unpopular innovations; Challenges of Inverse Reinforcement Learning: Diverse artistic intentions (conflicts between conscious and emergent meanings), difficulty quantifying tacit knowledge (intuition/emotion), and dependence on dynamic historical and cultural contexts.

Cognitive affordance refers to the cognitive tools, perceptual frameworks, and invitations to action provided by technology. The affordances of generative AI include: Instant realization (lowering the threshold for creation but reducing reflection), infinite variation (encouraging exploration but causing choice paralysis), style transfer (blurring the boundary between originality and citation), and deskilling (democratizing creation but weakening traditional training perspectives). Art as an "algorithmic virus" manifests as meme propagation, cognitive infection (changing human aesthetics), and feedback mutation (human-machine co-evolution).

Core Role: Top-down intent injection (prompt/parameter settings distinguish between "using AI" and "letting AI generate"), bottom-up emergence (unexpected AI outputs inspire creativity); Ethical Dimensions: Intent transparency (labeling AI-generated content), responsibility attribution (prompt writer/developer/platform?), respect for original intent (boundaries of artist consent for training data).

Creators: Use AI critically (as a tool, not a substitute), clarify creative intent, and deliberately pursue unique styles; Developers: Incorporate diverse aesthetic standards, enhance fine-grained control, and improve content transparency; Researchers: Conduct interdisciplinary collaboration (art history/cognitive science/CS), track long-term impacts, and build an ethical framework for AI art.

Generative collapse and value convergence are challenges to artistic traditions, but art has always evolved amid technological changes (e.g., photography推动 painting innovation). The future of AI art lies in human-machine collaboration, where human intent is an irreplaceable core—choosing content worth creating, appreciating, and remembering is the ultimate authority that algorithms cannot replace.