Between Predictability and Randomness: Seeking Artistic Inspiration from AI Generative Models

2025-06-14

AI Summary

This study investigates how AI-generated poetic lines function as catalysts for human artistic creativity, focusing on the mechanisms by which semantic openness and unconventional imagery combinations stimulate creative agency. To address this, we comparatively analyze fragmented poetic outputs from an LSTM-VAE model against coherent poems generated by large language models (LLMs), integrating close poetic analysis with practice-based creative experiments. Our key contribution is the first systematic articulation and empirical validation of “productive indeterminacy”—a construct encompassing fragmentation, semantic gaps, and syntactically anomalous collocations—as a more potent driver of human ideation and organic narrative emergence than technical fluency or structural completeness. We propose an “incompleteness-oriented” paradigm for AI-human co-creation and demonstrate empirically that original poems composed in response to LSTM-VAE fragments exhibit significantly greater structural spontaneity and imagistic resonance than those inspired by LLM outputs.

Technology Category

• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Humans and AI: Game Design — Procedural Content Generation & Storytelling
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Artistic inspiration often emerges from language that is open to interpretation. This paper explores the use of AI-generated poetic lines as stimuli for creativity. Through analysis of two generative AI approaches--lines generated by Long Short-Term Memory Variational Autoencoders (LSTM-VAE) and complete poems by Large Language Models (LLMs)--I demonstrate that LSTM-VAE lines achieve their evocative impact through a combination of resonant imagery and productive indeterminacy. While LLMs produce technically accomplished poetry with conventional patterns, LSTM-VAE lines can engage the artist through semantic openness, unconventional combinations, and fragments that resist closure. Through the composition of an original poem, where narrative emerged organically through engagement with LSTM-VAE generated lines rather than following a predetermined structure, I demonstrate how these characteristics can serve as evocative starting points for authentic artistic expression.

Problem

• Exploring AI-generated poetic lines for artistic inspiration
• Comparing LSTM-VAE and LLM outputs for creative stimuli
• Demonstrating LSTM-VAE's evocative impact through semantic openness

Innovation

• LSTM-VAE generates evocative poetic lines
• Combines resonant imagery with indeterminacy
• Semantic openness stimulates artistic expression