Sci-Fi Spark: A Human-AI Co-Creation System for Science Fiction Ideation

Paper Title

Sci-Fi Spark: A Human-AI Co-Creation System for Science Fiction Ideation

Publication Info

• Topic area: Human-AI collaboration in creative writing, specifically for science fiction ideation.
• Keywords: Human-AI co-creation, science fiction, ideation, large language models, novelty, consistency, visualization, speculative fiction, creativity support, iterative workflows.

Background and Problem

• Problem / challenge: Existing large language models (LLMs) used for creative writing often fail to generate novel and consistent ideas, especially in worldbuilding-intensive genres like science fiction. Current interaction paradigms do not support the fragmented and iterative workflows typical of speculative fiction ideation.
• Significance: Addressing these gaps is critical for enhancing the creative capabilities of writers and improving the usability of AI tools in complex narrative tasks.
• Motivation and related work: Prior research has explored LLMs for plot development and story infilling, but these approaches often produce predictable outputs and lack mechanisms for maintaining internal and temporal consistency. Visualization tools for writing have been limited to single relationship types, failing to represent the multi-dimensional relationships required for speculative fiction. This paper builds on these insights to address the specific needs of science fiction writers.

Solution

• Proposed approach: Sci-Fi Spark, a human-AI co-creation system, integrates an Ideation Canvas for visualizing idea relationships, a Novelty Generator for producing novel worldbuilding inspirations, and a Consistency Generator for generating context-aware storyline suggestions.
• Novelty:
  1. Introduction of the Ideation Canvas to visually represent and manage internal and temporal consistency relationships in a unified structure.
  2. Operationalization of computational creativity strategies (combinational and transformational) to enhance novelty in worldbuilding.
  3. Development of a system that supports nonlinear and iterative workflows for speculative fiction ideation.
• Procedure and key techniques:
  1. Ideation Canvas: Visual interface with idea blocks (worldbuilding or storyline), internal consistency lines (logical, spatial, societal, psychological, thematic), and temporal consistency lines.
  2. Novelty Generator: Uses combinational and transformational creativity strategies based on the TVTropes database and Boden’s theory of computational creativity.
  3. Consistency Generator: Leverages structured context from the Ideation Canvas to generate storyline suggestions that align with internal and temporal consistency.
  4. Workflow: Supports iterative and nonlinear ideation by allowing writers to alternate between worldbuilding and storylining, with adaptive inspiration from the generators.

Results

• Concrete findings:
  • Novelty Generator improved novelty by 55.4% (combination mode) and 56.3% (transformation mode) compared to baseline.
  • Consistency Generator achieved higher ratings for internal consistency (66.3%) and temporal consistency (61.4%) compared to baseline.
  • Adoption rate of system-generated ideas was significantly higher (45.1%) than the baseline (18.1%).
• Advantage over baselines:
  • Sci-Fi Spark outperformed Sudowrite in inspiring novel ideas and supporting structured visualization of narrative relationships.
  • Provided better support for nonlinear workflows and iterative co-creation.
• Experiments / evaluation:
  • Technical evaluation: Pairwise comparisons of novelty and consistency using GPT-4-generated outputs.
  • User study: 12 participants compared Sci-Fi Spark with Sudowrite in two ideation sessions, focusing on creativity, structure visualization, and co-creation experience.
  • Metrics: Generative utilization, structural connectivity, co-creation iteration, and subjective survey responses.
• Limitations and future work:
  • Cultural mismatch between the English-language trope database and Chinese participants.
  • Short-term study duration limited insights into long-term usage and cognitive impacts.
  • Lack of multimodal input (e.g., visual elements) for abstract or speculative concepts.

Summary

Sci-Fi Spark is a human-AI co-creation system designed to address the challenges of novelty and consistency in science fiction ideation. By integrating the Ideation Canvas, Novelty Generator, and Consistency Generator, the system supports nonlinear and iterative workflows, enabling writers to manage complex narrative structures and generate creative ideas. Technical evaluations and user studies demonstrate its effectiveness in enhancing creativity, aligning with writers’ cognitive processes, and outperforming existing tools like Sudowrite. Future work could expand the system’s cultural adaptability, support long-term projects, and incorporate multimodal inputs for richer creative support.