IVIE: A Neuro-symbolic Approach to Incremental and Validated Generation of Interactive Fiction Worlds
Summary
IVIE (Incremental & Validated Interactive Experiences) is a neuro-symbolic system designed to automatically generate complete and playable interactive fiction worlds. Building upon the PAYADOR framework, IVIE employs a four-stage incremental pipeline that leverages Large Language Models (LLMs) for creative elements like setting, character, and puzzle design, while a symbolic layer ensures world coherence through validation of spatial connectivity, type correctness, and objective solvability. Human evaluation with eight participants demonstrated that IVIE generates immersive, thematically coherent worlds, achieving 100% objective completion in "Generate" mode and 100% thematic coherence in "Inspiration" mode. The system successfully matched configured parameters (4 Locations, 4 Items, 2 Characters, 1 Puzzle) in 81.25% of generated worlds. Despite successes, challenges include LLM inconsistencies bypassing puzzle logic and validation gaps allowing structurally impossible objectives. IVIE also integrates RAG for improved long-term narrative coherence and uses Pydantic for robust schema validation.
Key takeaway
For AI Engineers developing generative systems that require both creative flexibility and structural integrity, IVIE's neuro-symbolic approach offers a robust blueprint. You should adopt an incremental, multi-stage generation pipeline with explicit symbolic validation at each step to prevent LLM inconsistencies. Consider integrating RAG for enhanced long-term narrative coherence and use schema-guided decoding with tools like Pydantic to enforce output structure, ensuring your generated content remains playable and consistent.
Key insights
IVIE balances LLM creativity with symbolic validation to generate coherent, playable interactive fiction worlds.
Principles
- LLMs provide creativity; symbolic systems ensure consistency.
- Incremental generation with validation gates improves coherence.
- Goal-oriented architecture guides content relevance.
Method
IVIE uses a four-stage pipeline (Adventure Core, World Structure, World Materialization, Challenges) with LLMs for creative content and symbolic validation (Pydantic, DFS) at each stage, plus RAG for memory.
In practice
- Use Pydantic for strict LLM output schema validation.
- Implement RAG for long-term narrative coherence.
- Log turn-by-turn states for debugging and replayability.
Topics
- Interactive Fiction
- Neuro-symbolic AI
- Large Language Models
- Procedural Content Generation
- Symbolic Validation
- Retrieval-Augmented Generation
Code references
Best for: AI Scientist, AI Engineer, Research Scientist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.