The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) AI Storytelling Systems with Mark Riedl - #478
Apr 26, 2021
Mark Riedl, a Professor at Georgia Tech, discusses his pioneering work in AI storytelling systems. He explains how AI can predict what happens next in a story by leveraging large language models like GPT-3. The conversation dives into the art of creating suspense and emotional resonance in narratives, as well as the challenges of aligning AI with human thought processes. Riedl also highlights the importance of model explainability and the potential of integrating symbolic systems with neural networks to enhance narrative coherence.
AI Snips
Chapters
Transcript
Episode notes
Human Storytelling vs. Computers
- Storytelling is fundamental to human experience, used for entertainment, education, and connection.
- Computers lack this rich communication modality, exemplified by Hemingway's six-word story.
LLMs and Storytelling
- Large language models (LLMs) excel at generating plausible-sounding text but lack planning and theory of mind.
- Stories require planning to achieve a goal and theory of mind to consider the listener's thoughts.
Improving Coherence with Neuro-Symbolic Models
- Current research uses neural language models with a symbolic control layer to improve coherence.
- This symbolic layer represents a reader model by tracking facts and ensuring consistency.