Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RevTogether: Supporting Science Story Revision with Multiple AI Agents

Yu Zhang, Kexue Fu, Zhicong Lu

TL;DR

This work tackles the challenge of translating complex scientific content into accessible narratives by enabling iterative revision with multiple AI agents. It introduces RevTogether, a multi-agent system comprising two human-like commentator agents (mad scientist and curious girl) and a writing assistant, all powered by GPT-4o, with emotion-enabled avatars and three tiers of user agency. Key contributions include a concrete agent design that captures diverse feedback perspectives, embedding four writing techniques, and a three-level revision workflow implemented in a Flask/React web app. A preliminary study with three non-expert writers suggests that transparent reasoning and emotional cues can enhance learning and collaboration in science storytelling, informing future HCI designs for human-AI co-creation in content creation.

Abstract

As a popular form of science communication, science stories attract readers because they combine engaging narratives with comprehensible scientific knowledge. However, crafting such stories requires substantial skill and effort, as writers must navigate complex scientific concepts and transform them into coherent and accessible narratives tailored to audiences with varying levels of scientific literacy. To address the challenge, we propose RevTogether, a multi-agent system (MAS) designed to support revision of science stories with human-like AI agents (using GPT-4o). RevTogether allows AI agents to simulate affects in addition to providing comments and writing suggestions, while offering varying degrees of user agency. Our preliminary user study with non-expert writers (N=3) highlighted the need for transparency in AI agents' decision-making processes to support learning and suggested that emotional interactions could enhance human-AI collaboration in science storytelling.

RevTogether: Supporting Science Story Revision with Multiple AI Agents

TL;DR

This work tackles the challenge of translating complex scientific content into accessible narratives by enabling iterative revision with multiple AI agents. It introduces RevTogether, a multi-agent system comprising two human-like commentator agents (mad scientist and curious girl) and a writing assistant, all powered by GPT-4o, with emotion-enabled avatars and three tiers of user agency. Key contributions include a concrete agent design that captures diverse feedback perspectives, embedding four writing techniques, and a three-level revision workflow implemented in a Flask/React web app. A preliminary study with three non-expert writers suggests that transparent reasoning and emotional cues can enhance learning and collaboration in science storytelling, informing future HCI designs for human-AI co-creation in content creation.

Abstract

As a popular form of science communication, science stories attract readers because they combine engaging narratives with comprehensible scientific knowledge. However, crafting such stories requires substantial skill and effort, as writers must navigate complex scientific concepts and transform them into coherent and accessible narratives tailored to audiences with varying levels of scientific literacy. To address the challenge, we propose RevTogether, a multi-agent system (MAS) designed to support revision of science stories with human-like AI agents (using GPT-4o). RevTogether allows AI agents to simulate affects in addition to providing comments and writing suggestions, while offering varying degrees of user agency. Our preliminary user study with non-expert writers (N=3) highlighted the need for transparency in AI agents' decision-making processes to support learning and suggested that emotional interactions could enhance human-AI collaboration in science storytelling.

Paper Structure

This paper contains 19 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Design and Implementation
  3. Designing the Commentator Agents
  4. Asking for Comments
  5. Emotional Reactions
  6. Designing the Writing Assistant Agent
  7. Embedding Writing Techniques in Agent's Suggestions
  8. Supporting Revision in Different Levels of User Agency
  9. Human-like comments for high agency
  10. Suggestions on writing techniques for medium agency
  11. Suggestions on revisions for low agency
  12. Implementation
  13. Preliminary User Study
  14. Procedure
  15. Findings and Future Directions
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Character Avatars with different affects for Commentator Agents. Each character has three avatars with positive, neutral, and negative affect respectively. They share "happy" and "calm" for positive and neutral affects, while the "mad scientist" character has "angry" while the "curious girl" character has "disappointed" as negative affect.
  • Figure 2: The Iterative Revision Workflow. RevTogether enables an AI-supported iterative revision process, which starts from free editing and allows the writer to receive supportive information on three levels of user agency throughout the process: (A) Human-like comments with high agency; (B) Writing technique suggestions with medium agency; and (C) Detailed Revision suggestions with low agency.