GhostWriter: Augmenting Collaborative Human-AI Writing Experiences Through Personalization and Agency
Project Overview
The document explores the role of generative AI, particularly through the example of GhostWriter, an AI-powered writing assistant that leverages Large Language Models (LLMs) to improve personalization and user engagement in educational writing tasks. It addresses the challenges inherent in utilizing LLMs, including the struggle to deliver personalized outputs and the limitations on user control during the writing process. Findings from user studies demonstrate that GhostWriter successfully aids in creative writing by enabling users to refine their writing style and context, ultimately fostering a greater sense of ownership and control over their work. The implications of this technology suggest that generative AI can enhance the educational experience by providing tailored support for students, encouraging creativity, and facilitating a more interactive and personalized learning environment. Overall, the document underscores the potential of generative AI in transforming educational practices by making them more adaptive to individual student needs and promoting active participation in learning.
Key Applications
GhostWriter, an AI-powered writing assistant
Context: Educational context for professionals involved in writing tasks, such as content designers and UX researchers
Implementation: Participants used the system in two sessions focusing on editing and creative writing tasks, with features allowing for style and context personalization.
Outcomes: Users reported enhanced creativity, personalization, and agency in writing, with positive feedback on the ease of use and effectiveness of the tool.
Challenges: Users experienced confusion regarding the roles of style and context, and the system's lack of contextual awareness in generating text.
Implementation Barriers
User Experience Barrier
Participants expressed confusion over how the system's style and context settings interacted and impacted generated text.
Proposed Solutions: Provide clearer guidelines and examples that delineate the roles of style and context in the writing process.
Technical Limitation
The system struggles with maintaining contextual awareness, often failing to generate text that seamlessly builds on previous content.
Proposed Solutions: Enhance the system's ability to understand and remember context within documents to improve text generation.
Feature Limitation
Users desired the ability to store and manage multiple writing styles, indicating a limitation in the current implementation of a singular style.
Proposed Solutions: Implement a feature allowing users to define and select from multiple styles and contexts for different writing tasks.
Project Team
Catherine Yeh
Researcher
Gonzalo Ramos
Researcher
Rachel Ng
Researcher
Andy Huntington
Researcher
Richard Banks
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Catherine Yeh, Gonzalo Ramos, Rachel Ng, Andy Huntington, Richard Banks
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai