GhostWriter: Augmenting Collaborative Human-AI Writing Experiences Through Personalization and Agency (2402.08855v2)

Abstract: Writing is a well-established practice to support ideation and creativity. While LLMs have become ubiquitous in providing different kinds of writing assistance to different writers, LLM-powered writing systems often fall short in capturing the nuanced personalization and control necessary for effective support and creative exploration. To address these challenges, we introduce GhostWriter, an AI-enhanced writing design probe that enables users to exercise enhanced agency and personalization. GhostWriter leverages LLMs to implicitly learn the user's intended writing style for seamless personalization, while exposing explicit teaching moments for style refinement and reflection. We study 18 participants who use GhostWriter for editing and creative tasks, observing that it helps users craft personalized text and empowers them by providing multiple ways to steer system output. Based on this study, we present insights on people's relationships with AI-assisted writing and offer design recommendations to promote user agency in similar co-creative systems.

• The paper introduces GhostWriter, an AI-assisted writing tool that enhances personalization and user agency through implicit and explicit feedback.
• It employs inline LLM-assisted features and context-sensitive prompting to dynamically adapt writing style and better integrate into user workflows.
• Study results from 18 participants highlight the significance of explicit user input in refining AI-generated content and meeting diverse writing needs.

Enhancing Human-AI Collaborative Writing with GhostWriter: A Study on Personalization and User Agency

Introduction to GhostWriter

GhostWriter introduces an innovative approach to AI-assisted writing, focusing on elevating user agency and personalization. Developed through a collaboration between Harvard University and Microsoft, GhostWriter leverages LLMs to adapt to and learn from users’ writing preferences both implicitly and explicitly. This is achieved by analyzing users’ text inputs, direct style edits, and annotated likes and dislikes. The paper involving 18 participants across various writing tasks sheds light on the potential of such systems to provide a more personalized and controlled writing assistance experience.

The Design and Functionality of GhostWriter

GhostWriter is designed based on a set of core principles aimed at preserving user agency, offering familiar interaction metaphors, integrating seamlessly into users’ workflows, and providing transparency regarding the system's understanding of the user's style. Key features include:

• Inline LLM-assisted Features: Abilities such as text rewriting, continuation, and context-sensitive prompting that all adapt to the defined user style and context, offering personalized text generation.
• Explicit Style Teaching Mechanisms: Users can directly inform the system of their preferred writing styles through likes, dislikes, and manual style description edits.
• Context and Style Summaries: The system provides summaries of the captured style and context, allowing users to understand and refine the system’s interpretations.

This design effectively turns the writing environment into a canvas where users can articulate their desired writing style and context, thereby enabling the LLM to generate content that aligns closely with their intentions.

Examination of User Interaction and Feedback

The paper presents several key findings:

1. Dynamic Style Adaptation Introduction: Users effectively utilized GhostWriter’s features to tailor the writing style, reflecting a strong desire for more personalized content from AI writing systems.
2. Importance of Explicit Personalization: The likes/dislikes feature was notably appreciated for its simplicity and effectiveness in refining the learned writing style, underscoring the value of direct user input in AI-mediated processes.
3. Demand for Multi-faceted User Agency: Feedback highlighted a need for finer-grained control over text generation and style application, suggesting that further enhancements in user-driven customization options could improve experience and outcomes.

Implications and Future Directions

GhostWriter's approach to personalization and user agency opens new avenues for the design of AI-assisted writing tools. It suggests that users are looking for more than just efficiency from these tools; they seek a partnership where their style and voice are recognized and respected by the AI. The paper also points to the importance of offering multiple mechanisms for style personalization, catering to diverse user preferences and writing stages.

Looking ahead, the development of AI writing systems could benefit from incorporating capabilities for handling multiple styles and contexts, adjusting AI outputs with greater specificity, and integrating format customization as part of style personalization. Moreover, understanding and mapping the evolving nature of the human-AI collaborative relationship in creative processes will be crucial in designing systems that are not just tools, but partners in the writing process.

GhostWriter's exploration into augmenting human-AI collaborative writing experiences through personalization and agency presents a promising paradigm. It demonstrates the potential of AI to serve as a capable and adaptable assistant in the creative writing process, responding to the user's stylistic preferences and offering personalized support that enhances rather than overrides the human element.