Ai.llude: Encouraging Rewriting AI-Generated Text to Support Creative Expression

David Zhou,Sarah Sterman

DOI: https://doi.org/10.1145/3635636.3656187

2024-05-28

Abstract:In each step of the creative writing process, writers must grapple with their creative goals and individual perspectives. This process affects the writer's sense of authenticity and their engagement with the written output. Fluent text generation by AIs risks undermining the reflective loop of rewriting. We hypothesize that deliberately generating imperfect intermediate text can encourage rewriting and prompt higher level decision making. Using logs from 27 writing sessions using a text generation AI, we characterize how writers adapt and rewrite AI suggestions, and show that intermediate suggestions significantly motivate and increase rewriting. We discuss the implications of this finding, and future steps for investigating how to leverage intermediate text in AI writing support tools to support ownership over creative expression.

Human-Computer Interaction

What problem does this paper attempt to address?

The problem this paper attempts to address is how to encourage authors to rewrite AI-generated text through the design of specific AI-generated text suggestions during the creative writing process, thereby supporting the authors' self-expression and creative process. Specifically, the researchers focus on:

1. **How to design AI suggestions to encourage rewriting**: The researchers hypothesize that by generating imperfect intermediate text, authors can be encouraged to engage in more rewriting activities, thereby promoting higher-level decision-making and reflection.
2. **How rewriting AI-generated text affects the author's personal writing process**: The researchers aim to understand the impact of these suggestions on the authors' creative process and final work by analyzing the authors' behavior when using AI suggestions.

To explore these issues, the researchers designed an experiment involving 27 creative writers using a custom text editor that provided two types of AI-generated suggestions:
- **Fluent continuation**: Smooth text that directly continues the user's current text.
- **Intermediate suggestions**: Composite text containing multiple fragmented ideas that require the author to modify them to integrate into the story.

By comparing these two types of suggestions, the researchers hope to reveal the impact of different types of AI suggestions on authors' rewriting behavior and explore how to design AI writing assistance tools to better support authors' creative processes and self-expression.