Where Do I ‘Add the Egg?’: Exploring Agency and Ownership in AI Creative Co-Writing Systems DOI: XXXXXXX.XXXXXXX Conference: ACM Conference on Human Computer ; Date; Location CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing HCI theory, concepts and models

Dashiel Carrera, Jeb Thomas-Mitchell, Daniel Wigdor · 2025

Read Where Do I ‘Add the Egg?’: Exploring Agency and Ownership in AI Creative Co-Writing Systems DOI: XXXXXXX.XXXXXXX Conference: ACM Conference on Human Computer ; Date; Location CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing HCI theory, concepts and models by Dashiel Carrera, Jeb Thomas-Mitchell, Daniel Wigdor (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-15440.txt.

Introduction

Dashiel Carrera, Jeb Thomas-Mitchell, and Daniel Wigdor ask who owns words written with an AI helper. Such tools blur who is in charge of a text, which slows their uptake. The team first looked at tools on the market. They then built three writing tools with the same core but three styles: one feels like an agent, one like a tool, and one like magic.

They spoke with eighteen writers, both paid and unpaid, as they used the tools. The styles shifted where writers felt control and who they saw as the author. Tool like designs moved the felt points of control, while agent like designs stressed ideas over words. The team end with design advice on how style shapes use.

Catalog record

CREATORDashiel Carrera, Jeb Thomas-Mitchell, Daniel Wigdor
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-15440

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Where Do I ‘Add the Egg?’: Exploring Agency and Ownership in AI Creative Co-Writing Systems DOI: XXXXXXX.XXXXXXX Conference: ACM Conference on Human Computer ; Date; Location CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing HCI theory, concepts and models CCS: Human-centered computing Empirical studies in HCI Dashiel Carrera, Jeb Thomas-Mitchell, Daniel Wigdor Abstract . AI co-writing systems challenge long held ideals about agency and ownership in the creative process, thereby hindering widespread adoption. In order to address this, we investigate conceptions of agency and ownership in AI creative co-writing. Drawing on insights from a review of commercial systems, we developed three co-writing systems with identical functionality but distinct interface metaphors: agentic, tool-like, and magical. Through interviews with professional and non-professional writers (n = 18), we explored how these metaphors influenced participants’ sense of control and authorship. Our analysis resulted in a taxonomy of agency and ownership subtypes and underscore how tool-like metaphors shift writers’ expected points of control while agentic metaphors foreground conceptual contributions. We argue that interface metaphors not only guide expectations of control but also frame conceptions of authorship. We conclude with recommendations for the design of AI co-writing systems, emphasizing how metaphor shapes user experience and creative practice. 1. Introduction The advent of Large Language Models (LLMs) has opened new artistic possibilities for creative writers, resulting in novel forms of poetry, generative novels, and computational literature ( Whalen 2023 ; Marche 2023 ) . This relatively new yet increasingly pervasive form of Artificial Intelligence (AI) has lowered the barrier to entry for amateur or unpracticed creative writers, enabling them to explore genres and forms that may have previously felt inaccessible. Despite the new artistic possibilities brought about by AI, writers’ perceptions of these platforms remain mixed, with many skeptical of whether AI can yield sufficiently creative results ( Chakrabarty et al. 2024 ) or whether the introduction of AI into their writing practice would disrupt or threaten their creative integrity ( Asmelash 2023 ; Wasi et al. 2024 ) . One of the ways writers perceive their interactions with AI co-writing systems is through their perception of agency when they use the system and their perception of ownership over the final writing product ( Gero et al. 2023 ) . A writer’s perception of agency is the extent to which they feel a sense of control while using the system, and a writer’s perception of ownership is the extent to which they self-identify as an author of the writing co-produced with the AI. To support broader adoption of AI co-writing systems among writers, it is crucial to consider perceived agency and perceived ownership, as these factors can influence a writer’s long-term satisfaction with such systems ( Gero et al. 2023 ) . Previous work has identified agency and ownership as key factors in how writers experience AI co-writing ( Lee et al. 2024 ) and has characterized what types of ownership matter to writers in general ( Reza et al. 2025 ) . Additional work has demonstrated that agency and ownership are generally correlated: writers who feel greater control while AI co-writing generally feel greater ownership over the results ( Draxler et al. 2024 ) . However, little is known about how creative writers conceptualize agency and ownership during interactions with AI creative co-writing systems. This is especially surprising due to creative writing’s rich history of deriving multiple conceptualizations of agency and ownership in the creative process: from listening to the muse ( Shelley 1910 ) to post-modern challenges of authorship ( Barthes 2016 ) to aleatory or probabilistic methods of creation ( Tzara 2018 ) . To explore agency and ownership during AI creative co-writing, we use interface metaphors as a design probe. An interface metaphor is a design concept where familiar objects or concepts are used to represent and facilitate interaction with digital systems, making them easier to understand ( Marcus 1992 ) . Phil Agre’s Critical Technical Practice ( Agre 1997 ) argues that by analyzing the particular metaphors and language of a field of technology, we can identify what the field marginalizes. Interface metaphors are one of the most effective means of altering user perceptions and mental models ( Carroll et al. 1988 ) , having previously been used in mobile interfaces and voice-based user interfaces ( Li and Luximon 2022 ; Desai et al. 2024 ) . We seek to understand how interface metaphors marginalize or afford particular conceptions of agency and ownership. Additionally, by presenting participants with different metaphors, we elicit a broader range of conceptualizations of agency and ownership than those currently available in the literature. In furtherance of this exploration, we introduce the following research questions: RQ1 : What are the different ways writers conceptualize agency and ownership in AI creative co-writing? RQ2: How does the use of interface metaphors in AI creative co-writing systems affect these conceptualizations? In order to determine which interface metaphors to study, we performed a review of popular interface metaphors used in commercial AI creative co-writing systems today. Based on this review and prior literature ( Lee et al. 2024 ) , we identified three interface metaphor archetypes that were the most prevalent: tool-like, agentic, and magic. We then created three AI co-writing systems, each with a different interface features that exemplified each metaphor. We conducted an exploratory study followed with a semi-structured interview (n = 18) where nine professional and nine non-professional writers with creative writing experience were presented with one of three interface metaphors. In the interview, we explored how writers conceptualized agency and ownership in their co-writing experiences. Our analysis revealed multiple forms of agency and control in AI creative co-writing, including three subtypes of control—explicit, implicit, and process-based—and three subtypes of ownership—stylistic, conceptual, and effort-based. Notably, agentic metaphors fostered conceptions of ownership that emphasized the AI’s conceptual contributions. This paper makes three contributions. First, it presents a preliminary review of interface metaphors in common commercial AI co-writing systems, which should ground future work on AI co-writing system interfaces. Second, it introduces a taxonomy of agency and ownership in AI creative co-writing. Finally, it offers design recommendations for future AI co-writing systems, highlighting how interface metaphors shape writers’ expectations of control and authorship. 2. Related Work In this section, we detail research on AI co-writing systems that describes the common dynamics and expectations writers have for these systems and examine conceptions of agency and ownership for users of these systems. We also review research that describes how interface metaphors shape a user’s expectations and interactions with an AI system. 2.1. AI Writing systems The idea that computational processes could contribute to writing has a long history, with a very early example being mechanical machines that could generate lines of Latin verse ( Sharples and y Pérez 2022 ) . Early work in Computer Science, by Roemmele, presented an AI co-writing system for writers that generated suggestions for the next sentences in a story using information retrieval techniques ( Roemmele and Gordon 2015 ) . Later work by Roemmele and Gordon extended this application to use neural networks ( Roemmele 2016 ) . A 2024 literature review of intelligent and interactive writing assistants by Lee et al. found that since

Read the complete text