Exploring Collaboration Patterns and Strategies in Human-AI Co-creation through the Lens of Agency: A Scoping Review of the Top-tier HCI Literature

Shuning Zhang, Hui Wang, Xin Yi · 2025

Read Exploring Collaboration Patterns and Strategies in Human-AI Co-creation through the Lens of Agency: A Scoping Review of the Top-tier HCI Literature by Shuning Zhang, Hui Wang, Xin Yi (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-2507-06000.txt.

Introduction

Shuning Zhang and co-authors review how agency splits between humans and AI in joint making. The work dates to 2025. As AI turns from tool to team mate, who holds the reins in each step counts more. Their field's lens shows the live push and pull of that split in use.

The team read over one hundred papers from top meets of the past twenty years. They built a frame that sorts agency forms, rein tools, and use scenes. A full list of rein tools shows how agency is set in code and in flows. A cross-scene map ties agency forms to varied joint crafts. The review ends with tips for trust and ethics.

Catalog record

CREATORShuning Zhang, Hui Wang, Xin Yi
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE155 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-06000

Browse the full library · About the Tower of Babel

Full text

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

Exploring Collaboration Patterns and Strategies in Human-AI Co-creation through the Lens of Agency: A Scoping Review of the Top-tier HCI Literature Shuning Zhang, Hui Wang, Xin Yi Abstract . As Artificial Intelligence (AI) increasingly becomes an active collaborator in co-creation, understanding the distribution and dynamic of agency is paramount. The Human-Computer Interaction (HCI) perspective is crucial for this analysis, as it uniquely reveals the interaction dynamics and specific control mechanisms that dictate how agency manifests in practice. Despite this importance, a systematic synthesis mapping agency configurations and control mechanisms within the HCI/CSCW literature is lacking. Addressing this gap, we reviewed 134 papers from top-tier HCI/CSCW venues (e.g., CHI, UIST, CSCW) over the past 20 years. This review yields four primary contributions: (1) an integrated theoretical framework structuring agency patterns, control mechanisms, and interaction contexts, (2) a comprehensive operational catalog of control mechanisms detailing how agency is implemented; (3) an actionable cross-context map linking agency configurations to diverse co-creative practices; and (4) grounded implications and guidance for future CSCW research and the design of co-creative systems, addressing aspects like trust and ethics. Figure 1. The research framework for the scoping review of human-AI co-creation analyzed through the lens of agency. Reflecting on core user-centric design principles, the framework integrates the following dimensions: context (including interaction stage and modality), patterns and distributions of agency (including agency patterns and distribution), control mechanisms (including categories encompassing input, action, output and feedback), application domains, and challenges & directions. Due to space constraints, we abbreviated the control mechanisms. 1. Introduction Artificial Intelligence (AI) are increasingly transitioning from tools performing static tasks ( Eulerich et al. 2023 ) to integral partners in collaborative processes like co-creation ( Zhang et al. 2023a ) and human-supervised autonomous systems ( Zhang et al. 2025 ) . This evolution marks a significant shift for Computer Supported Cooperative Work (CSCW), moving beyond traditional Human-Computer Interaction (HCI) paradigms ( Fitria 2023 ) to encompass scenarios where AI actively collaborates ( Wang et al. 2024a ; Pavlova and Kuzmin 2024 ; Fan et al. 2024 ) . As AI develops into a collaborative partner capable of proactive and potentially autonomous action ( Wu et al. 2021 ; Moruzzi and Margarido 2024 ; Fu and Zhou 2020 ) , it introduces fundamental challenges to the CSCW community in terms of the distribution and negotiation of agency (e.g. initiative, control), the operational mechanisms that enable this distribution , and how these factors perform in various co-creative contexts and application domains ( Miernicki and Ng 2021 ; Kahveci 2023 ) . The concept of agency provides a critical lens for addressing the aforementioned complexities. Agency is understood here as the ability to act intentionally and exert control within a specific context ( Pacherie 2007 ; Limerick et al. 2014 ; Ågerfalk 2020 ) , encompassing both human intentional action and the apparent agency perceived in machines. Although other valuable frameworks detail collaborative workflows ( Gao et al. 2024 ; Moruzzi and Margarido 2024 ) or communication patterns ( Rezwana and Maher 2022 ) , they often underexplored AI’s proactive capabilities. Consequently, they may not inherently provide the tool to systematically investigate the dynamic patterns of initiative or the concrete operational mechanisms through which control is enacted. The agency perspective is crucial, as it directly addresses these aspects, enabling a structured analysis of how collaboration unfolds in terms of distributed action capabilities within various co-creative contexts. Despite its importance, research applying the agency lens to human-AI co-creation is fragmented. Existing conceptualizations ( Wu et al. 2021 ; Fu and Zhou 2020 ) and frameworks ( Zhang et al. 2023a ; Moruzzi and Margarido 2024 ; Zhou et al. 2024 ) often lack granular detail on the specific operational control mechanisms needed to guide the design of practical systems. Although empirical studies ( He et al. 2023 ; Guo et al. 2023 ; Gao et al. 2024 ) offer valuable situated insights, their focus on individual contexts limits the cross-context synthesis required to identify overarching patterns and design principles applicable across the co-creative landscape. Furthermore, investigations into mediating factors (e.g., interaction design ( Rezwana and Maher 2023 ) , explainability ( Zhu et al. 2018 ) ) typically occur in isolation, limiting the structured mapping of how agency patterns, control mechanism, and contextual factors interrelate. This reveals a critical gap: the lack of a comprehensive and systematic overview mapping the interplay between agency patterns, the control mechanisms implementing them, and the co-creative contexts (modalities, domains) where they are employed. To address this critical gap–particularly the need for a systematic synthesis of agency patterns and the operational control mechanisms implementing them across diverse contexts, this paper presents a scoping review. We focus specifically on research published in prominent HCI venues (e.g., CHI, UIST, CSCW) over the past 20 years, as this body of literature provides a concentrated documentation of the interaction designs and implemented control strategies relevant to operationalizing agency in co-creation systems. Our analysis covers 134 papers, employing the methodology detailed in Section 3 and proposing a novel framework (Figure 1 ) to structure our investigation. The structure of the framework is informed by considerations of both system-level organization (integrating context, control, and application) ( Hu et al. 2025 ) and user-centric design principles ( Card 2018 ) . Applying our analytical framework, the analysis yields specific insights into how the agency is operationalized in practice. We first elaborate on the context dimension (Section 4), identifying key situational factors from the reviewed literature. We then identify and classify the principal forms of AI agency currently employed in co-creative systems, revealing a spectrum from passive tools to cooperative partners (Section 5). We systematically catalog the concrete control mechanisms documented across interaction stages (input, action, output, feedback) used to manage agency distribution (Section 6). Furthermore, our synthesis reveals recurring associations linking specific control strategies (e.g. transparency, iterative feedback loops) and agency patterns with particular application domains and reported collaboration dynamics (Section 7). These findings directly inform our core contributions: the integrated analytical framework itself, the comprehensive catalog of operational control mechanisms, an actionable cross-context map of current practices, and grounded implications for the CSCW community (Sections 8-9). 2. Related Work Agency is a foundational concept that is studied in diverse fields. Psychology clarifies its cognitive origins and perception mechanisms ( Wegner 2003 ; Synofzik et al. 2008 ) , sociology addresses complex questions of intentionality and responsibility ( List and Pettit 2011 ; Bratman 2013 ) , while AI examines emerging machine capabilities and their perceptual effects ( Sharma et al. 2024b ; Dai 2024 ) . These diverse inquiries collectively shape our understanding of human action, machine behavior, and their societal implications. Based on this broad context, our review adopts an HCI/CSCW lens. Here, AI’s growing capabilities drive human-AI partnerships, bringing agency, control, and autonomy into the CSCW community’s focus ( Adenuga and Dodge 2024 ; Cheon et al. 2021 ; Liao et al. 2019 ) . We synthesize literature on agency allocation, the distinction between agency and control, and examine co-creation frameworks relevant to this scope. 2.1. Agency and Its Distribution in Human-AI Collaboration Agency, central to HCI ( Shneiderman and Plaisant 2010 ) , yet often ambiguously defined

Read the complete text