PaperlayerView exact revision on arXiv

Source-linked AI summary

Where Does the Human End? Creative Agency with Generative AI across Five Years of Chinese Digital Painting

Yibo Meng, Ruiqi Chen, Shuheng Cao, Weijia Zhang, Chengxi Zang

arXiv:2609.09333v1cs.HC

TL;DR

Creative practitioners must decide where AI assistance ends and human authorship begins, yet much prior evidence is cross-sectional or short-term. This five-year interview study follows 17 Chinese digital painters as AI becomes more publicly visible and professionally relevant. It finds recurring but non-uniform patterns of resistance, delegation, and, for some, agency repartitioning.

  • Problem

    Prior research often captures artist-AI relationships at one moment, leaving open how creators revisit delegated tasks and preserved boundaries as AI becomes more publicly visible and economically embedded.

    Source: §1 · p. 1

  • Method

    The study follows 17 Chinese digital painters through annual semi-structured interviews from 2021 to 2025.

    Source: §1 · p. 2

  • Results

    Participants described protective resistance, pragmatic task delegation, and, for some, reflective agency repartitioning, shaped by peer norms and emotional climates.

    Source: §1 · p. 2

  • Takeaways & Limitations

    Longitudinal agency partitioning offers a lens for tracing how creative agency is allocated, revisited, and contested through workflow, authorship, peer circulation, and labor value.

    Source: §1 · p. 2, §6 · p. 8

    The findings trace patterns within a modest, self-selecting cohort of 17 baseline participants, including 13 five-wave records, in a context shaped by Chinese digital painting.

    Source: §5.5 · p. 8

Abstract

from arXiv · show

As generative AI enters creative work, practitioners must decide where AI assistance ends and human authorship begins. Human-agent interaction (HAI) research has examined AI as a tool, collaborator, consultant, and competitor. The longitudinal problem is how these roles are revised as systems become more capable, public, and economically embedded. We report a five-year interview study with 17 Chinese digital painters, based on annual semi-structured interviews from 2021 to 2025. Participants described recurring but non-uniform patterns of protective resistance, pragmatic task delegation, and, for some, reflective agency repartitioning. Early resistance protected observation, originality, signature, and ownership from AI. Later delegation placed AI in bounded tasks such as references, backgrounds, rough sketches, and client-facing drafts. By 2025, some participants built hybrid workflows around human-only zones, while others described fatigue, precarity, or difficulty locating a remaining human role. Peer norms, emotional climates, and production pressures shaped which delegations felt useful, acceptable, or exhausting. Copyright, authorship, and creative labor remained recurring limits on what participants were willing to delegate. We frame these accounts as longitudinal agency partitioning, the situated work of deciding which stages, responsibilities, values, and claims remain human in creative human-agent interaction. We discuss design implications for revisable agency-boundary controls, provenance scaffolds, and community-facing authorship norms.

1 Introduction

Generative AI raises questions about how creators allocate creative agency, authorship, and responsibility as systems become more embedded in work. This study traces how Chinese digital painters revisit those boundaries over five years.

1 Introduction

  • Prior evidence is largely cross-sectional, short-term, or based on prototype evaluations, leaving open how creators revisit judgments as AI becomes more public and economically embedded.The study asks what artists delegate, preserve, and renegotiate across multiple years.Source: §1 · p. 1
  • Longitudinal agency partitioning describes the repeated allocation and revision of tasks, responsibilities, values, and claims between humans and AI.The framing connects workflow choices with authorship, labor, and peer legitimacy.Source: §1 · p. 1
  • The five-year study follows 17 Chinese digital painters through annual semi-structured interviews from 2021 to 2025.It examines changing relationships with generative AI and the role of peer communities in shaping delegation.Source: §1 · p. 2
  • The study identifies protective resistance, pragmatic task delegation, and, for some painters, reflective agency repartitioning.Peer norms and emotional climates shaped whether AI use felt useful, acceptable, or exhausting, while copyright, authorship, and creative labor remained limits on delegation.Source: §1 · p. 2
  • The paper contributes a longitudinal account of creative AI experience, a temporal agency-partitioning lens, and design directions for revisable boundaries, provenance, and community norms.The account is grounded in one cohort and context.Source: §1 · p. 2

2 Related Work

Prior work documents creative AI’s opportunities and ethical tensions, and examines assigned AI roles and identity pressures. This paper distinguishes longitudinal agency partitioning by tracing how permitted tasks and claims are revisited over time.

2 Related Work

  • Prior work links creative AI interaction choices to authorship, labor, professional identity, and attributed system roles.These studies connect artist adoption and resistance with roles such as tool, collaborator, consultant, and competitor.Source: §2 · p. 2

2.1 Creative Practice, Resistance, and Ethical Boundaries

  • Creative AI can support prototyping, ideation, and stylistic experimentation, while artists also report resistance, labor concerns, and disputes over consent, credit, and compensation.Efficiency gains may pressure compensation, and market pressure can lead some artists to adopt AI despite opposition.Source: §2.1 · p. 2

2 Related Work

  • Existing studies capture opportunities and risks at particular moments, while this study follows practitioners as tools and norms change.Authorship, labor, and copyright remain boundaries around new practices.Source: §2.1 · p. 2

2.2 Agency Attribution in Creative Human-Agent Collaboration

  • Prior studies describe AI as an advisor, collaborator, or driver, with stronger attributed agency sharpening tensions around intent and authorship.Advisory uses can support human decision-making while preserving human authorship.Source: §2.2 · p. 2

2 Related Work

  • Longitudinal agency partitioning tracks how AI’s permitted tasks, responsibilities, values, and claims change with social expectations and production pressures.It connects workflow and claim boundaries with the repeated work of maintaining or revising professional identity.Source: §2.2 · p. 2, §2.3 · p. 3

2.3 Identity Work in Technology Adoption

  • Identity work concerns how professionals construct, maintain, and revise occupational selves amid shifts in competence, authenticity, and creative labor.Creative AI research links identity pressure to everyday delegation choices and tensions between market relevance and artistic integrity.Source: §2.3 · p. 3

3 Methodology

The study uses annual interviews and longitudinal qualitative analysis to examine changing AI practices among Chinese digital painters from 2021 to 2025. It preserves each year’s account and distinguishes observed changes from missing records.

3 Methodology

  • The five-year longitudinal design treats participants’ accounts as situated interpretations and uses a stable interview guide to compare changing understandings of AI, creativity, and agency.The design preserves the difference between accounts recorded at different moments.Source: §3 · p. 3

3.1 Participants

  • The 2021 baseline enrolled 17 Chinese digital painters, including 11 professionals and 6 hobbyists, and Table 1 summarizes their baseline demographics.The study defines digital painters as artists primarily creating two-dimensional visual works with digital tools.Source: §3.1 · p. 3, Table 1 · p. 3
  • Available records numbered 17, 15, 15, 14, and 14 from 2021 to 2025, with 13 participants contributing all five waves.Claims about change are limited to participants and years with available records.Source: §3.1 · p. 3

3.2 Procedure

  • Two researchers jointly conducted annual 60–90-minute online interviews, using the same four-block guide across all five waves.Interviews were recorded in Mandarin, transcribed verbatim, and cross-checked; analytically central quotations were checked against Mandarin originals.Source: §3.2 · p. 3
  • The study treats technological and conceptual drift as part of the phenomenon, distinguishing 2021 tools and discourse from prompt-controllable systems that circulated more broadly after 2022.Participants’ descriptions of tools, stages, and workflows anchored the analysis across those shifts.Source: §3.2 · pp. 3–4

3.3 Data Analysis

  • Thematic analysis used independently coded calibration transcripts, a codebook, team review, and documented decisions, with calibration agreement exceeding 80%.The agreement assessed consistent codebook application, not theme validity.Source: §3.3.1 · p. 4

3.3.2 Longitudinal Case Analysis.

  • Lexical memos tracked shifts in participants’ descriptors, while year-stamped quotations and longitudinal case comparisons remained the main evidence.The memos were contextual support because interview availability, transcript length, and vocabularies changed over time; the 2025 interviews were used to examine reinterpretation and consolidation.Source: §3.3.1 · p. 4, §3.3.2 · p. 4

3.3.1 Codebook-Based Thematic Analysis.

  • Agency coding marked delegated and human-reserved stages, responsibility, authorship, credit, labor value, and peer or client legitimacy while retaining each passage’s year.Longitudinal case files compared reinterpretation and stability without replacing earlier accounts with later ones; missing waves remained missing data.Source: §3.3.1 · p. 4, §3.3.2 · p. 4

4 Findings

The findings trace recurring but non-uniform agency-partitioning patterns across time. Delegation and resistance varied across tasks and participants, with social and practical boundaries shaping what remained human.

4 Findings

  • Protective resistance recurred in 2021–2022, pragmatic task delegation in 2022–2024, and reflective agency repartitioning for some participants by 2025.Divergent paths included reframing, procedural compromise, continued refusal, difficulty locating a human role, and labor-related precarity.Source: §4 · p. 4
  • Participants partitioned agency through delegated stages, human-only authorship claims, final-delivery responsibility, and aesthetic, economic, peer, or client legitimacy boundaries.Peer norms and emotional climates shaped delegation, while copyright and authorship remained recurring limits.Source: §4 · p. 4
  • The three phase labels are analytic anchors, not exclusive participant types; a person could protect one boundary while delegating another in the same year.The phases organize allocations across tasks, claims, and social settings while preserving divergent cases.Source: §4 · p. 4

4.1 Longitudinal Trajectory of Agency Partitioning

Across 2021–2025, participants moved from protective resistance toward bounded AI delegation, but later experiences diverged between deliberate human-AI boundaries and fatigue or precarity.

4.1.1 Protective Resistance from 2021 to 2022.

  • Early resistance defended observation, originality, signature, and ownership as creative capacities participants did not want absorbed into AI.P5’s critique shifted from AI not knowing what an artist wanted to denying that it observes; participants also raised signature, training-data, and livelihood concerns.Source: §4.1.1 · pp. 4–5

4.1.2 Pragmatic Task Delegation from 2022 to 2024.

  • From 2022 to 2024, participants increasingly delegated bounded tasks while retaining human selection, correction, signature style, and delivery responsibility.Uses included references, color schemes, backgrounds, rough sketches, and client-facing drafts; deadlines and team workflows shaped professional adoption.Source: §4.1.2 · p. 5
  • Commercial pipelines encouraged earlier or more frequent delegation among professionals, while hobbyists retained more freedom to remain skeptical or selective through 2024.Source: §4.1.2 · p. 5

4.1.3 Divergent Endpoints and Reflective Agency Repartitioning in 2025.

  • By 2025, some participants deliberately repartitioned work, while others reported fatigue, precarity, continued refusal, or uncertainty about their remaining role.Five of 14 participants with 2025 records reported shrinking salaries, fewer opportunities, or unemployment; six of 14 described monotony from AI’s technical polish.Source: §4.1.3 · pp. 5–6
  • Hybrid workflows made boundaries concrete by reserving steps such as sketching, correction, final details, or a character’s face for human work.P14 described a four-step workflow, while P6 reserved the face as an aesthetic-signature boundary.Source: §4.1.3 · p. 6

4.2 Conditions That Reopened or Preserved Agency Boundaries

Peer dynamics, changing emotional responses, and persistent ethical concerns shaped whether participants accepted, limited, or resisted AI delegation.

4.2.1 From Peer Pressure to Normalization and Polarization.

  • Peer influence shifted from pressure in 2022 toward normalization in 2023–2024, then fragmentation and withdrawal for some participants by 2025.Clients also began treating AI use in concept sketches as a baseline, while four of 14 participants with 2025 records reported community fragmentation.Source: §4.2.1 · p. 6

4.2.2 From Curiosity and Offense to Awe and Fatigue.

  • Emotional responses moved from curiosity and offense to awe and fatigue, and participants linked affect to whether they entered, avoided, or limited AI collaboration.Source: §4.2.2 · p. 6

4.2.3 Ethical Boundaries Persisted as Use Changed.

  • Copyright, data provenance, and authorship concerns persisted as AI use changed, extending from training-data and livelihood worries to client deliverables and hybrid-work credit.By 2025, some participants treated human-only steps as part of their legal and moral claims to a work.Source: §4.2.3 · pp. 6–7

5 Discussion

The discussion frames creative AI use as repeated, socially situated negotiation over where human agency ends, rather than a linear move from rejection to acceptance. It connects this account to design opportunities and limits in the study’s cohort and context.

5 Discussion

  • Across the study, painters described protective resistance, pragmatic delegation, and reflective repartitioning, shaped by peer norms, emotional climates, and boundaries around copyright, authorship, and labor.Whether AI felt useful, acceptable, or exhausting varied across these conditions.Source: §5 · p. 7

5.1 Longitudinal Agency Partitioning

  • Longitudinal agency partitioning captures repeated maintenance and revision of boundaries across six areas, as painters moved from resistance toward bounded delegation and, for some, hybrid workflows.Compensation concerns intensified late as participants described higher output without comparable pay growth; some found repartitioning difficult.Source: §5.1 · p. 7

5.2 Peer and Affective Conditions

  • Peer expectations and emotional climates shaped agency decisions: refusal could risk marginalization, while community saturation could make AI use exhausting and polarization encourage compromise or withdrawal.Anxiety appeared early, excitement and awe in middle waves, and fatigue and ambivalence later.Source: §5.2 · p. 7

5.3 The Chinese Digital-Painting Context

  • In Chinese digital-painting settings, production efficiency, platform-mediated peer circulation, and copyright debates made agency partitioning especially urgent.Industrialized pipelines made efficiency a visible business metric, while platforms circulated styles, copyright news, tutorials, and moral judgments.Source: §5.3 · p. 7

5.4 Design Considerations

  • Designers could make boundaries visible and revisable across workflow stages, layers, versions, posts, deliverables, and authorship claims.The proposed design space also includes provenance and authorship scaffolds and community-facing norms.Source: §5.4 · pp. 7–8

5.5 Limitations and Future Work

  • The modest, self-selecting cohort supports plausible patterns within this group, while the Chinese digital-painting context limits how far findings can be generalized across cultures and domains.Broader samples could examine whether professionals delegate more rapidly than hobbyists.Source: §5.5 · p. 8
  • Retrospective accounts may be reshaped by later experience and repeated interviewing, and the study did not isolate tool affordances from sociocultural identity work.Real-time diaries could provide finer-grained affective records; targeted system evaluations could complement the interviews.Source: §5.5 · p. 8

6 Conclusion

A five-year interview study of 17 Chinese digital painters traces recurring, non-uniform patterns of resistance, delegation, and agency repartitioning. It presents longitudinal agency partitioning as a lens for understanding how creative agency is negotiated over time.

6 Conclusion

  • The study found recurring but non-uniform patterns of protective resistance, pragmatic task delegation, and reflective agency repartitioning among 17 Chinese digital painters.Copyright, authorship, and creative labor remained boundaries on delegation, while peer norms and emotional climates shaped whether AI use felt useful, acceptable, or exhausting.Source: §6 · p. 8
Loading 2609.09333v1…