Man, Machine, and Masterpiece: Artistic Ownership in the AI Era

Technical Summary: Man, Machine, and Masterpiece: Artistic Ownership in the AI Era

Problem Statement
The integration of generative AI into creative workflows has precipitated a crisis regarding artistic ownership and authorship. While legal frameworks (such as EU copyright law) traditionally presume ownership originates from a singular human creator whose creative decisions are traceable to the final work, AI-mediated creation distributes agency across humans, tools, data, and algorithms. Current regulatory discourse relies on qualitative, loosely specified criteria like "significant human contribution," which fail to account for the interwoven nature of human and machine actions in processes involving prompting, iteration, and selection. There is a lack of consensus on how to define ownership when human and AI contributions are intertwined, and a prevailing tension exists between the desire to quantify these contributions and the artistic community's resistance to reducing creative intent to measurable metrics.

Methodology
The study employs a critical design approach centered on a "provotype" (a prototype designed to provoke critical reflection rather than solve a problem) named ArtSplit. The research followed a mixed qualitative methodology:

1. Preliminary Interviews: Semi-structured interviews were conducted with two academic artists (22 years of experience) to establish baseline perceptions of ownership and authorship.
2. Provotype Design: The authors designed ArtSplit, an AI-driven system that explicitly quantifies human and AI contributions across creative stages. Three distinct variants were developed to operationalize ownership differently:
   • Refinement-Oriented: Attributes ownership based on prompting and manual refinement (aligning with EU copyright doctrine regarding visible creative choices).
   • Input-Oriented: Attributes ownership based on prompting and reference image input, without further modification.
   • Input+Refinement-Oriented: Demonstrates how ownership attribution shifts dynamically as the creative process evolves (e.g., adding refinement steps later).
3. Main Study: Five participants (three academic artists and two content creators) were interviewed via semi-structured sessions. They viewed video vignettes depicting the three ArtSplit variants and were asked to reflect on the fairness of the ownership breakdowns, the quantifiability of their work, and the implications for their sense of authorship.
4. Analysis: Data was analyzed using emergent coding followed by thematic analysis, supported by MAXQDA.

Key Findings
The thematic analysis revealed five core themes regarding how artists interpret ownership in the context of quantification:

• Artistic Concept Determines Authorship: Most participants, particularly content creators, argued that ownership is defined by the underlying artistic concept rather than the volume of labor. However, academic artists were more skeptical, noting that "concept" is distinct from "instruction" (prompting) and that the creative process is a continuous feedback loop rather than discrete steps.
• Quantifiability of Ownership: Participants expressed deep skepticism about the feasibility of quantifying ownership. Academic artists argued that random processes (AI generation) and the intangible nature of "concept" make fair quantification impossible. Content creators were more open to quantification but relied on intuition and comparative fairness rather than fixed metrics.
• Appropriation vs. Motivation: Content creators viewed the system as a potential motivational tool to ensure they contributed meaningfully (avoiding "laziness"), whereas academic artists saw it as a tool for narrative critique or tracking copyright-relevant differences (e.g., ensuring a design is distinct enough to avoid infringement).
• Controllability: While some participants felt the system offered a "black box" glimpse that increased control, others warned it could create a false sense of control, potentially narrowing creative actions to only those measurable by the system.
• Perverse Intention: The study identified a risk where the incentive structure of the system could derail the artistic process. Artists might perform unnecessary actions solely to increase their ownership percentage, or conversely, ignore the metrics entirely if the desired image is achieved early, rendering the numbers meaningless.

Key Contributions

• ArtSplit Provotype: The design and evaluation of a system that explicitly operationalizes ownership through quantitative metrics, serving as a catalyst for critical discussion rather than a functional solution.
• Empirical Evidence on Quantification: The study provides empirical data showing that artists, particularly those with deep theoretical backgrounds, reject the premise that ownership can be reduced to measurable contributions. They argue that quantification risks diluting the understanding of creative intent and agency.
• Distinction between Contribution and Authorship: The research highlights a critical disconnect where systems track contribution (labor/actions), but artists define authorship through conceptual framing and vision, which are often unquantifiable.

Significance and Claims
The paper positions itself not as a proposal for better ownership-tracking systems, but as a critique of the impulse to transform historically and socially situated relations into technical problems. The authors argue that:

1. Quantification Fails to Align: Efforts to measure ownership through system-generated metrics fail to align with artists' understandings of creative intent and agency.
2. Risk of Dilution: Attempting to operationalize ownership risks diluting long-standing assumptions about how artists understand and practice creative work.
3. Technosolutionism Critique: The paper challenges the assumption that ownership disputes in AI art are problems to be solved via system design. Instead, it suggests these are complex socio-legal issues that resist technical resolution.
4. Methodological Reflection: The study demonstrates that presenting users with a system that appears to measure the unmeasurable can induce "false beliefs" about the feasibility of such metrics, raising ethical questions about the use of such provotypes in research.

Ultimately, the work suggests that the focus should shift from refining metrics for ownership to reflecting on whether the desire to computationally resolve these questions is appropriate given the nature of creative labor.