Trend AnalysisArts & Design
AI-Generated Art and Authorship Debates: Who Creates When Machines Paint?
Generative AI now produces visual art that wins competitions and sells at auctionโbut legal systems worldwide still cannot answer who the author is. Recent research reveals how artists, technologists, and the public diverge sharply on authorship attribution.
By Sean K.S. Shin
This blog summarizes research trends based on published paper abstracts. Specific numbers or findings may contain inaccuracies. For scholarly rigor, always consult the original papers cited in each post.
Why It Matters
When Jason Allen's AI-generated artwork won the Colorado State Fair's fine arts competition in 2022, the outcry was immediateโbut the legal questions remain unresolved three years later. Generative AI systems such as Midjourney, DALL-E, and Stable Diffusion can now produce images that are visually indistinguishable from human-made art, yet copyright law in most jurisdictions requires a human author. This gap between technological capability and legal framework is not merely academic: it affects how millions of dollars in art market transactions are structured, how artists protect their livelihoods, and how training data rights are negotiated.
The authorship debate also reveals deeper tensions about what creativity means. If an artist writes a detailed prompt, selects from hundreds of outputs, and refines the result through iterative human-AI dialogue, is the resulting work "authored" by the human, the machine, or neither? Recent research shows that answers depend heavily on who you askโand what their stake in the outcome is.
The Science / The Practice
Divergent Stakeholder Perceptions
Singh (2025) directly tests whether artists and technologists evaluate AI-generated art differently across five dimensions: authorship, ownership, ethics, public perception, and market impact. The findings reveal a significant professional divide. Artists consistently attribute less creative agency to AI systems and are more protective of human authorship claims, while technologists view AI as a legitimate co-creative partner. This divergence has practical consequences: if artists and engineers cannot agree on what authorship means, industry standards and legal frameworks will remain contested.
Egocentric Bias in Copyright Judgments
Lima et al. (2024), with 10 citations, provide perhaps the most striking finding. In a study of 432 participants, they demonstrate that laypeople's copyright judgments about AI art are shaped by egocentric bias: people are more likely to grant copyright protection to AI-generated works when they imagine themselves as the creator rather than as a competing artist. This suggests that public opinion on AI authorship is less about principled legal reasoning and more about perceived self-interestโa finding with implications for how copyright legislation will be shaped by democratic processes.
The Creativity Spectrum
Deckker and Sumanasekara (2025), with 6 citations, provide a comprehensive review positioning AI not as a replacement for human creativity but as a force that "expands authorship constraints." Their analysis spans visual art, music, literature, and performance, arguing that AI-generated content creates a creativity spectrum rather than a binary human/machine divide. The implication is that legal and ethical frameworks need gradient modelsโdegrees of human involvementโrather than the all-or-nothing authorship models that current copyright law provides.
Toward Royalty Frameworks
Rai et al. (2025) propose a concrete solution: a royalty framework for copyright protection and accountability in AI-generated art. The framework addresses two problems simultaneouslyโcompensating artists whose work was used to train AI models, and establishing clear ownership chains for AI-generated outputs. This represents a shift from abstract ethical debate to actionable economic mechanisms.
Authorship Attribution Models Compared
<
| Model | Who Owns It | Strength | Weakness |
|---|
| Human-only (current law) | Prompt writer or no one | Legal clarity | Ignores machine contribution |
| Joint authorship | Human + AI developer | Reflects collaboration | AI cannot hold legal rights |
| Work-for-hire | AI company | Simple licensing | Devalues human creative input |
| Gradient model | Proportional to contribution | Nuanced, fair | Difficult to measure contributions |
| Royalty framework (Rai et al.) | Shared via royalty chain | Compensates training data artists | Complex implementation |
What To Watch
The next two years will be decisive. The EU AI Act's provisions on generative AI, ongoing US Copyright Office rulings, and emerging case law in China and the UK will collectively establish the first generation of AI authorship standards. Researchers should watch for convergence or divergence across jurisdictionsโbecause if copyright frameworks fragment internationally, the global digital art market will face significant friction. The gradient model proposed by Deckker and Sumanasekara (2025) may prove prophetic if regulators move toward contribution-proportional rights rather than binary authorship.
Explore related work through ORAA ResearchBrain.
์ ์ค์ํ๊ฐ
2022๋
Jason Allen์ AI ์์ฑ ์ํ์ด ์ฝ๋ก๋ผ๋ ์ฃผ๋ฆฝ ๋ฐ๋ํ ๋ฏธ์ ๋ถ๋ฌธ์์ ์์ํ์ ๋ ์ฆ๊ฐ์ ์ธ ๋ฐ๋ฐ์ด ์ผ์์ง๋ง, ๋ฒ์ ์์ ์ 3๋
์ด ์ง๋ ์ง๊ธ๋ ํด๊ฒฐ๋์ง ์์ ์ฑ๋ก ๋จ์ ์๋ค. Midjourney, DALL-E, Stable Diffusion๊ณผ ๊ฐ์ ์์ฑํ AI ์์คํ
์ ์ด์ ์ธ๊ฐ์ด ๋ง๋ ์์ ์ํ๊ณผ ์๊ฐ์ ์ผ๋ก ๊ตฌ๋ณํ๊ธฐ ์ด๋ ค์ด ์ด๋ฏธ์ง๋ฅผ ์์ฐํ ์ ์์ง๋ง, ๋๋ถ๋ถ์ ๊ตญ๊ฐ ์ ์๊ถ๋ฒ์ ์ธ๊ฐ ์ ์์๋ฅผ ์๊ฑด์ผ๋ก ํ๋ค. ์ด๋ฌํ ๊ธฐ์ ์ ์ญ๋๊ณผ ๋ฒ์ ํ๋ ์์ํฌ ์ฌ์ด์ ๊ฐ๊ทน์ ๋จ์ํ ํ๋ฌธ์ ๋ฌธ์ ๊ฐ ์๋๋ค. ์ด๋ ์๋ฐฑ๋ง ๋ฌ๋ฌ ๊ท๋ชจ์ ๋ฏธ์ ์์ฅ ๊ฑฐ๋ ๊ตฌ์กฐ, ์์ ๊ฐ๋ค์ ์๊ณ ๋ณดํธ ๋ฐฉ์, ํ์ต ๋ฐ์ดํฐ ๊ถ๋ฆฌ ํ์ ๋ฐฉ์ ๋ชจ๋์ ์ํฅ์ ๋ฏธ์น๋ค.
์ ์๊ถ ๋
ผ์์ ์ฐฝ์์ฑ์ด๋ ๋ฌด์์ธ๊ฐ์ ๋ํ ๋ณด๋ค ๊ทผ๋ณธ์ ์ธ ๊ธด์ฅ์ ๋๋ฌ๋ด๊ธฐ๋ ํ๋ค. ์์ ๊ฐ๊ฐ ์์ธํ ํ๋กฌํํธ๋ฅผ ์์ฑํ๊ณ , ์๋ฐฑ ๊ฐ์ ์ถ๋ ฅ๋ฌผ ์ค์์ ์ ํํ๋ฉฐ, ์ธ๊ฐ๊ณผ AI์ ๋ฐ๋ณต์ ์ธ ๋ํ๋ฅผ ํตํด ๊ฒฐ๊ณผ๋ฌผ์ ๋ค๋ฌ๋๋ค๋ฉด, ๊ทธ ์ํ์ ์ ์์๋ ์ธ๊ฐ์ธ๊ฐ, ๊ธฐ๊ณ์ธ๊ฐ, ์๋๋ฉด ๋ ๋ค ์๋๊ฐ? ์ต๊ทผ ์ฐ๊ตฌ๋ค์ ์ด ์ง๋ฌธ์ ๋ํ ๋ต์ด ๋ฌป๋ ๋์์ด ๋๊ตฌ์ธ์ง, ๊ทธ๋ฆฌ๊ณ ๊ทธ๊ฐ ๊ฒฐ๊ณผ์ ์ด๋ค ์ดํด๊ด๊ณ๋ฅผ ๊ฐ์ง๊ณ ์๋์ง์ ๋ฐ๋ผ ํฌ๊ฒ ๋ฌ๋ผ์ง๋ค๋ ๊ฒ์ ๋ณด์ฌ์ค๋ค.
์ฐ๊ตฌ ๋ด์ฉ
์ดํด๊ด๊ณ์ ์ธ์์ ๋ถํ
Singh (2025)๋ ์์ ๊ฐ์ ๊ธฐ์ ์๊ฐ AI ์์ฑ ์์ ์ ์ ์๊ถ, ์์ ๊ถ, ์ค๋ฆฌ, ๋์ค ์ธ์, ์์ฅ ์ํฅ์ด๋ผ๋ ๋ค์ฏ ๊ฐ์ง ์ฐจ์์์ ์๋ก ๋ค๋ฅด๊ฒ ํ๊ฐํ๋์ง๋ฅผ ์ง์ ๊ฒ์ฆํ๋ค. ์ฐ๊ตฌ ๊ฒฐ๊ณผ๋ ๋๋ ทํ ์ง์
์ ์ธ์ ์ฐจ์ด๋ฅผ ๋๋ฌ๋ธ๋ค. ์์ ๊ฐ๋ค์ AI ์์คํ
์ ๋ํ ์ฐฝ์์ ์ฃผ์ฒด์ฑ์ ์ผ๊ด๋๊ฒ ๋ฎ๊ฒ ํ๊ฐํ๊ณ ์ธ๊ฐ ์ ์๊ถ ์ฃผ์ฅ ๋ณดํธ์ ๋ ์ ๊ทน์ ์ธ ๋ฐ๋ฉด, ๊ธฐ์ ์๋ค์ AI๋ฅผ ์ ๋นํ ๊ณต๋ ์ฐฝ์ ํํธ๋๋ก ๋ฐ๋ผ๋ณธ๋ค. ์ด๋ฌํ ์ธ์ ์ฐจ์ด๋ ์ค์ง์ ์ธ ๊ฒฐ๊ณผ๋ก ์ด์ด์ง๋ค. ์์ ๊ฐ์ ์์ง๋์ด๊ฐ ์ ์๊ถ์ ์๋ฏธ์ ํฉ์ํ์ง ๋ชปํ๋ค๋ฉด, ์ฐ์
ํ์ค๊ณผ ๋ฒ์ ํ๋ ์์ํฌ๋ ๊ณ์ํด์ ๋
ผ์์ ๋์์ผ๋ก ๋จ์ ๊ฒ์ด๋ค.
์ ์๊ถ ํ๋จ์์์ ์๊ธฐ์ค์ฌ ํธํฅ
Lima et al. (2024)์ 10ํ ์ธ์ฉ์ผ๋ก, ๊ฐ์ฅ ์ฃผ๋ชฉํ ๋งํ ์ฐ๊ตฌ ๊ฒฐ๊ณผ๋ฅผ ์ ์ํ๋ค. 432๋ช
์ ์ฐธ๊ฐ์๋ฅผ ๋์์ผ๋ก ํ ์ฐ๊ตฌ์์ ์ผ๋ฐ์ธ์ AI ์์ ์ ์๊ถ ํ๋จ์ด ์๊ธฐ์ค์ฌ ํธํฅ(egocentric bias)์ ์ํด ํ์ฑ๋๋ค๋ ๊ฒ์ ๋ฐํ๋ค. ์ฆ, ์ฌ๋๋ค์ ์ค์ค๋ก๋ฅผ ๊ฒฝ์ ์์ ๊ฐ๊ฐ ์๋ ์ฐฝ์์๋ก ์์ํ ๋ AI ์์ฑ ์ํ์ ์ ์๊ถ ๋ณดํธ๋ฅผ ๋ถ์ฌํ ๊ฐ๋ฅ์ฑ์ด ๋ ๋๋ค. ์ด๋ AI ์ ์๊ถ์ ๋ํ ์ฌ๋ก ์ด ์์น์ ๊ธฐ๋ฐํ ๋ฒ์ ์ถ๋ก ๋ณด๋ค๋ ์ธ์๋ ์๊ธฐ ์ด์ต์ ๋ ๊ฐ๊น๋ค๋ ๊ฒ์ ์์ฌํ๋ฉฐ, ๋ฏผ์ฃผ์ ๊ณผ์ ์ ํตํด ์ ์๊ถ ๋ฒ์ ๊ฐ ์ด๋ป๊ฒ ํ์ฑ๋ ์ง์ ๋ํ ํจ์๋ฅผ ์ง๋๋ค.
์ฐฝ์์ฑ ์คํํธ๋ผ
Deckker and Sumanasekara (2025)๋ 6ํ ์ธ์ฉ์ผ๋ก, AI๋ฅผ ์ธ๊ฐ ์ฐฝ์์ฑ์ ๋์ฒด์ฌ๊ฐ ์๋๋ผ "์ ์๊ถ ์ ์ฝ์ ํ์ฅํ๋" ํ์ผ๋ก ์์น์ํค๋ ํฌ๊ด์ ์ธ ๊ฒํ ๋ฅผ ์ ๊ณตํ๋ค. ์ด๋ค์ ๋ถ์์ ์๊ฐ ์์ , ์์
, ๋ฌธํ, ๊ณต์ฐ ์์ ์ ๊ฑธ์ณ ์์ผ๋ฉฐ, AI ์์ฑ ์ฝํ
์ธ ๊ฐ ์ธ๊ฐ/๊ธฐ๊ณ์ ์ด๋ถ๋ฒ์ ๊ตฌ๋ถ์ด ์๋ ์ฐฝ์์ฑ ์คํํธ๋ผ์ ํ์ฑํ๋ค๊ณ ์ฃผ์ฅํ๋ค. ์ด๋ ๋ฒ์ ยท์ค๋ฆฌ์ ํ๋ ์์ํฌ๊ฐ ํํ ์ ์๊ถ๋ฒ์ด ์ ๊ณตํ๋ ์ ๋ถ ์๋๋ฉด ์ ๋ฌด ๋ฐฉ์์ ์ ์๊ถ ๋ชจ๋ธ์ด ์๋๋ผ, ์ธ๊ฐ ๊ด์ฌ์ ์ ๋๋ฅผ ๋ฐ์ํ๋ ๋จ๊ณ์ ๋ชจ๋ธ์ ํ์๋ก ํ๋ค๋ ์ ์ ์์ฌํ๋ค.
๋ก์ดํฐ ํ๋ ์์ํฌ๋ฅผ ํฅํ์ฌ
Rai et al. (2025)๋ AI ์์ฑ ์์ ์ ์ ์๊ถ ๋ณดํธ ๋ฐ ์ฑ
์์ ์ํ ๋ก์ดํฐ ํ๋ ์์ํฌ๋ผ๋ ๊ตฌ์ฒด์ ์ธ ํด๊ฒฐ์ฑ
์ ์ ์ํ๋ค. ์ด ํ๋ ์์ํฌ๋ ๋ ๊ฐ์ง ๋ฌธ์ ๋ฅผ ๋์์ ๋ค๋ฃฌ๋ค. AI ๋ชจ๋ธ ํ์ต์ ์ํ์ด ์ฌ์ฉ๋ ์์ ๊ฐ์ ๋ํ ๋ณด์๊ณผ, AI ์์ฑ ๊ฒฐ๊ณผ๋ฌผ์ ๋ํ ๋ช
ํํ ์์ ๊ถ ์ฒด๊ณ ํ๋ฆฝ์ด ๊ทธ๊ฒ์ด๋ค. ์ด๋ ์ถ์์ ์ธ ์ค๋ฆฌ์ ๋
ผ์์์ ์คํ ๊ฐ๋ฅํ ๊ฒฝ์ ์ ๋ฉ์ปค๋์ฆ์ผ๋ก์ ์ ํ์ ์๋ฏธํ๋ค.
์ ์๊ถ ๊ท์ ๋ชจ๋ธ ๋น๊ต
<
| ๋ชจ๋ธ | ์์ ์ฃผ์ฒด | ๊ฐ์ | ์ฝ์ |
|---|
| ์ธ๊ฐ ๋จ๋
(ํํ๋ฒ) | ํ๋กฌํํธ ์์ฑ์ ๋๋ ์์ | ๋ฒ์ ๋ช
ํ์ฑ | ๊ธฐ๊ณ์ ๊ธฐ์ฌ ๋ฌด์ |
| ๊ณต๋ ์ ์ | ์ธ๊ฐ + AI ๊ฐ๋ฐ์ฌ | ํ์
๋ฐ์ | AI๋ ๋ฒ์ ๊ถ๋ฆฌ ๋ณด์ ๋ถ๊ฐ |
| ์
๋ฌด์ ์ ์๋ฌผ | AI ๊ธฐ์
| ๋จ์ํ ๋ผ์ด์ ์ฑ | ์ธ๊ฐ์ ์ฐฝ์์ ๊ธฐ์ฌ ๊ฐ์น ํ๋ฝ |
| ๋จ๊ณ์ ๋ชจ๋ธ | ๊ธฐ์ฌ๋์ ๋น๋ก | ์ธ๋ฐํ๊ณ ๊ณต์ ํจ | ๊ธฐ์ฌ๋ ์ธก์ ์ด๋ ค์ |
| ๋ก์ดํฐ ํ๋ ์์ํฌ (Rai et al.) | ๋ก์ดํฐ ์ฒด๊ณ๋ฅผ ํตํ ๊ณต์ | ํ์ต ๋ฐ์ดํฐ ์์ ๊ฐ ๋ณด์ | ๋ณต์กํ ๊ตฌํ |
ํฅํ ์ฃผ๋ชฉํ ์ฌํญ
ํฅํ 2๋
์ด ๊ฒฐ์ ์ ์ธ ์๊ธฐ๊ฐ ๋ ๊ฒ์ด๋ค. ์์ฑํ AI์ ๊ดํ EU AI ๋ฒ์ ์กฐํญ, ๋ฏธ๊ตญ ์ ์๊ถ์ฒญ์ ์ง์์ ์ธ ํ๊ฒฐ, ์ค๊ตญ๊ณผ ์๊ตญ์์ ํ์ฑ๋๊ณ ์๋ ํ๋ก๋ฒ์ด ํฉ์ณ์ ธ 1์ธ๋ AI ์ ์๊ถ ๊ธฐ์ค์ ํ๋ฆฝํ ๊ฒ์ด๋ค. ์ฐ๊ตฌ์๋ค์ ๊ฐ๊ตญ ๋ฒ์ ๊ฐ์ ์๋ ด ๋๋ ๋ถํ ์ฌ๋ถ๋ฅผ ์ฃผ์ํด์ผ ํ๋ค. ์ ์๊ถ ํ๋ ์์ํฌ๊ฐ ๊ตญ์ ์ ์ผ๋ก ๋ถ์ด๋ ๊ฒฝ์ฐ ๊ธ๋ก๋ฒ ๋์งํธ ๋ฏธ์ ์์ฅ์ ์ฌ๊ฐํ ๋ง์ฐฐ์ ์ง๋ฉดํ๊ฒ ๋ ๊ฒ์ด๊ธฐ ๋๋ฌธ์ด๋ค. Deckker and Sumanasekara (2025)๊ฐ ์ ์ํ ๋จ๊ณ์ ๋ชจ๋ธ์ ๊ท์ ๋น๊ตญ์ด ์ด๋ถ๋ฒ์ ์ ์๊ถ ๋์ ๊ธฐ์ฌ ๋น๋ก ๊ถ๋ฆฌ ๋ฐฉํฅ์ผ๋ก ๋์๊ฐ๋ค๋ฉด ์ ๊ฒฌ์ง๋ช
์ด ์๋ ๊ฒ์ผ๋ก ํ๋ช
๋ ์ ์๋ค.
๊ด๋ จ ์ฐ๊ตฌ๋ ORAA ResearchBrain์์ ํ์ํ ์ ์๋ค.
References (4)
[1] Singh, A. (2025). Creativity reimagined: ethical authorship, ownership and social impact of AI-generated art. Journal of Information, Communication and Ethics in Society.
[2] Lima, G., Grgic-Hlaca, N., & Redmiles, E. M. (2024). Public Opinions About Copyright for AI-Generated Art: The Role of Egocentricity, Competition, and Experience. Proceedings of CHI 2024.
[3] Deckker, D., & Sumanasekara, S. (2025). A Review of AI-Powered Creativity: The Intersection of AI and the Arts. EPRA International Journal.
[4] Rai, P., Gupta, K., & Gabralla, L. A. (2025). A royalty framework for copyright protection and accountability in AI-generated art. Information Retrieval Journal.