As generative AI (GenAI) lowers the barrier to artistic creation, we are forced to confront a messy question: What is this art actually worth, and who gets the check?
A new paper by Obinna et al., titled "What Is the Value of AI-Involved Art? Evidence From a Juried In-Person Exhibition," moves this debate out of the realm of abstract philosophy and into a real-world gallery. By studying how creators and evaluators from different professional backgrounds (Law, CS, and Art) value AI-involved works, the researchers reveal a fundamental disconnect in how we perceive creative labor in the age of the algorithm.
TL;DR
This study uses a juried, in-person exhibition to show that creators value AI art based on their actual "sweat of the brow" (effort), while evaluators price it based on perceived artistic merit. It also highlights a "disciplinary divide": legal training makes people more likely to reward training-data artists, while computer science training shifts credit toward the AI model itself.
What Problem Does This Paper Address?
Traditionally, the value of art is tied to visible traces of labor—the brushstroke, the chisel mark, the hours spent at a canvas. GenAI disrupts this because the final output often appears "instantaneous," hiding the iterative prompting, curation, and refinement that a human creator actually performed.
This creates an invisibility of creative labor. If an evaluator can’t see the effort, they may systematically undervalue the work. Furthermore, as policy debates rage over copyright and compensation, we don't yet understand how different stakeholders (like lawyers vs. engineers) distribute rights and rewards among the user, the AI developer, and the artists whose data trained the model.
Methods
The researchers conducted a unique "embodied" study involving 209 participants (105 creators and 104 peer evaluators) from a university community.
- The Creation Phase: Creators used a custom web-based platform (integrating DALL-E 3, GPT, and Claude) to generate art. The system logged every prompt, iteration, and minute spent. Creators could enter "AI-only" or "Mixed Media" categories.
- The Evaluation Phase: Peer evaluators (blind to the creation process) and expert jurors rated the works on creativity, technical quality, and visual appeal.
- The Exhibition: The top 12 works were displayed in a physical gallery, adding real-world stakes to the participants' judgments.
Datasets
The "dataset" in this study consists of:
- 105 AI-involved artworks generated by participants.
- Behavioral logs of the creation process (prompt word counts, session duration, number of iterations).
- Survey responses from 209 participants regarding valuation, rights, and rewards.
Key Insights and Ideas
- The Effort-Merit Gap: Creators price their work based on how hard they actually worked (e.g., total prompt word count). Evaluators, however, price work based on how creative or beautiful they think it is. Because evaluators can't see the "invisible labor" of prompting, their valuations often miss the mark.
- The "Sweat of the Brow" Paradox: Creators frequently invoked the "sweat of the brow" doctrine—the idea that labor deserves rights—even though this doctrine has been largely rejected by modern U.S. copyright law in favor of "originality."
- Disciplinary Silos: Your major matters.
- Law students were significantly more likely to support "distributed reward models" that include the artists whose work was used for training.
- CS students were less likely to want to compensate the human creator, often attributing more creative agency to the AI model itself.
- Training Data Artists are "Left Behind": Across the board, training data artists received the least support for rewards and rights, suggesting a major hurdle for those advocating for artist compensation in GenAI pipelines.
Key Results
- Willingness to Pay/Sell: 71.2% of creators were willing to sell their work, but 54% of evaluators stated they would not buy any of the art they saw.
- Pricing Correlations: For creators, asking prices were positively associated with observed effort ($B = 0.64, p = .008$). For evaluators, purchase prices correlated with perceived creativity and visual appeal.
- Reward Allocation: 62.9% of creators wanted to keep 100% of the rewards for themselves. In contrast, evaluators were more distributive: nearly 20% included training data artists and 38% included AI developers in the reward pool.
Limitations and Future Work
The study’s population was primarily university students, which may not represent professional artists or the general public. Additionally, the pricing was hypothetical; participants weren't spending their own "real" money.
The authors suggest that future AI systems should include provenance standards (like "model cards" for art) that document the human creative process. By making the "invisible labor" of prompting and curation visible, we might enable fairer market valuations and provide the evidentiary basis that legal frameworks increasingly require for copyright protection.
Summary updated