In early 2025 auction house Christie’s in New York City sold an unusual collection of art pieces. Surreal portraits, photorealistic images and cartoon-inspired creations, all generated by artificial intelligence, raked in more than $700,000, beating sale estimates. The first-of-its-kind event also sparked a backlash. More than 6,000 artists protested that the AI models used to create these works had been trained on copyrighted images without creator consent. Although the auction house had argued that the works demonstrated “human agency in the age of AI,” critics saw the event as an example of an industry rushing to commercialize technology built on uncompensated creative labor.
Other artistic and professional communities have also been worried. A report released last November found that more than half of novelists surveyed in the U.K. thought AI could end their career. Many authors believed that their work had already been used without consent to train large language models and that AI was flooding the market with low-quality prose. Audiences seem to have complicated feelings about the technology, too. As one survey found, many Americans are okay with AI as a tool for creative professionals but not as a replacement for their work.
A viewer’s comfort with AI art, however, may depend on how much they know about the way it was made. I study neuroaesthetics, a field that combines neuroscience, psychology, and our perception of beauty and art. My colleagues and I have found that the more people learn about how AI’s back end works—the datasets, training process and prompting—the less comfortable they are with the moral considerations surrounding these creations and the value of AI-generated pieces.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
I became curious about AI because its rapid proliferation in the art world has started to expose a gap between what the technology is and what people know about it. Past research has shown that people tend to give AI art lower ratings on creativity, value and emotional depth. And in my own work, I had studied how knowledge about art changes the way we view it. This background led me to wonder whether knowledge about AI shapes people’s judgments of AI-generated art and might help explain the often observed bias against it.
To investigate this question, my colleagues and I conducted three experiments, each involving 100 participants. We started by presenting people with AI-generated art images and asking about their morality and aesthetic value. For example, participants in two of these experiments had to rate how morally acceptable it was to use AI to produce such art, earn money or prestige from these works and label them as conventional art. People also had to rate how much they aesthetically appreciated the images we presented.
People don’t yet have a knee-jerk reaction to AI art. Moral resistance is something they learn over time.
In the first experiment, we showed our participants 20 landscapes and 20 portraits that were generated with DALL-E 3 from prompts based on the Impressionist art of Spanish painter Joaquín Sorolla. To ensure that the AI-generated images were highly similar to the artist’s work, we gave the AI very specific instructions. For example, one prompt was “Impressionistic painting of simple wooden Valencian fishing boats in shallow water, 1900s- style, large white sails, loose brushstrokes, Sorolla, master of light.”
Half of the participants viewed this AI art with no added context. The other half received a short text that gave them more information. It read: “This image was generated by an AI algorithm that produces images from textual descriptors. To accomplish that, several steps are required. First, the AI algorithm is trained by learning a large dataset of art images and their corresponding text descriptors, such as the artist’s name. Then, the AI algorithm is able to generate new images based on different textual prompts (e.g., artist’s name, artistic style, whether it depicts a seascape, landscape, or people).”
The additional information made a difference. When people knew how the AI system operated, they perceived the images it produced as less morally acceptable, especially when the creation of these images involved financial gain and artistic acclaim. But the aesthetic appeal of the images did not change, suggesting that learning how AI works made people reflect on ethics, not aesthetics.
Psychologists have found that people’s judgments about what is good or valuable can change when they learn something has earned awards or praise from experts. Authority bias, for example, makes us more inclined to agree with people who seem to be in charge or in the know. In addition, cues such as success or prestige can lead people to see something as more morally good. In our second study, we told participants that some of the AI art had been exhibited, sold or praised. But we were surprised to find that sharing a work’s success did not improve the moral acceptability of the image in the eyes of people who had learned about how the art was created.
In a final experiment, we tested people’s automatic judgments of AI-made versus human-made art. We used a tool from psychology called a go/no-go association task, in which people are asked to very quickly link one kind of prompt, such as an image, with another, such as the word “good” or “bad.” In this experiment, we showed participants images of Impressionist paintings (which were either AI-generated or human-created) along with object-category labels on the left (“AI art” or “human art”) and attribute labels on the right (such as “good” or “bad”). Participants needed to click a button if the image and labels were in alignment and to refrain from responding if they were not. This task needed to be done quickly and over many trials as a way to capture the most immediate associations. We worked with people who had not been given any additional education on AI to try to get a sense of what the average person might think.
We found no strong automatic tendency to see AI or human art as inherently better or worse. This finding tells us that people don’t yet have a knee-jerk reaction or deeply held opinion about AI art as opposed to that made by humans. It also underscores that, as our earlier experiments suggested, moral resistance to AI art is something people learn over time.
Overall, when people know how AI operates, they become more careful in judging its moral fairness. This finding suggests that educating audiences, artists, curators and policymakers about how technology works could shape the future of the technology in the art world. Artists working with AI tools can help in this effort by sharing information about the models, data or prompts they used and clarifying where their own human hand guided the process. Although such transparency may lead to critiques, it might also build credibility and equip people with the tools they need to think critically about technology.
