Bias against AI art is so deep it changes how viewers perceive color and brightness

New research suggests that simply labeling an artwork as created by artificial intelligence can reduce how much people enjoy and value it. This bias appears to affect not just how viewers interpret the meaning of the art, but even how they process basic visual features like color and brightness. The findings were published in the Psychology of Aesthetics, Creativity, and the Arts.

Artificial intelligence has rapidly become a common tool for visual artists. Artists use technologies ranging from text-to-image generators to robotic arms to produce new forms of imagery. Despite this widespread adoption, audiences often react negatively when they learn technology was involved in the creative process.

Alwin de Rooij, an assistant professor at Tilburg University and associate professor at Avans University of Applied Sciences, sought to understand the consistency of this negative reaction. De Rooij aimed to determine if this bias occurs across different psychological systems involved in viewing art. The researcher also wanted to see if this negative reaction is a permanent structural phenomenon or if it varies by context.

“AI-generated images can now be nearly indistinguishable from art made without AI, yet both public debate and scientific studies suggest that people may respond differently once they are told AI was involved,” de Rooij told PsyPost. “These reactions resemble earlier anxieties around new technologies in art, such as the introduction of photography in the nineteenth century, which is now a fully established art form. This raised the question of how consistent bias against AI in visual art is, and whether it might already be changing.”

To examine this, De Rooij conducted a meta-analysis. This statistical technique combines data from multiple independent studies to find overall trends that a single experiment might miss. The researcher performed a systematic search for experiments published between January 2017 and September 2024.

Free daily newsletter

The analysis included studies where participants viewed visual art and were told it was made by AI. These responses were compared to responses for art labeled as human-made or art presented with no label. The researcher extracted 191 distinct effect sizes from the selected studies.

De Rooij categorized these measurements using a framework known as the Aesthetic Triad model. This model organizes the art experience into three specific systems. The first is the sensory-motor system, which deals with basic visual processing. The second is the knowledge-meaning system, which involves interpretation and context. The third is the emotion-valuation system, which covers subjective feelings and personal preferences.

The investigation revealed that knowing AI was used generally diminishes the aesthetic experience. A small but significant negative effect appeared within the sensory-motor system. This system involves the initial processing of visual features such as color, shape, and spatial relationships. When viewers believed an image was AI-generated, they tended to perceive these basic qualities less favorably.

A moderate negative effect appeared in the knowledge-meaning system. This aspect of the aesthetic experience relates to how people interpret an artwork’s intent. It also includes judgments about the skill required to make the piece. Participants consistently attributed less profundity and creativity to works labeled as artificial intelligence.

The researcher also found a small negative effect in the emotion-valuation system. This system governs subjective feelings of beauty, awe, and liking. Viewers tended to report lower emotional connection when they thought AI was responsible for the work. They also rated these works as less beautiful compared to identical works labeled as human-made.

“The main takeaway is that knowing AI was involved in making an artwork can change how we experience it, even when the artwork itself is identical,” de Rooij explained. “People tend to attribute less meaning and value to art once it is labeled as AI-made, not because it looks worse, but because it is interpreted differently. In some cases, this bias even feeds into basic visual judgments, such as how colorful or vivid an image appears. This shows that bias against AI is not just an abstract opinion about technology. It can deeply shape the aesthetic experience itself.”

But these negative responses were not uniform across all people. The researcher identified age as a significant factor in the severity of the bias. Older participants demonstrated a stronger negative reaction to AI art. Younger audiences showed much weaker negative effects.

This difference suggests a possible generational shift in how people perceive technology in art. Younger viewers may be less troubled by the integration of algorithms in the creative process. The style of the artwork also influenced viewer reactions.

Representational art, which depicts recognizable objects, reduced the negative bias regarding meaning compared to abstract art. However, representational art worsened the bias regarding emotional connection. The setting of the study mattered as well. Experiments conducted online produced stronger evidence of bias than those conducted in laboratories or real-world galleries.

“Another surprising finding was how unstable the bias is,” de Rooij said. “Rather than being a fixed reaction, it varies across audiences and contexts. As mentioned earlier, the bias tends to be stronger among older populations, but the results show it is also influenced by the style of the artworks and by how and where they are presented. In some settings, the bias becomes very weak or nearly disappears. This further supports the observation that, much like earlier reactions to new technologies in art, resistance to AI may be transitional rather than permanent.”

A key limitation involves how previous experiments presented artificial intelligence. Many studies framed the technology as an autonomous agent that created art independently. This description often conflicts with real-world artistic practice.

“The practical significance of these findings need to be critically examined,” de Rooij noted. “Many of the studies included in the meta-analysis frame AI as if it were an autonomous artist, which does not reflect artistic practice, where AI is typically used as a responsive material. The AI-as-artist framing evoke dystopian imaginaries about AI replacing human artists or threatening the humanity in art. As a result, some studies may elicit stronger negative responses to AI, but in a way that has no clear real-world counterpart.”

Future research should investigate the role of invisible human involvement in AI art. De Rooij plans to conduct follow-up studies.

“The next step is to study bias against AI in art in more realistic settings, such as galleries or museums, and in ways that better reflect how artists actually use AI in their creative practice,” de Rooij said. “This is a reaction to the finding that bias against AI seemed particularly strong in online studies, which merits verification of the bias in real-world settings. This proposed follow-up research has recently received funding from the Dutch Research Council, and the first results are expected in late 2026. We are excited about moving this work forward!”

The study, “Bias against artificial intelligence in visual art: A meta-analysis,” was authored by Alwin de Rooij.