People generally prefer artworks labeled as “human-created” over those labeled as “AI-created,” according to new research published in the journal Cognitive Research: Principles and Implications. The study suggests that narratives and perceived effort play a crucial role in influencing people’s judgments of artwork. The findings provide unique insights into how people appraise art created by humans compared to art generated by artificial intelligence (AI).
Art is typically seen as a human phenomenon, involving emotions, individual and communal experiences, and social commentary, which are thought to be unique to humans. However, recent advancements in AI have allowed machines to produce high-quality artworks that can be indistinguishable from human-created ones.
“The topic of AI is a really hot, weighted one nowadays,” said study author Lucas Bellaiche, a PhD student at Duke University. “It dominates these huge environments like technology and politics, but it also comes up in your interpersonal conversations with family members and friends. A lot of people are scared of it, a lot of people are impressed by it, and many are both.”
“We actually started exploring the idea around September of 2021, before the big AI boom which occurred a couple months later when you started really seeing AI art and AI deepfakes over the internet. The timing worked out great in that regard. This paper essentially serves to put our finger on the pulse on what people’s perceptions are on AI ‘intruding’ into a seemingly very human form of expression: creativity.”
“So, in one vein, the topic is interesting because it shows the massive developments of AI in recent years, but what I am personally most interested in is actually the opportunity to explore what “creativity” even means to the layperson: is it a form of cognition that can be achieved by anything (including AI), or is it reserved for humans only, as a sort of valued anthropocentrism? What factors of creativity (in this case, art specifically) can be achieved by AI as compared to humans?”
The study aimed to explore whether people indeed have a preference for human-created art and to identify the factors that influence these aesthetic appraisals. To do this, the researchers designed two studies.
In Study 1, they recruited 149 participants from Amazon’s Mechanical Turk and presented them with 30 AI-created paintings labeled as either “human-created” or “AI-created.” The participants rated each artwork based on criteria like liking, beauty, profundity, and worth. They also had to guess whether the artwork was created by a human or AI. The researchers used AI-created artworks exclusively in this study to ensure a controlled and comparable stimulus set.
The researchers found that people tended to prefer artworks labeled as “human-created” over “AI-created” (even though all the artworks were actually created by AI). This preference was consistent across different criteria used to judge the artworks, such as liking, beauty, profundity, and worth. The participants tended to rate human-created artworks higher on all these measures compared to AI-created artworks.
The preference for human-created art was more pronounced when it came to deeper aspects of art, such as its meaning and worth. However, for surface-level aspects like liking and beauty, the differences between human and AI labels were not as significant.
The study also explored individual differences in participants’ judgments, such as cognitive reflection (the ability to override intuitive responses) and attitudes toward AI. Surprisingly, these individual differences did not significantly influence the preference for human-created art over AI-created art.
“There were two major takeaways from our paper,”
“Firstly, we reinforce that art serves two purposes: The first function is the very surface-level enjoyment of art afforded through the senses – the visual (or auditory, for music) properties can be pretty, ugly, symmetrical, etc, and we respond to that by simple ‘liking’ or judgements of ‘beauty.’ The second function is more complex: art serves as a communicative medium from artist to audience. What does the art tell us? Is there a deeper meaning besides what we simply see/hear? What emotion is being conveyed?
“Secondly, importantly, art presumed to be by AI seems to do well on this first function (judgements of simple “liking” and “beauty” are nearly equal to art presumed to be by humans; i.e., AI can indeed make a pretty picture), but not so much on the second function.”
“That is, the average person does not believe AI, compared to humans, to be able to communicate deep meaning very well through art, like emotions, narratives, worth, or profundity. Arguably, for people that are anti-AI, this should come as a relief, in that communicative properties of art will seemingly be reserved for the human and human only, on average.”
In Study 2, the researchers recruited 148 participants from Prolific and used the same 30 AI-created paintings from Study 1. However, in addition to the criteria used in Study 1, they included new judgment criteria such as emotionality, perceived narrativity, personal meaning, perceived effort, and estimated time to create the artwork. They also administered various questionnaires to assess participants’ cognitive abilities, empathy, attitudes toward AI, and beliefs about creativity.
In line with the previous results, the researchers found that people tended to prefer human-created art over AI-created art across all criteria. Interestingly, they also found evidence that narratives and perceived effort were important factors that affect how people judge and appreciate artworks.
When a painting was labeled as “human-created” and was perceived to involve more effort in its creation, people tended to like and appreciate it more compared to when the painting was labeled as “AI-created.” This means that people valued the human effort and labor put into creating the artwork, and it influenced their judgment and appreciation of the art.
When a painting was labeled as “AI-created” and had a higher level of narrativity (strong story or emotional connection), people tended to like and appreciate the artwork more. In other words, narratives seemed to make people appreciate AI-created art more, possibly because the narratives helped overcome biases against machine-made products and encouraged deeper engagement with the art.
“One interesting finding is from Study 2, looking at the moderating effects. We find that participants like and find AI-labelled art (compared to human-labelled art) more beautiful (Liking and Beauty being the two surface-level criteria we assessed) the more they could develop a narrative alongside it. This essentially shows that narratives serve as an engine to appreciate art (particularly by AI) on these surface-level properties.”
“It’s almost like we try to engage with the art by developing a story alongside it, and give it bonus points if it is made by AI as compared to by humans. Much more research is needed here to solidify this claim though, particularly because this analysis is inherently correlational and not causational.”
Moving forward, future research can build upon these findings and delve into more specific questions related to AI-generated art. Investigating the effects of different AI generative models, knowledge levels about AI, and the impact of knowing the true creator of AI-made art could provide deeper insights into audience perceptions and preferences.
“There are so many open questions from this. First, we really chose nine criteria that spanned from surface-level properties (e.g., Beauty) to very deeply engaging ones (e.g., Profound). Future research should focus on these, and others, with more direction rather than exploratorily. Additionally, we chose our 30 stimuli two years ago. AI generative models are so much better now, so those stimuli could be a bit outdated and poorer than what we may see today.”
“Lastly, we use a very reductive term of ‘AI’ in our study to get at the typical, baseline reaction from people on this topic. In reality, ‘AI’ can span so many different kinds of models, and people can have varying knowledge on what AI can actually do. Future work should assess if knowing more about AI influences these judgements. For instance, do people recognize that an AI-made piece of art actually includes some human production element too, like building the model or training the model on human-made pieces?”
Understanding how people perceive and judge AI-created art has implications beyond individual aesthetic preferences. The topic reaches into broader societal contexts, Bellaiche explained.
“I think this kind of research is really important to inform not just our one-on-one interactions with art, but also massive industries,” he told PsyPost. “The current strikes and negotiations in Hollywood, for instance, have a major point about AI replacement, and it is a very valid concern. On this point, this paper shows that if the audience knows the product was made by AI (and I strongly believe the audience does deserve a right to know if something is made by AI or not), then audiences will generally not respond as well to it on these communicative properties of art.”
“In other words, Hollywood should hire human writers if they want their art to communicate things like emotions or rich narratives. However, if you just want to make a pretty piece of art that people don’t engage with (think TikTok AI generators), then AI seems to do that job nearly as (but not completely) equally as humans. Note, however, that there is a whole other discussion of copyright infringement too: if we use AI to build art, we have to make sure the model doesn’t use art made by humans without permission or purchase.”
“But, it is crucial to emphasize that our studies used actual AI-made art, and simply switched the creator labels randomly,” Bellaiche added. “Though we didn’t explicitly ask the participants, they seemed to believe the labels. Thus, there were instances when participants got an actual AI-made piece, believed it was human-made by the label, and rated higher because of that very label. This goes back to my above point: audiences deserve to know the proper creator so they can make an honest judgement of the work of art. While the judgements are subjective, one of the most important factors we need to prioritize as AI continues to influence our society is that of transparency.”
The study, “Humans versus AI: whether and why we prefer human-created compared to AI-created artwork“, was authored by Lucas Bellaiche, Rohin Shahi, Martin Harry Turpin, Anya Ragnhildstveit, Shawn Sprockett, Nathaniel Barr, Alexander Christensen, and Paul Seli.
