People ascribe intentions and emotions to both human- and AI-made art, but still report stronger emotions for artworks made by humans

A new online experimental study examined how individuals perceive AI- and human-made art created to incite emotions. The results indicated that people attributed intentions and emotions to the creator of the artwork, regardless of whether it was produced by artificial intelligence or a human. However, they felt stronger emotions when viewing art made by humans. The study was published in Computers in Human Behavior.

In recent years, there have been significant advancements in AI’s practical applications. Ranging from AI chatbots to generative transformer AIs, and even AIs capable of tasks once believed to be exclusively human, AI technology is reshaping our society. One of these areas is art.

Back in 2018, the auction house Christie’s sold a painting for $432,500. This would not have been uncommon were it not for the fact that this painting, titled “Portrait of Edmond Belamy” was produced by an AI algorithm. Many were shocked by this development. However, in the 5 years since this event, AI algorithms that produce artwork have become fairly common.

Study author Theresa Rahel Demmer and her colleagues wanted to examine the emotional reactions people have to computer-generated (i.e., AI-generated) and to human generated art. Previous studies have suggested that humans might be better at conveying emotional experiences through artwork. Study authors wanted to know if individuals routinely report experiencing specific emotions or sensing emotional intentions in AI-generated art. They were also interested in whether individuals will be able to differentiate AI and human-made art when they look similar.

The study comprised 48 participants, sourced through social media and the online platform SurveyCircle. Of these, 25 were male, and their educational backgrounds ranged from secondary education to PhD.

Free daily newsletter

The art pieces shown to participants consisted of 32×32 black and white grids. These grids were either populated randomly by a computer using an online noise generator or by human artists, which included online visual artists, fine art students, and degree holders in the field. Human artists were tasked with selecting one or more emotions they aimed to convey through the artwork and subsequently reporting it.

Participants were shown these artworks and were informed beforehand whether each piece was created by a computer or a human. They were also briefed that they would be viewing an approximately equal number of human and AI-created art pieces. In the end, they saw 14 human-created and 10 AI-created artworks.

For every piece, participants indicated their appreciation, evaluated aesthetic attributes, and described specific emotions they experienced while viewing it. They also provided insights into the emotions they believed the artwork’s creator intended to communicate.

In the first two blocks of the experiment, participants viewed two sets of 12 artworks. Before viewing, they received a message indicating whether the upcoming artworks were human-made or computer-generated. In the experiment’s third segment, they viewed all 24 pieces without any information about the creator.

The findings showed that participants accurately identified the creator (human or computer) of the artwork 63.8% of the time. Though not extremely high, this percentage surpassed random guessing. However, this accuracy varied greatly between participants. There was one participant who was accurate 100% of the time and 2 participants who correctly guessed the author only 25% of the time (much below random chance).

There were no discernible differences in ratings for artworks labeled as human-created versus computer-generated. However, art pieces genuinely crafted by humans received marginally more favorable ratings across various experiential aspects such as beauty, interest, quality, pleasantness, meaningfulness, and clarity. The most pronounced differences were in beauty and interest evaluations.

Participants reported feeling at least some emotions with 77% of viewed artworks. 78% of participants reported not feeling any emotions for at least one artwork. They most frequently felt emotions for human-made artworks identified as such and least frequently for computer-generated pieces labeled accordingly. Subsequent analysis revealed that these differences were primarily influenced by the label rather than the actual creator. No significant variation existed in the intensity of emotions reported for AI-created versus human-made art, though slight differences emerged when evaluating individual emotions.

In 43% of cases, participants believed the creator did not intend to convey emotions. They most frequently ascribed this lack of intended emotions to AI-created artworks identified as such. Similarly, results revealed a pattern in the number of emotions participants believed the creator wished to communicate. However, participants often attributed emotional intentions to AI-created artworks.

“In conclusion, our study did find rather compelling, consistent evidence suggesting that participants reported feeling emotions and also ascribing intentions to artworks, regardless of the prime regarding whether these were from a computer or a human artist,” the study authors wrote. “Interestingly, such reports with computer-generated images contradict the belief that AI art lacks an ability to evoke emotional and intentional human elements—at least when considered from the actual perspective and self-reports of a viewer.”

“At the same time, the study did find that the evaluation and specific nature of emotional experiences of viewers were influenced by the provenance, with human-made art evoking stronger responses, and with participants exhibiting the ability to often recognize emotions intended by the human artists.”

The study makes an important contribution to the scientific understanding of how humans perceive AI-generated artwork. However, it also has limitations that need to be taken into account. Notably, the number of study participants was very small, the artworks used were of very limited complexity and constrained to be similar. Results might not be the same if the sample was larger and artworks created with full artistic freedom were compared.

The study, “Does an emotional connection to art really require a human artist? Emotion and intentionality responses to AI- versus human-created art and impact on aesthetic experience”, was authored by Theresa Rahel Demmer, Corinna Kühnapfel, Joerg Fingerhut, and Matthew Pelowski.