“Labels play an important role in people’s general evaluations of things.” mentions Bellaiche et al. (2023) in their introduction to their study of whether we prefer human-made or AI-generated art. Thus, they set out to find the answer to this question and why we may prefer one or the other. In order to do so, they designed the two following experiments.
Research design
With a relatively large sample size, thirty paintings generated via ArtBreeder AI were created. Half of them were representational while the other half were abstract. Since researchers hypothesized there was a chance that individuals with greater ability to override intuition showed less bias, a test to measure such cognitive ability was conducted beforehand for the first study, and a more exhaustive battery of tests was set in place for the second.
The paintings were presented in random order and individually assigned a label or either human- or AI-made (also randomly). A label-check question was also included to ensure the participants had correctly noticed assigned authorship of each piece.
| Study 1 | Study 2 | |
| Participants | 149 | 151 |
| Pre-tests | Cognitive Reflection Test | Cognitive Reflection TestToronto Empathy QuestionnaireGeneral Attitudes toward Artificial Intelligence ScaleCreative Mindset ScaleOpen-to-Experience subset of NEO Five Factor Personality Inventory |
| Rating values | LikingBeautyProfundityWorth | LikingBeautyProfundityWorthEmotionStoryMeaningfulEffortTime |
Results
Study 1 Results
As the researchers expected from previous literature, the results from this first study showed a bias against AI-labelled artworks: those pieces that had been assigned the “human” label, were better liked, and perceived to have more Worth and Profundity. On a side note, Beaty values were higher for human-labelled paintings when the Cognitive Reflection test scores were lower. This test’s scores were otherwise not statistically significant.
Study 2 Results
The results of this second study showed that Liking was positively linked to higher scores of Story, Emotion and Meaning, and confirmed the higher likeability of representational paintings. These three factors had parallel effects for Beauty, which was extra affected by participants’ age: well below average age correlated with higher Beauty values. In addition, for higher Story scores, AI-labelled works obtained higher Liking than human-labelled ones. Effort, however, obtained higher scores when the label was “human”.
As for the values of Profundity and Worth, they both seem similarly connected with higher scores of the five factors added to this second study (Emotion, Story, Meaning, Effort and Time). All five metrics were however higher for human-labelled than for AI-labelled. Additionally, Worth was higher for human-labelled pieces when positive attitude towards AI was well below average, but after reaching below average values, it was AI that presented a higher Worth. A similar phenomenon could be seen for Beauty in relation to “openness to experience”.
Conclusions
Since all the pieces in these two studies were AI-created, and considering AI art is hardly discernible from human-made art, it became apparent that labels had a strong influence on the judgement of visual artwork. However, this may become nearly insignificant at surface level when combined with communicative processes that emphasize Effort and Story, as AI seems to be better appreciated under quick, superficial conditions. That is, while the influence of labels appears stronger when engaging deeply (and the “human” label encourages this deeper engagement), enhancing Story through appropriate narratives and Effort by promoting better understanding of how AI art is created is likely to lessen the effect of said labels. Additionally, higher Story and perceived Effort do lead to better Liking and higher Beauty scores.
These, as pointed out by Bellaiche et al., present strong implications for marketing and product branding strategies. The researchers also add that it reminds of advertising practices in the early twentieth century, when the arrival of mass-produced goods by multinational companies left consumers hesitant to purchase such goods (Freeman, 2014; Pulizzi, 2012). At that time, too, backstories and understanding of the process was conducive to overcoming biases.
Bellaiche, L., Shahi, R., Turpin, M. H., Ragnhildstveit, A., Sprockett, S., Barr, N., Christensen, A., & Seli, P. (2023). Humans versus AI: Whether and why we prefer human-created compared to AI-created artwork. Cognitive Research: Principles and Implications, 8(1), 42. https://doi.org/10.1186/s41235-023-00499-6
Freeman, M. (2014). Transmedia Critical| Advertising the Yellow Brick Road: Historicizing the Industrial Emergence of Transmedia Storytelling. International Journal of Communication, 8(0), Article 0.
Pulizzi, J. (2012). The Rise of Storytelling as the New Marketing. Publishing Research Quarterly, 28(2), 116–123. https://doi.org/10.1007/s12109-012-9264-5
