The Computer, A Choreographer? Aesthetic Responses to Computer-Generated Dance Choreography

Abstract

Is artificial intelligence (AI) changing our culture or creating its own? With advancements in AI and machine learning, artistic creativity is moving to a brave new world of possibility and complexity, while at the same time posing challenging questions, such as what defines something as art, what is the role of human creativity in an automated world, and do we evaluate artificial art in the same way as art made by humans? Across two pre-registered and statistically powered experiments we shed light on the nature of aesthetic responses toward computer-generated art by investigating observer prejudices against computer-generated dance choreography, and the impact of expertise and pre-conceived beliefs about the origin of artistic creation. Our results provide substantive evidence that an explicit bias exists among dance experts against computer-generated choreography, and the mere belief about a dance work’s origin biases aesthetic responses toward artworks among both dance experts and dance naïve participants. The implications of the current study serve to inform several disciplines across the arts and sciences including but not limited to empirical aesthetics, artificial intelligence, engineering, robotics, and social cognition and neuroscience. Along with physical form and content of artificial agents and art productions, the viewers’ knowledge and attitudes toward artistic AI and artificial agents will need to be taken into consideration for effective human-computer/human-AI interactions.

Value attributed to text-based archives generated by artificial intelligence

“…Our results are in line with previous research in the domain of automated journalism and empirical aesthetics that suggest a bias against AI-generated productions-we found a trend for individuals to assign lower values to AI-generated archives compared with original archives, and individuals assigned lower values to archives they themselves categorized as AI-generated [6,[12][13][14]21,26,[45][46][47]. We also support previous findings of a relationship between positive attitudes toward AI and a higher acceptance of AI applications [28][29][30][31]48].…”

“…While the explicit bias was not higher than the implicit bias for archives that were, in reality, artificially generated, we predicted that when participants themselves categorized archives as AI-generated or original, they would continue to show a bias toward AI-generated archives. In line with this, and following on from previous evidence [ 12 , 13 ], we found that participants assigned lower ratings of value to archives they thought were AI-generated compared with those they thought were original. This bias was evident even when choosing to preserve or destroy archives—archives categorized as original were more likely to be preserved, both by participants who did the rating task first and those who did the categorization task first.…”

“…Consistent with previous literature and our preregistered hypotheses, participants were not able to accurately discern between AI-generated and original archives [13,14]. Accuracy for AI-generated archives was lower than original archives.…”

“…Research in the domain of empirical aesthetics suggests that creative content such as music, paintings or dance that are generated by machines are preferred less than human-generated content, especially when people are explicitly aware that such productions are computer-generated [ 12 , 13 ]. Even for creative text-based content such as poetry, some work has shown that individuals cannot distinguish between AI-generated poetry and poetry written by humans, yet still prefer human-written poems over AI-generated poems [ 14 ].…”

“…[27]) or how evaluations of creative productions made by computers are impacted by expertise in that creative domain (e.g. [13]). However, people's attitudes toward AI also impact their behavioural responses.…”