A recent study published in the journal PNAS Nexus sheds light on the impact of text-to-image generative AI tools like Midjourney, Stable Diffusion, and DALL-E on the artistic process. The researchers discovered that these AI systems tend to enhance artists’ productivity and lead to more favorable evaluations of their work by peers.
Interestingly, while the average novelty in AI-assisted artwork content declined over time, peak content novelty increased. The findings point to the possibility of “generative synesthesia,” a blending of human creativity and AI capabilities to unlock heightened levels of artistic expression.
The advent of generative AI has sparked a heated debate within artistic communities. While some view these technologies as a threat to the intrinsic human ability to create, others recognize their potential to augment human creativity. The research aimed to unravel whether AI adoption enables artists to produce more creative content and under what conditions it leads to more valuable artistic creations.
“In 2022, we witnessed the first truly high-fidelity generative algorithms that could produce creative artifacts that have traditionally been reserved for human faculties,” explained study author Eric Zhou, a PhD candidate in Information Systems at Boston University Questrom School of Business.
“I have several close friends who make a living off exercising their creative talents to produce digital artworks and video games, two domains which directly face the consequences of these technologies. Given I’ve been interested in generative AI and computational creativity, I was naturally drawn to studying how such technologies may be shaping the future of human creativity and expression and to identify under what conditions both traditional artists and AI-assisted artists can flourish.”
For their study, the researchers analyzed a large dataset from one of the largest art-sharing platforms on the internet. The dataset included the complete publishing histories of over 53,000 users, which amounted to more than 4.3 million digital artworks. It included over 5,800 artists known to have adopted AI tools in their creative workflow from January 2022 to June 2023.
To measure the novelty of the artworks produced, the study borrowed the concept of “conceptual spaces,” a method that uses geometric representations in vector space to determine how much an artifact deviates from or aligns with existing works. This approach was complemented by advanced algorithms and models, such as BLIP-2 for content feature extraction and DINOv2 for visual feature extraction, to analyze the artworks’ content and visual elements thoroughly.
The researchers found that artists who adopted generative AI tools experienced a notable increase in productivity, with a 50% rise in the number of artworks produced in the month of adoption and a doubling of output in the subsequent month.
Over time, AI-assisted artworks began to receive more favorable evaluations from peers, indicating an increase in the perceived value and attractiveness of these creations within the artistic community. This suggested that generative AI tools were not only enhancing productivity but also contributing to the production of quality work that resonated with viewers.
However, the researchers also uncovered some nuanced effects of AI adoption on creativity. There was a general decline in the average novelty of content and visual elements among AI adopters — indicating a trend toward conformity in themes or styles — but peak content novelty actually increased. This implies that, despite a broader trend toward homogeneity, some artists were leveraging AI to push creative boundaries and explore new ideas at the forefront of artistic innovation.
“Although generative tools provide many immediate benefits in creative pursuits like boosting productivity by automating the visual realization of ideas and improving the visual fidelity of artists’ works, the works produced by AI-assisted artists are becoming increasingly similar in content and styles over time,” Zhou told PsyPost.
“Still, there is evidence that the human behind the algorithm exercises significant influence over whether they are jointly successful with generative tools. Specifically, individuals who can successfully identify novel ideas when assisted by AI are able to produce more meaningful artworks.”
“Thus, such tools allow artists to focus on the ideas they are representing rather than how they represent it, opening new avenues for creative exploration. We call this ‘generative synesthesia,’ where the unique talents of a creative enables them to explore and exploit exceedingly novel features with AI tools.”
The study, like all research, has some limitations, including its focus on digital art on a single platform and the potential for unobserved factors influencing artist productivity and creativity. Future research could expand to other forms of creative expression and explore the long-term effects of generative AI on artistic innovation.
“Not a major caveat, but we do not observe the exact workflow that AI artists follow when producing their artworks,” Zhou noted. “We should expect high degrees of variation, where some users might be indiscriminately generating generic content while others formulate intricate workflows that maximize the capabilities of available tools. Pointing in this direction, our evidence suggests that humans’ baseline creative talents – which is likely correlated with the ability to extract novel features from AI tools – are key indicators of success with the technology.”
“This article is the initial exploration among a series of research where we seek to explore the unintended consequences of text-to-image generative algorithms on creative pursuits, both in terms of implications on the natural expression of ideas and the actual labor market dynamics that result from such technologies,” Zhou added. “Our long-term goal is to provide guidance on how text-to-image generative AI can be leveraged for human enrichment while also exploring ways for traditional artists to continue to succeed.”
The study, “Generative artificial intelligence, human creativity, and art,” was authored by Eric Zhou and Dokyun Lee.