Using artificial intelligence to create artworks increases artists鈥� productivity and generates more positive reactions, according to a study involving submissions to a popular art-sharing website by more than 50,000 users.
However, generative AI works are more likely to display stereotypical themes and depictions, reducing the novelty of the artist鈥檚 work.
and at Boston University examined work posted on an art-sharing site between January 2022 and May 2023, a period that covered the release of the AI image generators Midjourney, DALL-E and Stable Diffusion.
Advertisement
The researchers tracked 4 million works published on the platform by 53,000 users in that time. The users self-divided into those who continued to work using traditional methods 鈥� a sort of control group 鈥� and those who adopted AI. The latter, who numbered around 5800 people, were singled out by the use of tags on their work, such as 鈥淎I-generated鈥� or the names of the AI tools. Works posted into AI art communities on the site were also included.
Those who adopted AI tools saw their productivity 鈥� measured as the number of works posted 鈥� increase by 25 per cent over the study period. They also saw a 50 per cent increase in the number of 鈥渇avourites鈥� their work received over six months. But novelty, measured by the subject matter and specific details on the work, decreased for the AI-using group.
鈥淭he productivity effects were to be expected,鈥� says Zhou. 鈥淲hen it comes to 鈥榲alue鈥� 鈥� the conceptual perception of this artwork as acceptable in the current climate 鈥� there鈥檚 a lot of underlying mechanisms that could be going on.鈥�
He believes the human artists using AI might have found a subculture within the platform that is accepting of this kind of work. 鈥淥r it could be that the quality of the artwork is potentially indiscernible from those of the traditional artists,鈥� he says.
Zhou declined to name the art-sharing platform from which the works were analysed, citing a non-disclosure agreement, but he says it was a user-generated art website of a type similar to DeviantArt.
鈥淭he study completely tracks with my own experience with these tools and spaces,鈥� says at the University of Sussex, UK. 鈥淎t first it felt that you could do anything, bring anything in your head into existence, but the more I used them, the more I kept coming back to themes I liked.鈥�
Guadamuz points out that he isn鈥檛 judging the quality of the images produced by generative AI 鈥� nor their originality, nor the creativity involved in the process. All of those are often hotly contested, with some artists saying that generative AI is exploiting their works to use as training data and limiting their ability to be recognised as creatives. 鈥淚鈥檝e been thinking about it more as exploration and not so much as creation,鈥� says Guadamuz. 鈥淭hinking of latent space as an undiscovered country, using these tools is more an act of discovery. You uncover a new image.鈥�
Zhou is also considering these questions. 鈥淕enerative technologies are essentially giving everyone the same baseline skill,鈥� he says. 鈥淚t accelerates the ability to produce. But it raises other issues: are we foregoing the process of understanding what goes into being creative and producing something meaningful, in favour of just being able to brute force our way with technology?鈥�
PNAS Nexus
