Artificial intelligence has been a game changer across countless industries. Its disruptive impact is felt particularly strongly in the creative industry, especially among photographers and other visual artists. Digital creators face an increasing encroachment on their business by generative AI. Some platforms are now offering images created either by individuals or by AI. And the AI systems used to create the images are themselves trained using images provided by human creators.
To examine the effect AI is having on the creative field, Gies Business professor Nathan Yang and his coauthors, Lijuan Luo and Emaad Manzoor from Cornell University and Fan Yao from UNC Chapel Hill, explored how AI can affect whether – and how much – artists get paid for work that may later be used to train AI. Their paper “Platform Design When Creators Train Their AI Substitutes” warns that artists and photographers might be training the very AIs that take their jobs.
Modeling the economic impact of AI
The researchers focused on three questions for the creative marketplace in the face of increasing use of AI:
- Should creative content marketplaces integrate consumer-facing generative AI systems?
- How should marketplaces price content created using the generative AI system relative to human-created work?
- Should platforms integrating generative AI systems compensate creators for their voluntary contributions to fine-tuning data, or use the content without compensation?
To examine these questions, they built an economics model to analyze how AI affects profits, prices, and fairness when trained on human-made content. In the model, two creators (one low quality and the other high quality) sell their content on a particular platform. In addition to the work of these human creators, the platform also offers consumers the option to create similar work using AI.
Give the customers what they want
In their analysis, they set their model such that the generative model is perfectly adjustable to when both creators contribute to the fine-tuning data. Their results show the trade-offs that all stakeholders in creative marketplaces face when integrating generative AI.
Integrating generative AI as an option to a platform could increase profits for the platform by enabling consumers to generate better content for their needs. While the customer was provided two human-created options in their study, AI enables a third option that serves customers whose preferences lie somewhere between the two. Adding AI into the mix enables creation of something optimal for the customer – a result the customer really wants.
Yang and his colleagues suggest that a main insight derived from their study is that companies could charge more for AI-developed content.
“I know on the internet, AI content is often perceived as a cheaper option, because humans were not involved,” Yang said, “But the reality is that the quality might actually be better from a personalization standpoint. And so customers actually get something out of this that they might be willing to pay more, especially as this technology continually improves in terms of usability.”
Use of AI gives the customer exactly the kind of creative content they are looking for – something for which they may be willing to pay more. The AI use adds value to the final product.
One of the key concepts in the study is steerability: the ability of the user to guide the AI platform toward the desired output. The greater the steerability, the easier it is for customers to adjust an image to their needs or preferences. Greater steerability makes an AI platform more attractive as an option. And the use of AI could increase company profits.
The future of creator compensation
They also considered compensation for artwork used to train the AI algorithm. Some marketplaces compensate content creators in proportion to the share of content used in fine-tuning the data. Others do not provide compensation to creators. This is one of the areas that Yang and his colleagues examined: whether it is better for creators to be compensated or not. They also examined how fine-tuning of the data used for the AI would impact the entities involved: platform, creators, and consumers.
“What exactly determines that market rate that you're willing to pay that creator for that artwork?” he asked. “When you buy, you don't know exactly how much demand you're going to get. So these payments, at least in our model, simplify things, because we're simply saying, okay, they're going to get some fixed cut of the realized demand. But you might not know exactly what the demand will be.”
Another consideration is payment to the creators for using their images to train and fine-tune the AI algorithms so that customers can then prompt and create higher-quality AI-generated products. “You can imagine that when you're setting up the platform, it kind of collapses if no photographer wants to contribute. It kind of dies,” Yang said.
They did examine the scenario where creators are compensated to some degree when their work is used by AI to create something for a customer. In that case, creators are certainly better off, since they get paid. However, there is often no real incentive for platforms to pay creators. While platforms would pay for any direct sales of a creator’s content, there is nothing currently obligating them to pay for content that is augmented by AI.
“The platform would actually prefer just to keep the profits entirely from the AI,” Yang said. “I think at the end of the day, it might become as if platforms are just pirates about it. So I have a slightly pessimistic view.”