Artificial Intelligence Platforms and Training Methods
Artificial intelligence software has been widely adopted across technology, marketing, and other commercial sectors. These technologies use large datasets of existing text and images to learn to predict outputs, based on human input. ChatGPT is a large language model that generates text, whereas Stable Diffusion, an artificial intelligence image-generation system developed by Stability AI, generates images. Data used for training is sourced from publicly available internet content.
Because copyrights and trademarks are not always clearly identified or respected on the internet, a central legal issue arises: whether AI’s practical need for extensive training data affects trademark and other intellectual property rights.
Getty Images filed a federal suit in a Delaware federal court, alleging trademark infringement by Stable Diffusion. Stable Diffusion generates images by processing written descriptions submitted by human users and recreating them based on patterns learned from images scraped from the internet.
Allegations Against Stability AI
The Getty Images lawsuit is one of the first trademark challenges that directly address data scraping practices used by artificial intelligence developers. Getty alleged that Stability AI copied more than 12 million Getty Images photographs, along with associated metadata, descriptions, and search engine optimization content, to train Stable Diffusion.
According to the complaint, the copying was intentional, and the resulting AI-generated images sometimes contain modified versions of the Getty Images watermark. Getty claims the created content may create consumer confusion about the source of the images, and that consumers may believe that the AI tool is associated with or endorses Getty Images. Getty sought injunctive relief and damages associated with copyright infringement and asked that versions of Stable Diffusion trained on Getty Images content be destroyed.
Issues Arising from Ownership of AI-Generated Content
Novel legal questions extend beyond the Getty complaint. Whether artificial intelligence companies can lawfully use protected intellectual property as training data is one question, and whether content generated by artificial intelligence systems can itself be protected is another. Because legislation takes time, courts may be required to determine whether ownership lies with the technology's developer, the user, or another party altogether.
Trademark Litigation, AI, and Consumer Survey Evidence
In this new era, consumer survey research may still provide empirical evidence to inform trademark litigation strategies. Survey evidence can be used to measure whether consumers are confused by the source, sponsorship, or affiliation of the AI-generated images, especially if watermarks are visible.
If issues do arise that involve text-generating artificial intelligence systems trained on copyrighted books or branded materials, and the resulting legal claims rest on consumer perception, surveys may be employed to help adjudicators make decisions.
The Litigation Surveys and Consumer Sciences Group at IMS Legal Strategies provides rigorous consumer survey research to support trademark and emerging technology disputes. We work closely with counsel to develop survey evidence that aligns with judicial standards and withstands scrutiny. Contact IMS if you require defensible consumer survey research.
