Disney stated that its AI image generator Midjourney was developed using films like ‘The Lion King’ Maximum Film/Alamy
Since the launch of ChatGPT, OpenAI’s generative AI chatbot, three years ago, we’ve witnessed dramatic shifts across various aspects of our lives. However, one area that remains unchanged is adherence to copyright law. We still strive to uphold pre-AI standards.
It’s widely recognized that leading AI firms have developed models by harvesting data from the internet, including copyrighted content, often without securing prior approval. This year, prominent copyright holders have retaliated, filing various lawsuits against AI companies for alleged copyright violations.
The most notable lawsuit was initiated in June by Disney and Universal, claiming that the AI image generation platform Midjourney was trained using their copyrighted materials and enabled users to produce images that “clearly included and replicated Disney and Universal’s iconic characters.”
The proceedings are still underway, with Midjourney’s recent response in August asserting, “The limited monopoly granted by copyright must yield to fair use,” suggesting that the outcome would be transformative, permitting AI companies to educate models with copyrighted works.
Midjourney’s statements highlight that the copyright debate is more complex than it might seem at first glance. “Many believed copyright would serve as the ultimate barrier against AI, but that’s not entirely true,” remarks Andres Guadams from the University of Sussex, UK, expressing surprise at how little impact copyright has had on the progress of AI enterprises.
This is occurring even as some governments engage in discussions on the matter. In October, the Japanese government made an official appeal to OpenAI, urging the company behind the Sora 2 AI video generator to honor the intellectual property rights of its culture, including its manga and beloved video games like those from Nintendo.
Sora 2 is embroiled in further controversy due to its capability to generate realistic footage of real individuals. OpenAI recently tightened restrictions on representations of Martin Luther King Jr. after family representatives raised concerns about a depiction of his iconic “I Have a Dream” speech that included inappropriate sounds.
“While free speech is crucial when portraying historical figures, OpenAI believes that public figures and their families should ultimately control how their likenesses are represented,” the company stated. This restriction was only partially effective, as celebrities and public figures must still opt-out from having their images utilized in Sora 2. Some argue this remains too permissive. “No one should have to tell OpenAI if they wish to avoid being deepfaked,” states Ed Newton Rex, a former AI executive and founder of the campaign group Fairly Trained.
In certain instances, AI companies face legal challenges over their practices, as highlighted by one of the largest proposed lawsuits from the past year. In September, three authors accused Anthropic, the firm behind the Claude chatbot, of deliberately downloading over 7 million pirated books for training its AI models.
A judge reviewed the case and concluded that even if the firm had utilized this material for training, it could be considered a sufficiently “transformational” use that wouldn’t inherently infringe copyright. However, the piracy allegations were serious enough to warrant trial proceedings. Anthropic ultimately decided to settle the lawsuit for at least $1.5 billion.
“Significantly, AI companies appear to be strategizing their responses and may end up disbursing a mix of settlements and licensing deals,” Guadams noted. “Only a small number of companies are likely to collapse due to copyright infringement lawsuits,” he adds. “AI is here to stay, even if many established players may fail due to litigation and market fluctuations.”
topic:
Source: www.newscientist.com
