The European Union’s AI Act has added a new provision that forces generative AI companies to disclose copyrighted material within the datasets they use.
According to a report from the Reuters news agency, companies such as Midjourney will have to reveal the material used to train its artificial intelligence (AI) models. It will be the same for generative language models like ChatGPT.
However, copyrighting words is more ambiguous than copyrighting photos, the latter is black-and-white in terms of intellectual property.
The text-to-image model Stable Diffusion uses the open-source dataset LAION, which is publically searchable and has led to lawsuits against it by Getty Images.
However, companies like Midjourney and DALL-E do not publically disclose their data sources. But, the former has admitted to using “hundreds of millions” of images without consent.
The EU’s AI Act would force Midjourney and DALL-E to reveal its data pool, which would inevitably reveal millions of copyrighted photos taken by professional and amateur photographers.
DIY Photography notes that if the law goes through, it could potentially trigger a rush of lawsuits against the AI image generators.
The European Commission began drafting the AI Act nearly two years ago to regulate nascent AI technology; which has exploded in popularity in the last 12 months thanks to the likes of Midjourney and ChatGPT.
EU lawmakers and member states are currently thrashing out the final details of the bill as they try to push the draft through to the next stage, they hope the legislation will be passed in 2023.
Under the proposals, AI tools will be classified according to their perceived risk level: from minimal through to limited, high, and unacceptable. Areas of concern could include biometric surveillance, spreading misinformation, or discriminatory language.
While high-risk tools will not be banned outright, the companies deploying them will need to be highly transparent in their operations.
“Against conservative wishes for more surveillance and leftist fantasies of over-regulation, parliament found a solid compromise that would regulate AI proportionately, protect citizens’ rights, as well as foster innovation and boost the economy,” says Svenja Hahn, a European Parliament deputy.
An analyst from Macquarie tells Reuters that the EU’s proposal was “tactful” rather than a “ban first, and ask questions later” approach proposed by some.
“The EU has been on the frontier of regulating AI technology,” says Fred Havemeyer.