The European Union's AI Act has been passed recently, which will encourage companies to be more transparent about the data used to train AI systems, potentially unlocking one of the most fiercely guarded secrets in the industry.

From$Microsoft (MSFT.US)$Since OpenAI's ChatGPT was released to the public 18 months ago, participation and investment in generative AI has surged. However, as the industry continues to flourish, questions have arisen about how AI companies obtain data for training models, and whether using best-selling books and Hollywood movies to train AI without the creators' permission constitutes infringement.

The European Union's recent AI Act will be phased in over the next two years, giving regulators ample time to implement new regulations and for companies to fulfill a range of new obligations, but it is still unclear how some of these provisions will operate in practice.

One of the most controversial provisions in the bill requires organizations deploying general purpose AI models, such as ChatGPT, to provide a "detailed summary" of the content used to train the model. The newly established AI office plans to publish a template for organizations to reference after consulting with stakeholders, by early 2025.

Although the details are still being finalized, AI companies are resistant to disclosing model training content, claiming that such information is proprietary and that making it public would give their competitors an unfair advantage.

Matthieu Riouf, CEO of AI image editing company Photoroom, said, "Seeing a competitor's dataset would be a dream come true for us, and it would be the same for them to share our dataset."

He added, "It's like cooking, where some parts of the recipe are secrets that top chefs won't share, the 'secret' that makes dishes different from everyone else's."

The granularity at which the transparency provisions are ultimately enforced will have a significant impact on businesses in the industry, both small AI startups and large tech companies such as$Alphabet-A (GOOGL.US)$ and $Meta Platforms (META.US)$, which have made the technology a cornerstone of their future operations.

Sharing trade secrets

In the past year, many well-known tech companies, including Google, OpenAI, and Stability AI, have faced lawsuits from creators claiming their content was improperly used to train AI models.

While US President Biden has issued several executive orders focusing on AI security risks, copyright-related issues have not been thoroughly examined. Calls for tech companies to pay for data to rights holders have gained bipartisan support in Congress.

Under increasing scrutiny, tech companies have signed numerous content licensing agreements with media and websites. OpenAI has signed agreements with the Financial Times and the Atlantic Monthly, while Google has agreements with the media giant$News Corp-A (NWSA.US)$and social media website$Reddit (RDDT.US)$have reached agreements.

Nevertheless, OpenAI came under criticism in March when its Chief Technical Officer, Mira Murati, refused to answer the Wall Street Journal's question about whether YouTube videos were used to train its video generation tool, Sora. If true, this would violate the company's terms of service.

Last month, OpenAI faced further criticism when the AI-generated voice used during the public demonstration of its latest version of ChatGPT was described by actress Scarlett Johansson as sounding "eerily similar" to her own voice.

Thomas Wolf, co-founder of the artificial intelligence startup Hugging Face, said he supports increased transparency, but it has not been accepted by the entire industry. He said, "It is difficult to know what the outcome will be, and there are still many things to be decided."

There are still differences among senior legislators on the European continent. Dragos Tudorache, one of the legislators responsible for overseeing the drafting of the AI Act in the European Parliament, said that AI companies should be forced to disclose their datasets.

"These datasets must be detailed enough for Scarlett Johnson, Beyoncé or anyone to know whether their works, songs, sounds, art or science have been used to train algorithms," he said.

An official on the committee said, "The AI law recognizes the need to strike an appropriate balance between the legitimate need to protect trade secrets and the promotion of the EU's legal rights, including those of copyright holders."

Under the leadership of French President Macron, the French government privately opposes provisions that could hinder the competitiveness of European AI startups.

In May of this year, Bruno Le Maire, French Minister of Finance, said at the Viva Technology conference in Paris that he hopes Europe will become a world leader in the field of artificial intelligence, rather than just a consumer of American and Chinese products.

"This time, Europe, which has established controls and standards, needs to understand that innovation must precede regulation. Otherwise, there is a risk of regulating technologies that have not yet been mastered, or of conducting bad regulation because these technologies have not yet been mastered."