The European Union's (EU) artificial intelligence (AI) bill has passed the ultimate vote in parliament and is now set to enter effect. Nonetheless, what attracted more attention was the statement that AI corporations were strongly against disclosing the copyright of information used for model training.
Tech Crunch reported on the thirteenth (local time) that consequently of the ultimate vote by EU lawmakers, the 'AI Act' was passed with overwhelming support with 523 votes in favor, 46 against, and 49 abstentions.
Accordingly, if the ministers of the 27 EU countries give final approval next month, it can be published within the Official Gazette and take effect in May, 20 days later. Under the terms, it can be applied in 4 stages starting six months from the tip of this 12 months, with rules affecting general-purpose AI systems reminiscent of chatbots taking effect as much as 36 months later to offer AI providers time to comply. Subsequently, the AI law will likely be safely applied in May 2027.
Moreover, if the ban on AI use will not be followed, a fantastic of as much as 7% of world annual sales will likely be imposed.
Before today's vote, the EU executive was asked intensive questions on whether corporate lobbying had much influence throughout the bill negotiations.
That is on account of the news that some corporations were engaging in intense lobbying activities throughout the negotiation process at the tip of last 12 months. Specifically, last month, a partnership between Microsoft (MS) and French startup Mistral AI was announced, raising suspicions about lobbying connections.
AI Special Committee Chairman Dragos Tudorace, who led the negotiations, said there was no negative impact from the lobbying. “We made a compromise that we thought was reasonable,” he said, asserting that “what the corporate did was its alternative, and it didn’t affect us in any way.”
As an alternative, he said, “the eye of all corporations developing AI models has been focused on maintaining a ‘black box’ of coaching data.”
The AI Act features a provision requiring disclosure of the copyright source of information used to learn large language models (LLM), which is some of the sensitive issues for corporations.
For that reason, some have argued that copyright ought to be left to the copyright laws of every country. Some complained that there was virtually no method to discover the source of the training data.
Nonetheless, Chairman Tudorache said, “We especially want copyrighted materials to be disclosed transparently, because that’s the only method to guarantee the rights of authors.”
Reporter Lim Da-jun ydj@aitimes.com