The European Parliament voted at noon on the Artificial Intelligence Act. Where does Europe stand on regulating artificial intelligence?
The EU is pushing further the Artificial Intelligence Act today. To this end, a vote on the AI Act draft was scheduled in the European Parliament this afternoon. The draft has been in the works since 2021.
The Act aims to identify and address the dangers of artificial intelligence to the public, particularly in copyright and privacy. For this purpose, each AI product or platform will be categorized, which will determine the risk factor. Products and platforms with unacceptable risk will not enter the EU at all.
‘Historic vote’
The vote was introduced by Brando Benifei and Dragoş Tudorache, MEPs at the forefront of the AI Act: “It will be a historic vote in which we not only address the biggest problems with AI but also send a signal to the world about how we believe our democracy should be protected while supporting innovations.”
In doing so, the MEPs put another spotlight on the fact that the AI Act will be the first set of regulations around artificial intelligence globally.
MEPs are proposing three changes to the bill that the Commission previously approved. First, they want the list of prohibited AI practices to be reviewed and expanded. The same goes for the list of high-risk applications.
Finally, the House would like to see the addition of transparency and security provisions for foundation models and generative AI systems, such as ChatGPT. In this way, the European Parliament would like to increase understanding of such AI applications and the potential risks such as discrimination and spilling sensitive information. To that end, it should become possible to categorize such AI applications into one of the four categories of the AI Act: minimal risk, limited risk, high risk and unacceptable risk.
Tip: ‘ChatGPT based on illegal sites, private data and piracy’
Clear majority
A clear majority approved the bill. 499 members voted in favor, 28 against and 93 opposed. Roberta Metsola, president of the European Parliament, begins the press conference positively: “I would like to thank Benifei and Tudorache for a balanced and human-centred approach to the legislation.”
The president goes on to emphasize that Europe is leading the way and takes responsibility in an era of AI. “Many things cannot be digitized: emotions, will and judgment. We must use these things to set a global example.”
Then the floor goes to Benifei. He sees that a draft was approved today that will spark further conversation. In addition, he emphasizes that the law should guarantee Europe’s residents that they are using secure AI systems and products. “We had the courage today to ban AI players that we believe don’t accept EU values.”
Dangers of harmful AI content
Furthermore, the Commission’s proposal has been supplemented because the House considers marking AI content in the digital space essential.
AI content is essentially not always correct or safe. Last week, the EU asked all social media sites to flag AI content. This is currently already necessary to counter the flow of disinformation from Russia, the EU judges.
The AI law will complement the Digital Services Act (DSA), which will be in force by the end of August 2023. The DSA already handles this partially because the European Parliament fears artificial intelligence will affect the 2024 elections if regulation is not implemented in time.
The legislation voted on today will supplement the DSA with rules around biometric identification. The states do not agree on this topic, Bonifei adds: “There was discussion in the House today about the ban on biometric identification. There was interest in turning this into a propaganda tool, but we won in the House and can now secure ourselves from mass surveillance.”
What about ChatGPT?
If the regulations take effect today, ChatGPT would have several issues to sort out. Namely, OpenAI will have to transparently disclose to the EU which datasets they used in training the language model and point out which are copyrighted. Furthermore, the EU must be informed of how the AI product works.
Generative AI systems must also clearly state under the AI Act that AI created the content. Finally, such systems must set up barriers to prevent generating illegal content.
‘Encourage innovation, not inhibit it’
Tudorache further adds, hoping to address developers’ concerns about the AI Act partially, “We served the agenda of the European Parliament and at the same time the agenda of encouraging innovation. We consider the latter as important as protecting EU residents.” He generally names the draft as “a balanced proposal”. The AI Act has come under fire from developers and experts because it could put a major brake on the development of AI in Europe.
The draft further takes a definition of AI and aligns it with the one that applies in America. This simplifies collaborations between companies for AI innovations. “Furthermore, there is a clear mandate for a standard framework. That was missing in other EU legislation, such as the GDPR, so companies struggled to translate the legalisation to technical standards. We decided to do this bottom-up and will involve developers and companies already active in the field of AI in creating these standards.”
Finally, he wants to clarify that the law must ensure that every resident of the EU is on board with the AI transformation. “We must invest massively so that every organization embraces AI.”
Tinkering until 2025
The legislation must be generally applicable by 2025. Years of discussion about the content of the law will precede that. Today, the House voted on a draft of the legislation, which could provide a clear framework to get the legislation up and running.