Вештачка интелигенција

The Future of Artificial Intelligence Regulation in the EU

Summary

After extensive negotiations, the negotiators from the European Union have finally reached a political agreement on a new AI law, which will be the world’s first and potentially the most influential set of rules for the artificial intelligence sector. While […]

The Future of Artificial Intelligence Regulation in the EU

After extensive negotiations, the negotiators from the European Union have finally reached a political agreement on a new AI law, which will be the world’s first and potentially the most influential set of rules for the artificial intelligence sector. While the details are yet to be announced, we have a good idea of what to expect when the law comes into effect in a few years. In the meantime, a voluntary “AI Pact” will be implemented to bridge the current regulatory gap.

Basic models like GPT-4, along with widely-used systems like ChatGPT, will have to be shipped with technical documentation and comprehensive summaries of the content on which they were trained. Systems deemed to pose systematic risks will undergo “adversarial testing” and adhere to rules on cyber security and energy efficiency. The agreement also introduces government-supported “regulatory sandboxes” to allow startups and smaller companies to develop and train new AI systems and test them in real-world conditions before deployment. It’s worth noting that fully open models will be subject to less stringent regulations compared to proprietary models.

High-risk AI systems in fields such as insurance and banking, as well as those that can influence voter behavior, will have to undergo impact assessments, and European citizens will have the right to request explanations on the decision-making processes that may affect them. Some practices will be directly prohibited, including social scoring systems, workplace emotion recognition systems, AI systems that exploit vulnerable individuals, and biometric categorization systems that use characteristics such as race and sexual orientation.

While there are satisfied individuals with the outcome, including negotiators who managed to successfully conclude this process on time, there are also critics who are dissatisfied with the regulation. Technology lobbyists and human rights activists have expressed their disapproval. Tech companies and associations such as Microsoft, Amazon, Meta, Google, Apple, and the Computer & Communications Industry Association believe that EU lawmakers have acted hastily and should slow down the enactment of legislation until they have a better understanding of what they are regulating. On the other hand, human rights activists and organizations like the European Consumer Organisation and European Digital Rights argue that the law fails to adequately address concerns related to the power of artificial intelligence and the potential for manipulation, discrimination, and mass surveillance.

How regulation will develop and how much it will impact the artificial intelligence sector remains to be seen in the coming weeks when the technical drafts of the law will be published. Regardless of the varying views and reactions, one thing is certain – the new AI law will have a significant influence on Europe’s digital future.

### Frequently Asked Questions

**What is the AI Act?** The AI Act is the first and most comprehensive set of rules for the artificial intelligence sector in the EU.

**Who will be regulated by the AI Act?** Basic models like GPT-4 and widely-used systems like ChatGPT will be regulated by the AI Act, along with high-risk AI systems in fields such as insurance and banking.

**When will the AI Act come into effect?** The AI Act will come into effect in a few years, but a voluntary “AI Pact” will be established to bridge the current regulatory gap.

**What are some of the prohibited AI systems?** Prohibited AI systems include social scoring systems, workplace emotion recognition systems, AI systems that exploit vulnerable individuals, and biometric categorization systems that use characteristics such as race and sexual orientation.