May 11, the European Parliament adopted the world’s first rules for artificial intelligence (AI ACT). The rules are aimed at ensuring security, transparency and ethics of AI in Europe. The rules provide for a risk-oriented approach and establish obligations for suppliers and users of AI, depending on the risk level that can create AI.
AI ACT has wide jurisdiction and applies not only to EU companies, but also to organizations that are located in third countries if their II systems are intended for use in the EU or fall under the EU law under international law. This means that American companies and developers who offer their AI products or services in the European market must comply with all the requirements of the EU act.
AI ACT can have especially serious consequences for American companies and developers that are engaged in generative and models such as GPT-3 from Google Openai or Bert, which are used to generate text, images, sound and other formats. It is worth noting that American companies that offer their models through the API or according to the Open Source model will face great difficulties, since AI ACT contains several prohibitions:
- AI ACT prohibits access to generative models through the API, which allows third parties to use the model without installing it on their equipment;
- It is also forbidden to spread such models through open sources (Open Source) if they are not Certification in the EU has passed;
- also AI ACT requires the disclosure of confidential technical information about the model to third parties that want to get a license to use the model;
- ai ACT requires the redistribution of models if they are if they are Show unexpected abilities or receive additional training. This can be a problem for those models that use the Lora (Learning on Random Access), which allows you to quickly and cheaply add new information and capabilities to the model;
AI ACT is an ambitious and controversial bill that can lead to dangerous consequences for the development and use of AI in Europe and abroad.
The law is trying to balance the goals of the promotion of AI and the solutions of risks associated with the use of AI. The act can also undermine the innovative potential and competitiveness of the European AI sector, create barriers to cooperation and exchange of knowledge with American partners, as well as contribute to the development of unsafe AI. The bill requires further discussion and clarification before its final adoption.