Then, the draft recommends programmers to prepare a risk management strategy should an "AI use disaster" happen as well as reporting and recovery mechanisms from the incident. However, the draft does not mention further in detail on the loss or disaster coming from the use of AI. The ethical guideline also includes privacy and humanity aspects, that any utilization of AI, including AI product and marketing development, needs to support innovation without replacing humans.
Just like Indonesia, the European Union has reached an agreement on what is called the "Artificial Intelligence Act" in a marathon discussion for 36 hours, which ended on Friday (8/12/2023). This reflected a very big political and commercial interest behind it. Europe is now in the frontline in terms of AI regulation, something that has been voiced by many renowned AI experts across the globe.
A completed detail on the achieved agreement was yet to be known. The European Parliament and members of the European Union also still have to add final words. But still, its framework has been defined straightforwardly so that it explains clearly how Europe wants to limit AI in the coming year, while creating a global standard.
The fundamental principles of this act is that the bigger an AI-related risk is, the tighter its regulation will be.
One of the hot debates was the discussion on regulating the strongest AI models, which is known as the "General Purposes AI (GPAI)", including the well-known chatbot ChatGPT. In recent weeks, Germany, France, and Italy have launched many tech lobbies to suggest erasing these models from the act and to only impose "autonomy regulation" onto those models. According to them, too-tight regulations will hold up innovation in Europe.
The lobby was not fully successful. GPAI models will have to comply with the act. It means every company behind them must, among many, be transparent about the data used to train the model. They must also follow the European intellectual property act and explain when the text, image, or voice is produced by AI.
If the new regulation is going to be officially passed in early 2024, this act is predictably in effect by the end of 2025 or 2026.