EU AI Regulation adopted

On the night of Saturday, after over 36 hours of negotiation, the pieces finally fell into place for the new AI regulation. Here are the most significant changes, including .dem relevant to the education sector.

The European Parliament has decided to use the OECD's proposal to define artificial intelligence instead of the previously proposed definition. A Danish translation of the definition is not yet available:

"An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in autonomy and adaptiveness after deployment."

It has also been agreed that the Regulation cannot be used outside the scope of EU law and should under no circumstances affect Member States' competencies for national security.

In addition, the main changes to the final AI regulation are as follows:

It is generally forbidden to use facial recognition in public spaces. Still, it is now open for member states to use the technology to search for victims and suspects of serious crimes and to try to prevent terrorist attacks. The use of emotion recognition systems in workplaces and educational institutions is also prohibited but may sometimes be used at border crossings. Also, one must refrain from systematically scraping facial images from the Internet or surveillance footage to create facial recognition databases. AI systems classified as high-risk must now undergo a mandatory assessment of their impact on fundamental rights.

Transparency is required for general-purpose models, such as the model ChatGPT is based on, before they are marketed. The agreement introduces a tiered model to keep an extra eye on the most potent AI models (e.g., GPT-4, Google's Gemini, and future models).

When using AI systems such as chatbots, users must be aware that they are interacting with a machine. Deep fakes and other AI-generated content must be clearly labeled so that users are informed about what they see. In addition, providers must design systems in such a way that synthetic audio, video, text, and image content is marked in a machine-readable format (a form of watermark) that can be detected as artificially generated or manipulated.

Special and stricter rules are made for "high-impact general-purpose" AI models that may cause systemic risk in the future, like for high-risk AI systems. A new office will be established within the European Commission to enforce the rules governing these models.

However, it is essential to understand that only one political agreement has been reached. The details of how the companies will comply with the individual parts still need to be put in place. There will be a two-year phase-in period before a fine can be imposed.

The provisional agreement on the new EU legislation on artificial intelligence sets different levels of fines for companies that do not comply with the law: if a company uses AI technologies that are banned, it can be fined up to €35 million or 7% of their total global annual turnover, whichever is higher. For violations of other obligations of the AI Act, for example, not following the rules of high-risk AI systems, the fine can be up to €15 million or 3% of global turnover. If a company provides incorrect information about its AI systems, it can be fined up to €7.5 million or 1.5% of global revenue. Important to note is that the agreement takes into account smaller companies and startups. This means that if they violate the provisions of the law, they could potentially receive lower fines, which are more appropriate to their size and financial capacity.