Race against time for European AI law: Is an agreement emerging?

After a marathon session, a breakthrough in AI law appears to have been found. A concession to open source models must reconcile all parties. Time is of the essence to get everything done on time.

The various European institutions met yesterday at around 2 p.m. in Brussels to discuss the AI ​​law again. A bill has been on the table for several months, but the final details are causing great debate. Yesterday’s meeting did not go smoothly. A broken drinks machine and too little coffee caused great frustration, writes Reuters. After almost 24 hours of negotiations, those involved are close to reaching an agreement.

The Member States are a hindrance

Things looked good for the AI ​​law in the spring. In the first decisive vote in the European Parliament, the bill received a convincing majority. The law was allowed to go to the European Council, but from then on the engine stalled. Suddenly Germany, France and Italy, not least member states of the Union, began to obstruct each other. The subject of discussion is what about the so-called basic models on which many AI applications are built, must be done.

There were fears that the proposed rules would affect not only American companies but also European players in the AI ​​space. It is no coincidence that two of the largest European parties in the field of artificial intelligence, Mistral and Aleph Alpha, come from France and Germany. There was a feeling among European parliamentarians that a real technology lobby was active to undermine the proposal.

A counter-proposal from the opposing member states was to give companies the opportunity to regulate themselves. This was absolutely not an option for the European Commission. Companies would then act out of self-interest and not enforce the necessary standards for their products, it is said. There are industry initiatives to work together in a coordinated manner, such as the newly founded AI Alliance. However, you may be wondering how impactful this project can really be if one of the current key players, OpenAI, does not support it.

Solomon’s judgment

Yesterday’s meeting attempted to find a compromise that all parties could agree on. The basic principle remains fundamentally unchanged: The AI ​​law must prevent or at least restrict the use of systems whose use entails “systemic risks”. Companies that train and offer LLMs for commercial purposes must also communicate transparently about how they acquired and processed their data.

An exception to the rule could be open source models. Open source generally means that access to a model is completely free and the developers behind it therefore have no commercial motives. This would mean Mistral, Aleph Alpha and other European open source providers would be subject to less stringent rules unless Europe considers their technology to be high risk. It is not yet known what has emerged about other parts of the AI ​​law, such as the use of biometric identification in public spaces.

Race against the clock

In any case, time is not the European Union’s friend. The AI ​​law has been in the works for years, but technology is evolving so quickly that the law is constantly lagging behind. In the space of nine days, we saw AWS launch a new chatbot, ChatGPT celebrate its first anniversary, and Google launch its most powerful AI yet. The AI ​​storm isn’t going away any time soon.

There are also political arguments for finalizing the text of the law this year. There are elections for the European Parliament next year and all the election madness will probably temporarily push the AI ​​law into the background. As President of the European Council, Spain definitely wants to take this step. Belgium will take over on January 1st. If there is one thing in which Belgium is European champion, it is political compromises.