EU lawmakers struggle to reach agreement on AI rules. According to four sources with knowledge of the situation, European legislators have not yet reached consensus on several topics about new artificial intelligence regulations ahead of a major conference on Tuesday, putting any settlement delayed until December.
The European Parliament and the nations that make up the European Union must approve the draft AI regulations. In trilogies—meetings between parliament and EU states to hammer out the final texts of laws—they have thus far been considered three times.
On Tuesday, the day after EU legislators are set to debate their negotiating position regarding foundation models and high-risk AI systems, sources added, a fourth trilogue meeting will occur.
AI systems with foundation models, like OpenAI’s ChatGPT, are trained on massive data sets and can learn from fresh data to carry out various tasks.
To hasten the process, Spain, now in charge of the EU until December, has been pressing for an agreement and offering concessions.
They include a tiered system for regulating foundation models with more than 45 million users, according to a document that Reuters has seen.
Spain also wants extra requirements, such as frequent screening for possible vulnerabilities, for very capable foundation models (VCFM) like ChatGPT.
Smaller platforms, according to opponents, might be equally dangerous.
Before the fourth trilogy, Spain claimed to have discussed potential concessions with other EU nations. However, according to the sources, it is doubtful that a final agreement would be reached at that meeting.
Early in December, a fifth trilogue is scheduled to take place. If an agreement cannot be reached, talks may be postponed until early next year. The June elections for the European Parliament might further stall discussions.
The EU industry commissioner, Thierry Breton, and the co-rapporteurs for the AI Act, Dragoş Tudorache and Brando Benifei, have expressed optimism that the bill will pass by the end of the year.
In 2021, the EU began drafting the AI Act. In May this year, the European Parliament approved a draft law that set additional restrictions on face recognition, biometric monitoring, and other AI uses.
According to the ideas, AI technologies would be ranked low to unacceptable regarding perceived danger. Depending on the categorization, governments and businesses employing these technologies will have varied responsibilities.