Is ChatGPT doing the Commission’s homework? European Parliament asks after robotic replies
The questions about AI use underscore the ongoing tensions between the two institutions.
European Parliament President Roberta Metsola questioned whether the European Union’s executive arm was using artificial intelligence to respond to lawmakers’ queries during a closed-door meeting of political leaders, two officials from the Parliament told POLITICO.
The queries came after a complaint from Renew Group Chair Valérie Hayer about the low quality and slow speed of written answers from the European Commission to lawmakers during a recent Conference of Presidents gathering. POLITICO has seen notes from the meeting.
The Commission has a legal duty to answer such questions but is not bound by time or quality.
Some members of the Parliament argue that the institution is routinely ignored or overlooked by the EU’s executive arm. Others have gone as far as to say that Commission President Ursula von der Leyen “hates” the Parliament, noting the tense relationship between the institutions due to an imbalance of power that favors the executive.
Metsola said during the meeting that she would gather examples to submit to the Commission, the two officials confirmed; one reported that a majority of group leaders agreed to raise the questions. Both were granted anonymity to speak freely about a confidential meeting.
Metsola’s spokesperson declined to comment on specifics from the meeting.
“What I can say more generally is that the quality of replies to written questions and the timeliness of the replies have been longstanding issues,” the spokesperson said.
In recent months, the Commission has encouraged the internal use of AI to boost productivity. Officials, including heads of unit and directors general, use a large language model tool known as GPT@EC — a walled-off version of ChatGPT — for a range of tasks including internal briefings, according to an EU official who was granted anonymity to speak freely about such use cases.
AI “is currently being tested, for instance, for the correct attribution of questions among Commission services or the identification of precedent replies that could be relevant to a specific case,” a spokesperson for the Commission said when asked about its use in written questions from the Parliament.
According to the Commission spokesperson, staff are not allowed to “replicate the output of a generative AI model in public documents, such as the creation of Commission texts, notably legally binding ones.”
The use of AI supports 30,000 staff and contract employees in the Commission to create drafts, summarize documents, brainstorm or generate software code, an official said last October.
Metsola will raise the issue of questions from the Parliament to Commissioner Maros Šefčovič, who is in charge of inter-institutional relations, the spokesperson for the Parliament president said.
No AI was used in the writing of this article.