The government wants to ban officials from using Large Language Models (LLMs) for generative AI applications, such as Chat-GPT and Midjourney AI. This is due to concerns around possible privacy and copyright infringements that developers and users of these models may commit. The government is already working on an alternative plan.
This was reported by the Volkskrant on Thursday. Journalists from that newspaper were given access to a draft proposal from the outgoing State Secretary for Digitalization and Kingdom Relations (1). In the proposal, Van Huffelen shows concern about the protection of sensitive information that civil servants often have to handle. The reputation of these language models for privacy and information security has been repeatedly questioned over the past year (2). Other concerns relate to the copyright infringements these companies often commit when training their models with protected works. If the Council of Ministers approves the proposal, the ban is likely to be imposed on government agency suppliers as well.
Van Huffelen argues that there should not be a definitive ban and that there are "opportunities for useful use of generative AI within government," provided it is done in a "safe manner." Therefore, the secretary of state wants to allow the use of LLMs only where individual agreements have been made with vendors. Several pilot projects, to be completed by the end of next year, will provide guidelines for the responsible use of AI in government agencies and their contractors.
Earlier this month, TNO, SURF and the Netherlands Forensic Institute reported that they will jointly develop their own LLM using government funding (2). The consortium plans to launch the language model, GPT-NL, by the end of 2024. Asked about the possible deployment of GPT-NL with government agencies, a project leader at TNO says that no agreements have been made with the government at this time, although they have been in contact with several ministries.
(1) https://www.volkskrant.nl/tech/regering-bereidt-verbod-voor-op-gebruik-ai-software-door-ambtenaren~b8948cc7/
(2) https://privacy-web.nl/nieuws/consortium-werkt-aan-nederlands-alternatief-voor-chatgpt-dit-weten-we/.
