Cabinet had AI write speeches — Involvement of ChatGPT and Bard?

Several members of the Dutch government are utilizing generative artificial intelligence (AI) for communication, including Minister Rob Jetten, who applied AI for social media posts. The use of AI varies among cabinet members, with some having entire speeches written by algorithms. These developments have significant implications for the government and the public, especially within the framework of the European AI regulation.

Cabinet members have used AI for various writing activities, from brainstorming to final editing. This often involves chatbots, such as OpenAI's ChatGPT and Google's Gemini, which generate texts based on prior data. While these systems provide speed and variety in formulation, final editing and fact-checking remain crucial to ensure the quality of the final texts. The deployment of AI raises questions about the authenticity of speeches: who is speaking, the minister or the algorithm?

Concerns also arise with the use of AI for writing full speeches regarding hallucinations, where models generate incorrect information, and bias, where prejudices from training data are adopted. Transparency about the use of AI assistance is not a legal requirement but could enhance public trust, especially for significant policy announcements. The recent European AI regulation imposes obligations on governments regarding clarity and accountability in the use of AI.

To comply with GDPR requirements, it is essential that no unnecessary sensitive data is included in prompts for speechwriting tasks and that data is managed properly. The Dutch government is currently working on guidelines for the use of generative AI in the public domain and is adapting existing tools to ensure risks and rights are safeguarded. This includes attention to model documentation, the provenance of training data, and the necessary training for staff. Thus, the democratic responsibility remains with humans and not with the model.

Read the full article from AI Insider.