A few Thoughts on the Use of ChatGPT, GPT 3.5, GPT-4 and LLMs in Parliaments: Reflecting on the results of experimenting with LLMs in the parliamentarian context

Abstract

Starting in November 2022 with the free provision of ChatGPT, large language models (LLM) are now publicly available. This has significantly increased the number of publications which scopes potential changes caused by the application of generative artificial intelligence (AI) in various societal domains. The private use of AI and the economic integration of generative LLMs have increased significantly. However, for parliamentarians and parliamentary professionals, the technology often remains abstract, impacting everyday work only peripherally. Due to the special responsibility of parliaments, governments, and administrations as the organizational instances of society, and through the inherent legitimations by society itself, there is a necessity to examine the implications of the use of generative LLMs within these institutions and traditional structures as well as their influence on political system logic. The paper analyzes the responses that the generative LLMs GPT 3.5 and GPT 4 have provided via ChatGPT, based on the same input command (prompt) over different times. The responses help to assess how LLMs can be used in the parliamentary context, to reflect what dangers exist as well as to respond to the question on how a business model of an AI department in parliament might look like. Furthermore, it shall be explored whether there are fluctuations in the quality of the responses and how these should be evaluated against the backdrop of the need for accurate and precise workflows in parliamentary operations. Ultimately, the paper aims to provide an answer as to whether the application of ChatGPT together with the LLMs GPT-3.5 and GPT-4 could already deliver this necessary quality and consistency for the parliamentarian working environment today.