Governments around the world are exploring the use of artificial intelligence (AI) chatbots to automate services and provide advice to citizens. With the emergence of generative AI, there is a renewed vision of efficient public service where human-like advisors can respond to questions about government services at all hours.
However, the use of generative AI chatbots like ChatGPT has raised concerns about accuracy and reliability. In the UK, the Government Digital Service (GDS) tested a ChatGPT-based chatbot called GOV.UK Chat, which showed promising results but also generated incorrect information in some cases.
Portugal has also introduced an AI-driven chatbot called the Justice Practical Guide, funded by the European Union’s Recovery and Resilience Facility. While the chatbot has been successful in answering basic questions, it still lacks trustworthiness in more complex scenarios.
Experts like Colin van Noordt and Sven Nyholm caution against relying too heavily on AI chatbots to replace human civil servants. They emphasize the importance of accountability and reliability in public administration, which AI chatbots may not be able to provide.
Estonia, known for its advanced digital services, is developing chatbots for state services under the name Bürokratt. Unlike ChatGPT, Estonia’s chatbots use Natural Language Processing (NLP) technology, which allows for more control and transparency in how the chatbots operate.
While NLP-based chatbots may not have the conversational quality of LLM-based chatbots like ChatGPT, they are less likely to provide wrong or misleading answers. The balance between conversational quality and control is a key consideration in the development of AI chatbots for public services.