Governments worldwide have been trying to use chatbots to make their services and advice more automated. Initially, these chatbots were very basic and offered limited conversational abilities. Colin van Noordt, a researcher in the Netherlands who studies AI in government, says that the early chatbots were simpler with limited conversational abilities. However, In the past two years, generative AI technology has brought back the idea of making public services more efficient. Now, human-like advisors can provide 24/7 support on issues such as benefits, taxes, and other government services to the public.
Generative AI is advanced enough to offer human-like responses. When it gets trained on enough quality data, it can theoretically handle a wide range of questions about government services. Despite this, generative, generative AI often makes mistakes or gives odd answers, which is called hallucinations.
Trials and Challenges in the UK
The UK’s Government Digital Service (GDS) has been experimenting with a ChatGPT-based chatbot named GOV.UK Chat. This chatbot aims to answer citizens’ questions across a broad spectrum of government services. According to a blog post about the early findings of their trials, nearly 70% of participants found the responses useful. However, there were issues with the system occasionally generating incorrect information and presenting it as fact. The blog also highlighted concerns about misplaced confidence in a system that can sometimes be wrong.
The GDS said that the answers given were not accurate enough for a site like GOV.UK, where getting factual accuracy is very important. They are rapidly making changes to fix these accuracy and reliability problems.
Portugal’s AI-Driven Justice Practical Guide
In 2023, Portugal introduced a new AI chatbot called the Justice Practical Guide. This chatbot helps people with basic questions about topics like marriage and divorce. The €1.3 million project, funded by the European Union’s Recovery and Resilience Facility (RRF), uses OpenAI’s GPT-4.0 language model. In the first 14 months, it answered 28,608 questions.
The chatbot performed a good job with simple questions but had trouble with more complicated ones, such as legal questions about starting a company for minors. A source from the Ministry of Justice said that although wrong answers are rare, it is still hard to make sure the chatbot is always reliable. The source said they hope these issues will be fixed by making the answers more confident and reliable.
Expert Caution and Ethical Concerns
Despite the advancements, many experts urge caution in deploying AI chatbots. Colin van Noordt warns that problems arise when chatbots are used to replace people and reduce costs. He advocates for using chatbots as an additional service, a quick way to find information. Sven Nyholm, a professor of the ethics of artificial intelligence at Munich’s Ludwig Maximilians University, underscores the problem of accountability.
Despite the advancements in AI chatbots, many experts suggest being careful about using them. Some point out that issues occur when chatbots replace human jobs and reduce costs. It is recommended to use chatbots as an additional service for finding information quickly. Others highlight the problem of who is responsible when things go wrong with AI.
A chatbot is not interchangeable with a civil servant. A Human being can be accountable and morally responsible for their actions, whereas AI chatbots cannot be held accountable for what they do. Public administration demands accountability, which inherently requires human involvement.
Additionally, there are concerns about the reliability of newer chatbots, which can create an illusion of intelligence and creativity. These chatbots may occasionally make silly or dangerous mistakes, especially if people place undue reliance on their recommendations.
Estonia’s NLP-Based Approach
Estonia is trying an alternative approach to using chatbots. Since the 1990s, the country has been a pioneer in digitizing public services. They are now creating a series of chatbots called Burokratt. Unlike chatbots like ChatGPT which rely on large language models (LLMs), Burokratt uses Natural Language Processing (NLP) technology. This means it breaks down user requests into small segments, identifies keywords, and figures out what the user wants. If Burokratt can’t answer a question, a real customer support agent will take over the chat.
The head of Estonia’s Information System Authority says that their chatbots might not sound as natural or understand language subtleties as well, but they are less likely to give wrong answers. Another expert points out that chatbots based on large language models can have better conversations and more detailed answers, but they are harder to control and may be less consistent.
In conclusion, the use of AI in government chatbots is evolving rapidly. While generative AI offers promising potential for more efficient and responsive public services, challenges related to accuracy, reliability, and accountability remain significant. As countries like the UK and Portugal test these technologies, and Estonia continues to refine its NLP-based systems, the future of AI-driven government services will likely involve a balanced approach that combines technological advancements with human supervision.