Ever since the early days of Alexa, Siri, and Google Assistant, I’ve been fascinated by the idea of an AI-powered, government-vetted chat interface (voice or text) that could provide plain-language answers to specific questions about government services, eliminating the need to navigate and search through departmental websites.
While individual Government of Canada (GoC) departments and agencies have long experimented with simpler “programmatic” chatbots on their websites and social media, most of these never really took off. They were quite basic and required human intervention very early in the conversation (i.e. as soon as the question fell outside the basic preset parameters).
Fast-forward to 2024, and the use of AI chatbots for government service delivery is now a hot topic. Here in Canada, the government has updated its AI guidance resources and is working on an AI strategy that encourages departments to experiment with generative AI, including chatbots, to improve public services.
On the international front, there are numerous examples of government chatbots at various levels of government. Many have been around for years but have gone through multiple iterations. Noteworthy efforts include:
- Dubai: Uses the chatbot Rammas, a virtual employee of the Dubai Electricity and Water Authority (DEWA). It has been enhanced with ChatGPT, allowing it to handle a large volume of inquiries (over 9.6 million as of 2024).
- Government of Singapore: Uses its in-house Virtual Intelligent Chat Assistant (VICA) platform, which powers various agency chatbots (e.g. Captain Green) as well as the centralized AlphabotSG, and integrates with and learns from the various agency-specific chatbots.
- Australian Government: The Australian Taxation Office uses a chatbot named Alex, which assists with various taxation-related queries and improves user navigation on the agency’s website.
While these and many others not listed above tend to work quite well, it’s worth pointing out that most are still limited to individual departments or agencies of government rather than centralized government-wide efforts (with the exception of Singapore).
Current options for Canadians seeking government information via AI
As Canadian citizens become accustomed to interacting with increasingly intelligent AI chatbots developed by the private sector (general and task-specific), they will likely expect a similar capability on government-run websites.
When looking to use a chatbot to inquire about government services, Canadians already have several options:
- Using a free, publicly available generative AI chatbot, such as Perplexity, ChatGPT, CoPilot, Gemini, etc., and instructing it to only use the information found on gc.ca or canada.ca domains for its sources.
- Using a customized AI chatbot developed by an individual or organization for a specific purpose. An excellent example is TaxGPT, which Paul Craig developed to help Canadians with their tax filing questions.
- Developing a basic AI chatbot independently using OpenAI’s GPT builder or a similar easy-to-use tool. An example of this is my plain-language Canadian Government Services Navigator, which I created to assist Canadians in finding and using federal government services.
The elephant in the room
So why on earth should the Government of Canada (GoC) invest in creating or managing its own AI chatbot(s) when citizens can already easily use existing AI tools to access government information? What’s to stop government chatbots from facing the same fate as GoC website search tools, which have been largely abandoned in favour of people preferring to use Google?
The case for a GoC AI chatbot
When looking for information about government services, especially those that don’t involve providing sensitive private information, most Canadians simply want accurate and up-to-date responses as quickly as possible. However, they also need to trust the source. While I’ve tried to achieve this with my own chatbot by instructing it to only provide information from official government sources, this assumes that the sources themselves are up to date and that the chatbot is actually following instructions. The responsibility lies in the user to ensure that the responses are indeed reflective of the provided government sources.
Having AI chatbots reside directly on government websites or apps would add an extra layer of credibility. The gc.ca and canada.ca domains are powerful brand components encapsulating the trust and authority of the Government of Canada, which is generally high—though I’m setting aside the 2024 Edelman GoC trust barometer score of 49%; let’s leave that for another post.
Even though there is always a risk of hallucination and errors, which can be addressed in a disclaimer, a chatbot backed by the authority and credibility of the GoC (i.e. residing on canada.ca) will have some key advantages over third-party tools:
- It will be perceived as more trustworthy when it comes to the accuracy of information due to the oversight that would be required.
- It will be used by significantly more Canadians, especially those who don’t currently use AI tools to find government information.
- It will provide the GoC with valuable usage data, allowing for continual improvement in online service delivery. Note: I’m only referring to chatbots used for conveying publicly available information rather than handling sensitive personal information requests.
- It will be the only one that can ensure up-to-date information is being used, as it will be connected directly to official government data sources. Perhaps other AI tools can be blocked from scraping these.
- Ideally, it will seamlessly integrate with various services and databases across departments, providing a more cohesive user experience (something a third-party tool wouldn’t have access to).
The end of government websites
AI chatbots may end up being just a temporary blip in the evolution of online government service delivery, but I believe they will play a significant catalyst role in the shift away from government websites as we know them today.
Eventually, I can see us having access to a single, secure, multi-modal (text, voice, video) AI-based government virtual agent of some sort that offers a fully personalized experience and remembers our interactions. This means we’d need to feel comfortable sharing sensitive personal information with the AI, just like we already do with secure online government forms, websites, and applications. While we might not be at that comfort level yet, I believe it will happen sooner than we think. We’ll just need the “erase AI memory” button to always be a click away — and trust that it will actually work.
