Peter Loewen is the director of the Munk School of Global Affairs & Public Policy. Gillian Hadfield is the director of the Schwartz Reisman Institute for Technology and Society.
Take a moment to imagine your next application for passport renewal. Rather than heading to a passport office, the government now allows you to apply online. If the passport office wants to follow up, instead of inviting you to visit in person, they send you a text, asking you to call a number.
Your call is connected immediately, and the agent is pleasant, speaking your language fluently with a slightly hard-to-identify accent. She asks you benign but interesting questions about your upcoming trip. At the end of the conversation, she lets you know that your application will be approved. You thank her enthusiastically, and she wishes you a safe journey.
This scenario is the kind of ideal government interaction that Canadians dream of. But what if that pleasant person who helped you was in fact a highly intelligent computer? Would it change our feelings about the level of service?
What if, instead of a passport application, you were interacting over a medical need; a question of a child support payment; or a request for employment insurance after a job loss? What if you were trying to speak to your local member of Parliament, asking them for assistance in a public matter or to express your opinion on an issue?
Whither AI’s potential to create a bright and hopeful future?
The reality is that governments are not far from having access to such services. Large language models – made famous by OpenAI’s ChatGPT – are improving at a breathtaking pace. Speech technology and voice recognition are developing at a similar rate. When the linguistic fluency of a language generator is combined with speech technology, the capacity exists to have a conversation with a computer that differs undetectably from one with a human. These digital agents can seamlessly incorporate the information they are receiving in real-time to make judgments that their owners – in this case, the government – program them to make. A world of digital agents who can replace public servants is closer than we think.
Should we help that world develop or hold it back? Of course, we would all rather deal with a real human who behaves like our imaginary artificial agent – quickly, empathetically and accurately. But for many users of government services, that’s not the right comparison case. Which would you prefer: The scenario we described above with an intelligent chatbot? Or the scenario in which you get a notification that you have to head to the passport office (which involves finding it and either securing an appointment or waiting in line) to talk to someone? Or the alternative – to wade through phone trees and hold music to talk to someone who may be at the end of a difficult day and not all that interested in solving your problem, not able to speak your language idiomatically, or unable to explain things in terms you understand?
AI can build you an ETF portfolio in less than 20 seconds – should you trust it?
In the passport example, the constraint on providing better services with an intelligent chatbot would not be the availability of workers to process passport decisions, but the capacity of this technology to scale up. Marginal costs are low here.
To be sure, there are challenges in using these technologies. Their advantages are only realized when more discretion is given to the digital agent. We would have to allow it to make decisions. How do we audit the decisions of robots? And who is accountable for the decisions which they make? What is the recourse when they make the wrong calls, or even do harm through their choices?
These are the kinds of choices governments will need to make about how they are willing to deploy digital agents to deliver services. There will come a moment in the future, perhaps the near future, where the cost of such agents will be low enough and the need for more government services will be high enough, that saying no to such machines will be impossible. Before that time comes, governments ought to decide what principles will guide their use.
There are multiple ways to achieve this. Governments could engage in substantial public consultations and hearings, with both experts and regular citizens. They could convene groups of citizens to deliberate over the principles and rules for the deployment of digital agents. They could run small, open trials, where citizen use of these technologies is entirely voluntary and the results of decisions are open to public scrutiny.
However governments decide to tackle these future choices, the decision must be made a priority now. The aspiration of democracies has long been a government for the people, but also by the people. And it’s up to democracies to decide if the same rule should apply to public services.