Listen to this article
Estimated 5 minutes
The audio version of this article is generated by AI-based technology. Mispronunciations can occur. We are working with our partners to continually review and improve the results.
As Canada looks ahead at six December byelections and the United States rolls through midterms, Elections Canada is sounding the alarm that artificial intelligence is challenging how voters can get accurate information during elections.
“AI is making it increasingly difficult not only for individuals to distinguish synthetic content from real images or voices,” Chief Electoral Officer Stéphane Perrault said to a House of Commons committee on Tuesday morning.
Over the past few weeks, the procedure and House affairs committee has been studying how AI is affecting and being used at Canada’s democratic institutions.
According to Perrault, Canadians are already using the technology to help make their choices on the ballot.
WATCH | Chief electoral officer on AI risks:
Voters face ‘challenge’ of evolving AI use in elections, says chief electoral officer
Canada’s Chief Electoral Officer Stéphane Perrault told a parliamentary committee studying artificial intelligence uses in Canada’s democracy that voters should be equipped to navigate the ‘very complex and rapidly evolving’ information environment. They ‘do not have the tools to know for sure whether something comes to them from an AI agent or a human being,’ said Perrault.
“Canadians are increasingly turning to AI chatbots that organize and summarize information from various sources to respond directly and precisely to their questions,” Perrault said.
“This includes information about the electoral process or comparisons of competing political platforms of candidates.”
Beyond making it hard to discern if information is real or fake, Perrault said AI is also making it difficult “to understand who is behind that information and why they are receiving it, and what sources are used by chatbots to formulate responses.”
The problem isn’t just being seen by Elections Canada. Perrault cited an incident from last week, when Spain’s main opposition party released a political ad that included a hyperrealistic AI-generated video — or deepfake — of the prime minister’s wife.
“AI can be used not only to create high-quality content very quickly and at little cost, but also to distribute and promote information whether that information is true or false, reach targeted audiences and amplify conversations, real or synthetic,” Perrault said.
Perrault also highlighted how bots and social media play into the equation, especially as AI can take advantage of both. “If you create a large number of fake accounts, parroting similar worldviews, then electors become caught in that fake consensus … that can undermine trust.”
The capabilities for information sharing are obvious, and so are the risks.Chief Electoral Officer Stéphane Perrault
Perrault repeated his 2024 recommendation that transparency labels should be legally required for synthetic (AI) content in electoral communications such as ads.
He noted that this recommendation is now within the scope of Bill C-34, which is being studied at the Canadian heritage committee. The EU adopted similar legislation in August for AI providers and operators.
“If the line between legitimate speech and co-ordinated manipulation is blurred, if the filtering by AI tools is not transparent, the ability of electors to make informed choices and to trust democratic outcomes is at risk,” Perrault said.
“A sound democratic process requires a minimum level of transparency regarding the information that feeds it.”
He also recommended chatbots be required to point to official authoritative sources when responding to questions about the voting process.
“Parliament should also consider whether this should include sources like party platforms or candidate publications when responding to queries regarding political positions,” Perrault said.
After hearing Perrault’s testimony, committee member and Conservative MP Blaine Calkins noted that previous committee witnesses said an AI agent had created a webpage consulted by voters during the Quebec election campaign.
Radio-Canada reported in September that an AI-generated site that contained errors was frequently cited by ChatGPT when asked questions about the Quebec election.
Élections Québec oversaw that election, but Elections Canada has six byelections next month in Ontario, Quebec and Saskatchewan. Perrault said the issue in Quebec was amplified by the fact that AI agents do not have access to “regular media,” as sites can block AI from accessing their data either on purpose for security reasons or unintentionally if websites believe the agent to be a data scraper.
“We are concerned about making sure that the information on the voting process is proper,” Perrault said. “We’re revisiting our content to make sure it is being accessed by AI agents, that it is providing information about the electoral process that comes from Elections Canada.”
Assistant deputy minister of Library and Archives Canada Jennifer Schofield also spoke about the possible risks of AI being used on Canadian archives, adding that the department needs to “be clear” about where the information is coming from.
“We work with historical records that … reflect views and language that are [no] longer used and may perpetuate biases if used to train AI models,” Schofield said.
“Without that distinction there is a risk that people might accept an AI-generated explanation as historical fact without knowing where the information came from or whether the evidence support it.”
Committee member and Liberal MP Arielle Kayabaga raised concerns about how AI can provide different answers when the same chatbot is asked the same question.
“I find the AI systems are not caught up on providing correct information, which can erode our democracy,” Kayabaga said. “Including when you are archiving information.”
Right now, according to Perrault, the biggest challenge is “equipping voters with the tools they need to navigate this very complex and rapidly evolving information environment.”
“As I said in my remarks, they right now do not have the tools to know for sure whether something comes to them from an AI agent or a human being,” Perrault said. “Whether it’s synthetic content or real content. And I think there needs to be more transparency so that they can navigate that.”