Build Canada’s AI ‘experiment’ reignites debate over political communication ethics
As the third-party advocacy group Build Canada becomes the second politically adjacent organization in as many weeks to publicly experiment with artificial intelligence-generated communications, technology experts and digital strategists say the technology’s rapid evolution demands stronger ethical standards as legislation alone is unlikely to keep pace.
Just over two weeks after the Conservative Party released a partially AI-generated social media video criticizing reports of a "technical recession," Build Canada—a non-partisan "civic movement" whose backers include major tech entrepreneurs—unveiled its own “experiment” with the technology, featuring a fully generated “general message” to “Build Canada – Actually.”
Opening on a ribbon-cutting ceremony for a non-specified transit construction site, the roughly minute-long video depicts a trio of artificially generated politicians shaking hands and smiling for the artificially generated press before cutting to a young boy waiting in the crowd and a time-lapse sequence as he grows into an old man while the construction site remains unfinished, and a large "Coming Soon" billboard fades and collapses in front of it.
Like the Conservatives’ earlier video, the response to Build Canada's video followed partisan lines, from condemnations of the choice to use the technology over human actors and critiques of the quality as “slop” from the left, to the more positive responses from the right, including reposts from Conservative Leader Pierre Poilievre (Battle River–Crowfoot, Alta.), and former party leader Erin O’Toole, who said the video shared an “uncanny similarity to the history of Lakeshore East GO Train expansions that have been made dozens of times since I was in high school.”
In an interview with The Hill Times, Build Canada CEO Lucy Hargreaves emphasized that her organization’s video was neither a response to the Conservatives’ video, nor aimed at any particular party or level of government.
“All levels of government in this country love to announce things, they’re really great at that, but they have a really hard time following and delivering results,” Hargreaves said, adding that the video is not “anti-Carney” or “anti-insert whatever government.”
“It's something that governments at all levels struggle with in Canada,” Hargreaves explained. "It's a basic message.”
While Hargreaves said she expected “some mixed reactions” and some level of controversy over the new medium crowding out the intended message, “getting mixed reactions is nothing new to us.”
“Since we started Build Canada in early 2025, we've always been testing new, innovative ways to convey our message to Canadians, and so the way we view this is really as an experiment,” said Hargreaves, a former federal Liberal cabinet staffer. “It's a new medium and a new way to reach people and to share our message of growth and what is holding Canada back … it’s just part of our continuum of trying to test new ways to reach Canadians.”
Hargreaves said that the range of feedback on the video’s format is “kind of to be expected when any organization tries new things and new ideas,” but noted it had nonetheless generated “incredible traction” with more than 250,000 views on X alone.
While Build Canada will continue producing content using conventional filming techniques—including its June 24 World Cup-themed Home Turf video featuring real actors and locations—Hargreaves said the technology will allow her organization and others like it to convey their messages at a fraction of the cost.
Declining to disclose the video's exact budget or the software used to create it, Hargreaves estimated that most organizations would spend between $6,000 and $10,000 to produce a comparable AI-generated video. Producing the same concept with actors, sets, and visual effects, she said, would likely cost between $250,000 and $300,000.
"What this technology is doing is enabling many different organizations, groups, and individuals to convey their message to Canadians at a fraction of the cost," Hargreaves said. "That's really the democratization of communications."
Digital strategists say political communicators need a ‘higher ethical standard’ for AI generation
Liberal digital strategist Nathaniel Arfin, who called the Build Canada video “gross,” told The Hill Times that, beyond his critiques of the quality of the video—and remaining unconvinced it couldn’t have been better produced with real environments and human actors with AI enhancements implemented in post-production—his primary concern is the “exponential improvements” the technology has made in the past few years.
Arfin, a software and technology expert with a background in enterprise AI deployment and IT security, pointed to the so-called “Will Smith eating spaghetti” benchmark—which compares a now-infamous AI-generated video of the actor from March 2023 and examples of the same video generated since—as a demonstration of how quickly the technology has improved and the need for further guardrails on how it's used before “it starts to get weird.”
"If we don't consider the ethical framework around using these things to generate people, identifiable or not, especially in political communications, we're going to have a huge problem," said Arfin, founder of digital campaign firm Chester Hill Solutions.
While he acknowledged that AI is already widely used in political campaigns at every level in less-significant ways, such as minor digital touchups and alterations to images, or for translating audio, its use to generate the entire video—or the majority of it in the case of the Conservatives’ previous video—enters into new ethically fraught territory, moving from simple enhancement to replacement of human involvement, Arfin said.
“Instead of using it to skip some work or enhance some steps, you're taking away the human involvement, and it creates more distrust in what you're seeing and feeling about it,” Arfin said. “That to me is where it crosses a really, really big line.”
It’s too late to “unplug” the technology from political communications, Arfin said, adding he doesn’t think legislation to further limit its use in political campaigns or the private market would be an effective approach, “but we really do have to hold ourselves to that higher standard.”
“I think it would make sense to limit the generation of human beings or other anthropomorphic characters to deliver a political message," Arfin said. “I think we have an opportunity to show leadership and recognize that this is not a use of AI that is aligned with our values and does not represent responsible use of AI.”
Bill C-25, the Liberals' recently adopted Strong and Free Elections Act, introduced by Government House Leader Steven MacKinnon (Gatineau, Que.) on March 26, modernizes the Canada Elections Act to address digital threats to elections like AI deepfakes, as well as criminalizing activities to “knowingly spread false or misleading information” about elections or the voting process, with the intent to disrupt an election or affect its results.
Speaking before the Senate Legal and Constitutional Affairs Committee on May 27, MacKinnon said that while he finds deepfakes impersonating politicians troubling, the legislation allows for satirical content.
“When citizens cannot trust in what they are seeing, then we have a fundamental problem in our democracy,” MacKinnon said. “I don’t mind being made fun of. That goes with the job. But having someone say I said something that I obviously did not say is another matter.”
However, Arfin said that as the technology advances, that distinction will matter less and less. Unlike an actor portraying a politician on This Hour Has 22 Minutes, generative AI can recreate someone's appearance and voice as accurately as possible—and is becoming more convincing with each new model, he said.
"Unless you're specifically asking for a parodied version and making all of those things clear and transparent, there's no real way to make it clear to someone that that is not him," Arfin said. "It's completely subjective, and many people—especially when they're being encouraged by others acting in bad faith—will treat it as real. Those things can and do spiral out of control."
Even in instances where the videos are generating “random” people, Arfin argued that it still raises many of the same concerns.
"These aren't deepfakes in that they're not replicating a specific human being, but they're absolutely replicating an amalgam of them," Arfin said, adding that because AI image models are trained on enormous datasets of existing photographs and artwork, there is always the possibility they could inadvertently reproduce real people or reinforce existing demographic biases.
Megan Buttle, founder of AI advisory firm Real Signal, agreed that no AI system is truly neutral.
"There's absolutely no such thing as an unbiased AI," Buttle told The Hill Times. "They're all trained by individuals, and those individuals' biases shape their training, whether known or not. Even the most sophisticated predictive modelling engines and content-creation tools are biased. It's baked in from the beginning, and it's impossible to remove."
Even if developers could eliminate every trace of bias during training, she said, new biases would emerge through use—a phenomenon she described as the "mirror effect."
"A model isn't frozen in time," Buttle said. "There's ongoing user bias every time you use it. It continues to learn based on the questions you ask, the tone you use, and the terminology you're inserting into it."
Like Arfin, Buttle said there should remain room for satire involving AI-generated politicians. However, she questioned whether current disclosure requirements are sufficient to distinguish between legal and ethical uses.
"Legally, right now, we have a loophole that exists for parody and satire, as long as a reasonable person understands it to be a joke or political commentary," Buttle said. "Our rational brain can digest that, but particularly when messages are designed to hit Canadians emotionally, a tiny AI label is not good enough to moderate that."
Buttle said that beyond videos intended to embarrass or ridicule politicians, the greater concern will be increasingly convincing AI-generated content designed to provoke emotional responses while skirting the line between "parody" and authentic deception.
However, while she said she also doesn't believe that more legislation will be effective in countering the unethical use of AI-generated communication, Buttle also expects that the phenomenon will elicit, and has elicited, a greater counter-demand in the market for "the real thing."
As AI-generated political communications become cheaper, faster, and increasingly indistinguishable from reality, both Arfin and Buttle said the challenge facing political actors is no longer whether the technology will be used, but how responsibly they choose to use it.
“We'll see those who lean into AI-generated content, but I also think we'll see others do the opposite by really leaning into putting their own face out there and doing it in high volume,” Buttle said. “AI will remain part of our political advertising landscape, but it will also encourage others to invest in creating real, authentic content as a direct contrast.”
The Hill Times