How ready are we for deep fakes? Researchers call for changes to AI to protect women

In the photo, a blonde woman in a bikini is standing on the beach. Then a line flashes on the screen showing her naked image.

“Use AI to strip a girl for free!” reads the description on the site.

Although it says consent is required, it takes a few clicks to upload a picture and see the person in it undress.

Since last summer, the number of sites with publicly available AI imaging tools has grown, gaining millions of views, and cases of AI-doctored photos of underage girls have already been shared by high school students. London, Ont.and Winnipeg. No charges were filed in either case.

But a criminal case has been opened in Quebec for misuse of the technology. Last year, a Sherbrooke man in the Eastern Townships was jailed for three years for making at least seven deeply fake videos depicting child pornography.

According to intellectual property lawyer Gaspard Petit, Quebec, like the rest of the country, may not be ready to deal with this advanced AI technology.

As Ottawa moves to regulate harmful online content, researchers are calling for more diversity and transparency to ensure women aren't targeted by technology without their consent.

According to Petit, he is keeping a close eye on the development of AI technology, which is still developing.

“I think there's a general consensus in Quebec, in Canada as a whole, that we're not ready,” he said.

Gaspard said Quebec statutes and laws already have protections in place to protect people's privacy and reputation.

He said barefaced cases can fall into a legal gray area where it's not always clear whether the person producing or distributing them can be prosecuted — something Canadian lawmakers are debating how to improve.

One problem, Gaspard says, is that the onus is on the victim to prove they were harmed and who is responsible, and then sue if they can.

But he says the biggest problem is preventing the images from spreading in the first place.

Correcting gender inequality

Dunyan Lin, a researcher at the MILA Institute for Artificial Intelligence in Montreal, studies the relationship between neuroscience and AI. She says these deep fakes are “a perfect example of the absence of women in the decision-making process.”

As a result, he says, these companies have blind spots in their thinking about how the technology will be used “after mass commercialization.”

Affecting machinesDeveloped at Concordia University, the project seeks to close the gender gap by promoting women's work in AI and STEM.

Lindsay Rogers, a knowledge mobilization consultant at Concordia's Applied AI Institute, is one of the people involved in the project.

“Gender diversity is critical to having AI systems that are representative of the population that uses them,” she said.

“It's not just about numbers like AI (lab) hiring more women or non-binary people in the room, it's about creating a culture and atmosphere where they can succeed and work well and be valuable team members,” she said. In tech, women make up about a quarter of the industry's workforce, barely growing over the past two decades.

Ethical training, other solutions

Along with stricter regulations and public hearings on the use of artificial intelligence, Lin said, mandatory ethics training would help AI developers better understand how the technology could be used by the public.

Banning sites that use deepfake technology is also an option, but experts like Sasha Lucchioni, a Montreal-based researcher at AI company Hugging Face, point to tools that allow users to be banned in their country of residence.

Other technical solutions are also on the table, such as rendering images unusable for AI models, but none of these solutions address the problem at its core, Luccioni says.

The root cause of the problem, she says, is how people decide to use the technology available, including using it to objectify women's bodies.

He says that the solution to this problem is to educate and inform the people.