Months after a parliamentary committee called on Ottawa to require artificial intelligence companies to get permission before using copyrighted works to train their models, the federal government has yet to settle on how news organizations should be compensated when their journalism is used by AI. 

“I don’t think we found a path for all of it,” Liberal MP Lisa Hepfner, chair of the House heritage committee, told iPolitics. 

Earlier this year, the federal government unveiled the long awaited AI Strategy, which aims to expand AI adoption and create thousands of jobs. Dubbed as “AI for All,” it tackled provisions related to privacy and data protections, but neglected any mentions of copyright. 

Lots of AI models use huge amounts of information to learn how to generate texts, images and other content, the committee report stated. A lot of them end up using copyrighted materials like news articles, books, music and images, which creators argue that some of these AI systems are using without their permission. 

To address that, the committee recommended under three principles: authorization, remuneration and transparency (ART), which essentially require companies to ask permission, pay for the work and disclose what was used. 

Paul Deegan, President and CEO of News Media Canada, said AI platforms and its training models pose a different set of challenges when it comes to news. 

“What we’re concerned about is that the AI companies are basically engaging in the brazen theft of our copyright protected material, and the government is doing nothing about it,” Deegan said in an interview with iPolitics. 

In the past week, several major media companies have registered to lobby the federal government on the issue, with one of their priorities being greater clarity in the Copyright Act around text and data mining. 

Text and data mining, or TDM, refers to the large-scale collection and analysis of information, including potentially copyrighted material, that can be used to train AI models. 

For news publishers, the concern is that their journalism can be used to build and improve commercial AI products without their permission or compensation. 

The problem extends beyond model training. AI-powered search tools can also generate answers based on information gathered from across the web, potentially giving users the information they are looking for without the need to click through to the original news source. 

Publishers can take steps to restrict automated systems from accessing their website, but Deegan said doing so presents its own trade-off: limiting access can also make their journalism harder to discover online. 

“If people are stealing your work, no one is going to subscribe… How does your employer pay you as a journalist if people aren’t subscribing?” Deegan said. 

In its response to the committee, Artificial Intelligence Minister Evan Solomon acknowledged concerns about copyrighted works being used to train AI systems without authorization or compensation, but stopped short of endorsing the committee’s proposed solution. 

Currently, the government said that it “takes note” of the recommendation, while the committee recommended requiring prior authorization before copyrighted works can be used for text and data mining or AI model development. 

iPolitics reached out to the Solomon’s office last week, but as of Thursday, still hadn’t received a response on how the government intends to respond to concerns around AI systems using copyright materials.