Ottawa rolls out voluntary AI code of conduct

MONTREAL – The federal government on Wednesday unfurled a voluntary code of conduct for generative AI, as anxiety persists over its proliferation and pace of development.

The self-imposed safeguards will “build safety and trust as the technology spreads,” Innovation Minister Francois-Philippe Champagne told a crowd of techies at the All In artificial intelligence conference in Montreal. So far, executives from a dozen Canadian companies and organizations have signed on, including BlackBerry, OpenText and Telus.

The document lays out measures businesses can take when working in advanced generative AI _ the algorithmic engine behind chatbots such as San Francisco-based OpenAI’s ChatGPT, which can spit out anything from term papers to psychotherapy.

The measures range from screening datasets for potential biases to assessing for potential “malicious use of the system.” They also align with six key principles that include equity, transparency and human oversight.

“We have witnessed technology advancing at what I would say is lightning speed,” Champagne said. “The mission we should give ourselves is to move from fear to opportunities.”

Amid both excitement and angst over the seemingly boundless scale of AI advancement, the federal government in June tabled a bill that adopts a more general approach to guardrails around machine learning and leaves details to a later date. Ottawa has said Bill C-27 will come into force no sooner than 2025.

Artificial intelligence pioneer Yoshua Bengio, who has stated the legislation puts Canada on the right path even as progress remains too slow, said public anxiety still hangs over the sector.

“Fear might be a starting point, but we need to act,” the Universite de Montreal professor said on Wednesday. “There is basically zero AI regulation that’s actually in effect right now,” he added in an interview.

Bengio, who in 2019 won the Turing Award – known as the Nobel Prize of the technology industry – called on governments to start tracking parts of the AI industry, such as purchases of graphics processing units. Specialized GPUs power the complex calculations made by large AI models.

“If you go to the bank and you make a cheque (for) a million dollars, they will have to report it to the government. If you buy a lot of GPUs, the government should be able to know so they can make sure they can enforce whatever regulations are going to be applied,” he said in an interview.

To ensure safe development and deployment of deep learning systems, Bengio said more government funding for research and security services is essential, on top of the creation of oversight bodies at the national and international level.

“We know that Russia and China are doing things with cyberattacks, but now they’re going to have these AI tools,” he said. “I don’t think, currently, our national security agencies have the equipment and the people to help protect us against this.”

In May, Bengio called on the federal government to begin rolling out rules immediately against certain threats, such as “counterfeiting humans” using AI-driven bots.

“Who’s going to decide what to do with it? Right now, those decisions are happening behind closed doors and in private hands,” he told the audience.

Criticized as vague by some legal experts, the Liberals’ Artificial Intelligence and Data Act lays out a framework for responsible AI development that aims for agility amid the technology’s constant evolution.

The law, part of a broader bill on consumer privacy and data protection, would ban “reckless and malicious” AI use, establish oversight by a commissioner and the industry minister and impose financial penalties. But definitions around key terms such as “high-impact AI systems” and specifics on how they would have to adhere to human rights laws would be developed down the line.

Even after it takes effect, the act would focus initially on education, guidelines and helping businesses comply voluntarily.