A number of Canadian technology companies are signing on to the federal government’s new code of conduct for generative artificial intelligence, a set of voluntary guidelines aiming to limit harm until formal AI regulation comes into force in two years.
Blackberry BB-T, OpenText Corporation OTEX-T, Ada-AI, Cohere and Coveo CVO-T will be among the first signatories of the Guardrails for Generative AI Code of Practice, committing to a range of measures on safety, equity and transparency. The finalized guidelines were announced by Industry Minister François-Philippe Champagne at an AI conference Wednesday in Montreal, after a series of stakeholder consultations conducted by Innovation, Science and Economic Development Canada over the summer.
Ottawa says the code applies to all firms developing or managing generative AI, which draws on large datasets of text, images or audio to produce new content. The technology has been widely criticized for the risks it poses to privacy, equity and copyright.
ISED says the measures will help companies to prepare their processes and products before formal regulation takes effect. The department has projected that the provisions of The Artificial Intelligence and Data Act will come into force no sooner than 2025. The act is currently being considered by the House of Commons standing committee on technology and industry.
Governments around the world are speeding to develop AI policies that balance innovation and consumer protection, even as that technology is rapidly evolving. ChatGPT, the popular generative AI tool, was only made available to the public last November, five months after AIDA was tabled.
The new code of conduct lists 18 measures that apply variously depending on the capabilities of the generative AI. Broadly, companies who sign the guidelines will commit to testing broadly to identifying security vulnerabilities, beefing up cybersecurity protections and assessing how generative AI systems could become biased with low-quality datasets.
The code emphasizes transparency. It instructs companies to develop and implement a method to detect content generated (watermarking) and prominently identify systems that could be mistaken for humans as AI systems.
In a new addition since publishing a draft of the guidelines in August, ISED has asked AI developers to publish a description of the types of data used train their system. Blair Attard-Frost, a PhD candidate studying the governance of artificial intelligence at the University of Toronto’s faculty of information, said this could reference questions about how to prevent copyright infringement, but that it doesn’t go far enough to make that clear.
In submissions to ISED as part of the consultation for the code, groups such as the Screen Composers Guild of Canada urged the government to include provisions that specifically protect artists, musicians and writers.
The measures resemble those announced by the U.S. in July, and which were signed by large technology companies including Amazon.com, Inc. AZAM-Q, Alphabet Inc. GOOG-Q, Meta Platforms, Inc. META-Q (formerly Facebook) and Microsoft Corporation.
The measures were developed through a series of consultations with academia, civil society, Canada’s AI research institutes and industry. But while the code of conduct is only intended to be “sufficiently robust” as to span the gap before the AI act comes into force, according to ISED, some industry experts have criticized the closed nature of the consultations, saying that the public should have been consulted.
Natasha Tusikov, an associate professor of criminology in the department of social science at York University, called the process of developing the guidelines “rushed” and “relatively opaque.” Voluntary agreements amongst industry actors “are not effective when industry’s enforcement efforts are expected to go against their commercial interests,” Prof. Tusikov added.
Michael Geist, law professor and Canada Research Chair in internet and e-commerce law at the University of Ottawa, said the code is “a non-binding document that adds little to the global conversation on addressing generative AI.”
Shopify Inc. SHOP-T chief executive officer Tobi Lütke in a tweet called the code a “case of EFRAID” and said he won’t support it.
“We don’t need more referees in Canada. We need more builders. Let other countries regulate while we take the more courageous path and say ‘come build here,’” Mr. Lütke said.