Leading AI expert urges swift passage of Canada's AI law, emphasizing need for immediate oversight amid advancing technology risks. (ParlVu)
A prominent Canadian expert in artificial intelligence, Yoshua Bengio, has stressed the importance of swiftly enacting Canada's proposed AI legislation. Bengio, who serves as the scientific director of Mila, Quebec's Artificial Intelligence Institute, expressed his support for the Artificial Intelligence and Data Act (AIDA) during a session with the House of Commons industry committee.
While some other experts have recommended either withdrawing AIDA for a comprehensive overhaul or conducting further public consultations, Bengio urged Parliament to proceed with passing the legislation, albeit with certain amendments. He emphasized that having an imperfect law with provisions for subsequent regulations is preferable to delaying enactment altogether.
Bengio underscored the urgency of AI oversight by suggesting that certain regulations should take effect immediately upon AIDA's enactment, rather than waiting for the typical two-year period during which detailed regulations are formulated by the government. For instance, he proposed that businesses be required to disclose information about AI systems above a certain capacity, including safety measures and security assessments. This, he argued, would shift the responsibility of ensuring safety onto developers rather than taxpayers.
The discussion also delved into the potential risks associated with advancing AI technologies. Bengio warned that the trajectory of current AI development poses significant societal risks even before the potential emergence of superhuman AI with artificial general intelligence capabilities, which experts estimate could occur within the next few decades. He emphasized the need for establishing regulatory frameworks to foster innovation while mitigating potential harms.
AIDA, as outlined, would govern categories of "high-impact" AI systems, including those involved in employment, individual services provision, biometric data processing, content moderation on online platforms, and law enforcement activities. Deploying AI systems likely to cause serious harm, whether physical, psychological, or economic, would be prohibited under the proposed legislation. Developers of high-impact systems would be required to implement measures to identify, assess, and mitigate potential risks or biased outcomes.
In addition to advocating for the immediate implementation of the registry, Bengio proposed two key amendments to AIDA. Firstly, he suggested broadening the definition of high-impact AI systems to include considerations of national security risks and societal threats. Secondly, he recommended that AI developers demonstrate the safety and security of their systems before full deployment, emphasizing the importance of early risk identification in the AI development lifecycle.
