The Artificial Intelligence and Data Act (LIAD) proposed by the federal government last year, described as an “empty shell” that comes “too late,” is “not at all satisfactory,” according to experts.
While Europe and the United States seek to regulate the artificial intelligence industry as best as possible, Canada, through LIAD, also seeks to establish clear guidelines on the regulation of this technology, from its development to its uses through its commercialization.
However, since the bill was introduced in the House of Representatives in June 2022, artificial intelligence specialists, scientists and organizations have expressed strong concerns about the scope and content of this law.
One of the bill’s key measures is at the heart of their concerns: leaving the role of self-regulation to the artificial intelligence industry in Canada.
What is LIAD?
- The Artificial Intelligence and Data Act (AIDA) is the third part of Bill C-27.
- Bill C-27 primarily aims to reform the Canadian Digital Charter and strengthen the protection of Canadians’ personal information.
- The aim and “strength” of LIAD is to legally require companies operating in the AI space to assess the level of risk associated with their artificial intelligence tools based on their design.
- The federal government has stated that the bill will not come into force until 2025.
A colorful bill to support innovation?
It’s not satisfactory […], especially in systems that pose significant risks, says Andréanne Sabourin Laflamme, researcher at the International Observatory on the Societal Impacts of AI (OBVIA) and co-founder of the Digital and AI Ethics Laboratory (LEN. AI). Of course you can’t trust the companies [leurs] serve commercial objectives so that they can regulate themselves in a way that protects citizens.
According to this specialist, a concrete legislative measure is needed to regulate the activities of this sector. There will be a step backwards at the state level, she adds.
His criticism adds to concerns expressed by the League of Rights and Freedoms and 45 other organizations in a letter to the federal government dated September 25. (New window)
These groups, among other things, regret the lack of public consultation and consider it inappropriate to place the regulation of artificial intelligence entirely under the auspices of Innovation, Science and Economic Development Canada (ISED), whose mandate is to support the economic development of the industry artificial intelligence.
We feel that [le Canada] wants to monitor, but above all we don’t want to slow down innovation. […] The bill is somewhat influenced by this desire.
For its part, Innovation, Science and Economic Development Canada (ISED) maintains that Canadian law contains many important measures that will give Canadians confidence that their privacy is protected, that AI will be used responsibly, and that at the same time the bill will enable its use of innovation potential, a sign of a strong economy.
Contains AI-generated disinformation
While the Canadian Center for Cyber Security (CCC) identifies disinformation as a risk associated with generative artificial intelligence, several experts are sounding the alarm and encouraging the government to enact legislation to mitigate these risks.
Last May, during the C2 conference in Montreal, artificial intelligence expert Yoshua Bengio encouraged the federal government to introduce provisions into Canadian law to combat the threat of disinformation caused by, for example, images, videos and voices created by artificial intelligence Intelligence can be generated.
Open in full screen mode
Yoshua Bengio, here in Montreal during the C2 Montreal conference
Photo: (Christinne Muschi/The Canadian Press)
The federal government appears to have heeded the call.
On September 26, during the Standing Committee on Industry, Science and Technology’s (INDU) consideration of Bill C-27, Minister François-Philippe Champagne proposed amendments to the Artificial Intelligence and Data Act (LIAD), including amendments that Canadians can identify AI-generated content.
These changes require the industry to take measures to help the Canadian public correctly identify audiovisual content and text generated by AI.
There is a lack of transparency about the use of AI, so we don’t necessarily know whether generative AI even exists [créer] Images, videos or texts, explains Céline Castets-Renard, a law professor at the University of Ottawa and holder of the university’s research chair in globally responsible artificial intelligence.
What changes with generative AI is that a part of the population is not necessarily aware of the existence of these fakes and therefore for a part of the population there is always the idea: “If I see it, if it is said, then that’s why, “Because it’s true,” the researcher adds.
Minister François-Philippe Champagne responded to the accusation on September 27 by announcing a voluntary code of conduct for the artificial intelligence industry. Capacity is mentioned [d’acteurs malveillants] Produce realistic images and videos or pretend to be real people [de manière à] They enable deception on a scale that can harm important institutions, including the democratic system and the criminal justice system.
“Honestly, I don’t think it will change the situation,” says Lahcen Fatah, a doctoral student and researcher at the Interuniversity Center for Research on Science and Technology (CIRST).
The Canadian bill is ill-defined and much less ambitious than in Europe
The submission of the Canadian bill follows other initiatives to regulate artificial intelligence abroad, particularly in Europe with the EU AI law.
Through its legal framework, the European Union aims to regulate the design and use of artificial intelligence tools through an analysis and classification of the risk that these tools may pose. For example, AI tools that classify people based on a social score and AI biometric identification systems pose an unacceptable risk and will be banned. Some tools could also be classified as high or limited risk.
The law project [canadien] is much less ambitious than the European draft law, explains Frédérick Bruneault, professor at UQAM Media School and researcher at OBV.IA and LEN.IA.
Open in full screen mode
EU Commissioner for the Internal Market Thierry Breton during a press conference on artificial intelligence (AI) in Brussels on April 21, 2021. He is responsible for European policy regarding digital technology, including AI.
Photo: pool/afp via getty images / –
On the other hand, the bill [canadien] Frédérick Bruneault, professor at UQAM Media School and researcher at OBV.IA and LEN.IA, does very little to define the different AI systems and their uses.
In fact, the role of the artificial intelligence sector in Canadian law is to assess whether these tools have a high impact and pose a risk of harm or biased results, as we can read in the text of the bill.
However, the term “high incidence” is not even defined in Canadian law, regrets Ms. Sabourin-Laflamme. It is planned to clarify this term later by regulation.
The changes proposed by Minister Champagne in the parliamentary committee bring much-awaited clarification by experts on the meaning of high-incidence systems. This seven-point list includes, among other things, AI systems that could be used to make decisions on the labor market or systems that would be used in the healthcare system or in a court.
The European draft law, in turn, categorizes the risk levels and also presents a comprehensive list of artificial intelligence systems that are classified as particularly at risk by field of activity, such as justice, immigration, education, national security, etc.
Despite the bill’s shortcomings, progress is being made
Despite the criticism and despite the shortcomings in the draft law, researcher Céline Castets-Renard argues for a more pragmatic approach.
She says she disagrees with the proposal from some organizations to remove LIAD from Bill C-27 and subject it to revision.
I’d rather we move forward [avec le projet de loi actuel] and that messages can be delivered at this time during parliamentary debates and committee hearings.
According to this researcher, Minister Champagne appeared before the Standing Committee on Industry, Science and Technology [INDU] and made some changes [au projet de loi] to respond to the concerns of multiple stakeholders.
This takes into account the criticism expressed, she says.
When asked about the differences between the European and Canadian approaches to regulating AI, Ms. Castets-Renard believes that the Canadian bill has the advantage of being more flexible than the European initiative. Although this may be the right approach, the researcher has reservations.
Open in full screen mode
Céline Castets-Renard, Research Chair in Responsible Artificial Intelligence, University of Ottawa
Photo: Radio-Canada
Minister Champagne is trying to say that he wants agile regulations so that he doesn’t have to start from scratch every two years [à cause de la rapidité et de l’imprévisibilité du développement des technologies de l’IA], She says. But we still need to know what matters, what we mean by a “high incidence” system, real control, etc [par une véritable mise à] Execution of laws.
Due to the much more specific nature of the European text, in particular the classification of risk sectors, the researcher estimates that the risk of legal obsolescence is much higher in the European regulation. Because European law is much more precise, specific and restrictive, there is a risk that it will not be able to keep up with new innovations in the AI field.
Finally, Céline Castets-Renard addresses the criticism of corporate self-regulation triggered by the Canadian bill. This is not self-regulation, she makes clear. We make a commitment [juridique] the private individual for self-assessment.
However, the researcher assumes that Canadian law provides for concrete measures to develop controls and sanctions and actually enforce them, otherwise it will not deter anyone.