The Artificial Intelligence and Data Act needs a reset

Have you recently been rejected when applying for a job or a loan, but can't get a satisfactory reason? The best you're told is that the decision was automated using “artificial intelligence." Maybe you're among the many people who find that social media platforms claiming to use AI recommender systems attuned to your interests push inflammatory but hard-to-resist material? Was your facial image among the billions that Clearview AI scraped from the web and in effect sold to law enforcement agencies as a potential suspect for matching in virtual line-ups? Or perhaps you're just worried that inscrutable algorithms harnessed to massive quantities of fine-grained personal information are contributing to greater societal unfairness and divisiveness. If so, the good news is that the federal government has begun to act on the growing public concern about the misuse of complex, data-driven algorithmic systems. In particular, it has recently introduced the Artificial Intelligence and Data Act, aiming to “prohibit certain conduct in relation to artificial intelligence systems that may result in serious harm to individuals or harm to their interests.” The bad news is that the federal government has written the Artificial Intelligence and Data Act (AIDA) too hastily. It skipped over the normal public consultation—and it shows. The government has also inappropriately inserted it into the Digital Charter Implementation Act, Bill C-27, the long overdue modernization of Canada's private sector privacy law. In the few places AIDA is specific, the scope is far too narrow to address the widely debated potential harms of AI systems, while the principal substance is left to be worked out later entirely at the discretion of the industry minister. As Bloc Québécois MP René Villemure correctly observed in the House of Commons debate on Nov. 4, AIDA “is more of a draft than a law,” and should be separated from the rest of Bill C-27 for substantial reworking. The challenge of ensuring that artificial intelligence—or more accurately, automated decision (and recommender) systems (ADS)—enhance human well-being in the context of the fast growing digital sector, is urgent and daunting. Fortunately, the government can learn much from other comparable jurisdictions, such as the European Union. Unfortunately, there is little evidence that the drafters paid close attention to developments elsewhere. Or perhaps they did but listened more to their business constituency, which fears that regulation might interfere with their now largely unfettered ability to exploit data and powerful algorithms at scale. The government's hasty, minimalist approach is not legitimate lawmaking. Nor is it likely to garner the public trust it claims to seek. To do so some fundamentals need to guide re-drafting, such as: Base the act on genuine public consultation. The EU's proposed Artificial Intelligence Act offers a valuable model. In the Canadian context, such consultations should include community advocates, researchers, lawyers, and representatives of equity-deserving populations most likely to be at risk. Ensure the independence of regulatory enforcement and oversight. The overall regulatory authority, currently envisioned as residing with the industry minister, properly belongs with an officer of Parliament, such as the privacy commissioner of Canada. Broaden the scope of relevant harms beyond just individuals to include collective harms. The proposed law too narrowly defines high-risk systems exclusively in terms of their quantifiable impacts on individuals. It should also include harms to groups and communities, such as those affected by biased training sets or “algorithmic amplification” in recommender systems. Broaden the scope of relevant algorithmic practices to focus on function and not just a narrow set of specific techniques. Presently, AIDA only seeks to regulate the use of a very narrow set of novel “AI” techniques, far more restricted than those of the EU's AI Act. AIDA, therefore, misses many of the potential harms such as those caused by algorithmic amplification, which can function with familiar techniques in use for decades. It would be clearer and more effective to re-frame the law to cover all algorithmic practices with the potential to affect human well-being, and then regulate those practices based on the degree of prospective harm. In short, rather than press ahead with this deeply flawed legislation, the Artificial Intelligence and Data Act should be sent back to the drawing board and done properly. Andrew Clement is a computer scientist and professor emeritus in the Faculty of Information at the University of Toronto. The Hill Times

The Artificial Intelligence and Data Act needs a reset

Rather than press ahead with this deeply flawed legislation, the Artificial Intelligence and Data Act should be sent back to the drawing board and done properly.

Legislation that keeps up with AI may require special committee with expert panel, says NDP critic

How to dismantle a privacy bomb

Liberals on collision course to entrench anti-Blackness

Legislation that keeps up with AI may require special committee with expert panel, says NDP critic

Canada needs a modern, national health data architecture

To address online harms, Canada must commit to digital media literacy

A cure for multiple sclerosis could be just around the corner

Renewable energy is the way to go, says letter writer

Supporting local broadcasters is critical for our democracy

To take AI to the next level, we need a dose of wisdom

Ethical AI can unlock social and economic opportunity for Canada

Artificial intelligence another avenue for government to thwart transparency

‘Data, privacy, trade’: ministers tasked with AI file contending with fast, fundamental changes

Why Canada needs to take the plunge on AI

News

Get free news updates

Opinion

Submit a guest opinion article

resources

Advertise

subscribe

get help

The Artificial Intelligence and Data Act needs a reset

Rather than press ahead with this deeply flawed legislation, the Artificial Intelligence and Data Act should be sent back to the drawing board and done properly.

Get The Weekend Point of View Newsletter