Between an AI rock and a hard place: balancing privacy and artificial innovation

Bill C-27 cleared a regulatory hurdle last week, but still has a long way to go before it ushers in updated privacy law and the new Artificial Intelligence and Data Act (AIDA). After months of debate at second reading, the House of Commons voted on April 24 to advance the bill to the House Industry and Technology Committee for further review and revision. This progress may satisfy public demand for more regulation to protect personal information and limit its use in digital technology such as artificial intelligence (AI). However, Parliament should tread carefully to avoid stifling innovation in this new and complex space. AI promises to collectively leapfrog society by generating new ideas, innovating new products and personalized services, and generally improving the lives of Canadians. Plot twist for those concerned about privacy: AI is also a powerful tool to fight against cybercrime and protect personal information, by detecting and thwarting online fraud and other cyberthreats. The Canadian government clearly recognizes the benefits of AI tools, underscoring its importance in its companion piece to AIDA, and more importantly, investing significantly in AI innovation in Canada. Laila Paszti a partner at Kirkland & Ellis LLP in their Bay Area technology and intellectual property transactions group. Photograph courtesy of Laila Paszti By creating impractical and technically arduous rules around the use of personal information, Parliament risks squandering its (and the public’s) investment in AI innovation. A roundly criticized part of Bill C-27 is the requirement for transparency—providing the public with sufficient information to understand the capabilities, limitations, and potential impacts of certain AI systems. This requires an AI developer to explain, in layman’s terms, how an AI tool—inspired by neurons and neural networks in the brain—employs millions and sometimes billions of parameters to perform intensive tasks, such as image recognition. To comply with this obligation, a developer may employ a less complex system lending to transparency at the cost of performance. A compelling example of this trade-off is a fraud detection system that explains how it detects fraudulent online activity using a less complex model at the cost of not correctly detecting fraudulent online activity. One hopes that the Standing Committee will carefully consider the merits of imposing the transparency obligation. I am not unsympathetic to the task ahead of the Industry and Technology Committee as they try to balance the desires of the public with broader societal objectives. They might be guided by how regulators addressed public concerns at the turn of the 20th century, when technology—the automobile—evoked a dystopian future. The first automobile manufacturers in the United Kingdom had to contend with the Locomotive Act of 1865, which required the operator of any mechanically propelled vehicle, including an automobile, to employ a flag man to carry a red flag and walk about 55 metres ahead of the vehicle. This distinctive feature of that law resulted in similar impractical laws directed at the nascent automobile industry at the time to be termed "Red Flag" laws. At least one U.K. manufacturer—the inventor of the Petro-cycle, a single-seat tricycle—abandoned production versions, publicly blaming the "Red Flag Act." On the other hand, in both Canada and the United States, there was limited—if any—application of "Red Flag" laws. The most colourful example of attempted "Red Flag" laws is found in Pennsylvania in 1896, when lawmakers voted for arguably one of the worst traffic laws, ultimately vetoed by the governor. Upon chance encounters with livestock, drivers in Pennsylvania avoided having to immediately stop the vehicle, disassemble it, and then conceal the various components out of sight behind nearby bushes until the livestock were out of sight. Canada avoided the majority of such laws, with limited exception. Prince Edward Island's provincial legislature banned automobiles until 1913, when the law was repealed. Thereafter, to quell public unease, a revised law permitted automobiles on public roads only if the driver was able to secure the consent of 75 per cent of the people living along the road or municipality in which the driver wished to motor. Most other provinces adopted regulation that simply set a speed limit, with Ontario setting an initial speed limit of 24 km/h in 1903, increasing highways speeds to 40 km/h in 1923. This type of clear rule setting allowed automobile manufacturers to innovate freely and to consider practical challenges to the technology, such as better braking systems, rather than developing a car that could be quickly dissembled to comply with proposed Pennsylvania legislation. AI innovation and adoption is a path—if not a highway—to wide-sweeping improvement in many facets of Canadian society. As the House committee begins its task of balancing public and industry objectives in Bill C-27, the lessons from the annals of innovation should not be forgotten: broad, unclear, and impractical laws are "Red Flags" impeding our journey on the AI highway. This summary provides high-level, general information and is not intended to provide legal advice. Laila Paszti is a lawyer and professional engineer, with graduate degrees in machine learning. She is currently a partner at Kirkland & Ellis LLP in their Bay Area technology and intellectual property transactions group. The Hill Times