Quebec’s Bibliothèque et Archives nationales du Québec (BAnQ) is taking a significant step towards creating a comprehensive database that will encompass cultural and governmental content in French and Indigenous languages. This initiative aims to improve artificial intelligence’s comprehension of Quebec’s unique society, culture, and languages. Following a successful feasibility study, BAnQ has entered the experimental phase of this ambitious project, addressing the pressing need for more relevant data in AI training systems.
Addressing AI Limitations
Generative AI systems have often faced criticism for their inability to accurately reflect Quebec’s societal landscape, primarily due to the scarcity of localized data. Valérie D’Amour, who headed the feasibility study, emphasised the importance of collaborative input from cultural stakeholders, data owners, and providers to validate the project’s potential. “All scenarios are a little bit on the table right now,” she remarked, highlighting the exploratory nature of the initiative.
BAnQ’s president and CEO, Marie Grégoire, reinforced the goal of the project: to ensure that AI systems better represent the diversity of Quebec’s culture. She stated, “That means having Quebec references, whether in small models or large models, whether they come from research or from the business community.” By starting with its own collections, BAnQ aims to develop a robust foundation before considering additional data sources.
Learning from Global Examples
Similar efforts have been observed globally, notably in Sweden, where extensive collections of Nordic-language texts have been compiled to enhance generative AI models for Scandinavian languages. This international precedent highlights the importance of localised content in AI development and sets a benchmark for BAnQ’s own initiative.
The impetus for this project stems from a recommendation in a report by Quebec’s innovation council, which identified the limited availability of Quebec-centric data in existing AI datasets as a significant barrier to accurate representation. Destiny Tchéhouali, a prominent researcher and co-holder of a chair focused on French-language AI, pointed out that Quebec’s cultural contributions are often underrepresented in the AI landscape. He warned of the dangers of perpetuating linguistic and cultural biases, particularly concerning Indigenous peoples.
Cultural and Copyright Concerns
As BAnQ develops its database, concerns surrounding copyright and the protection of creative works loom large. Grégoire contended that the proposed platform could offer artists enhanced protection compared to the current landscape, which she described as “the Wild West” of data harvesting. The initiative could establish a centralized system that ensures creators are fairly compensated for their contributions, addressing long-standing frustrations within the cultural sector.
However, some artists voice apprehension about the potential consequences of sharing their work for AI training. Maxime Harvey, a postdoctoral researcher, noted that while artists might receive income, they risk contributing to a system that could ultimately jeopardise their livelihoods. “Even if artists earn income from it, they are still feeding the beast that will eventually be used to replace contracts they may lose because of AI,” Harvey cautioned.
Future Outlook and Funding
BAnQ aims to have the platform operational by 2029, with ongoing assessments of the timeline following the experimental phase. The initiative is projected to require nearly $10.5 million over five years, covering both operating and capital expenses. To support this endeavour, BAnQ received $340,000 from the Quebec government for the feasibility study and an additional $750,000 for the subsequent 12-month experimentation phase.
Why it Matters
This initiative represents not just a technological advancement but also a cultural necessity. By prioritising the inclusion of Quebec’s diverse voices and languages in AI systems, BAnQ is taking pivotal steps to ensure that future technology respects and reflects the province’s unique heritage. As AI continues to evolve, ensuring equitable representation in its training datasets could help prevent the marginalisation of local cultures, ultimately enriching the global digital landscape.