Federal Investment to Address the Rise of Synthetic Information
The Government of Canada has announced financial support for the Université de l'Ontario français, located in Toronto, to launch an immersive project dedicated to countering disinformation and the risks of generative artificial intelligence. Supported by MP Chi Nguyen, this institutional backing signals a major realization: in a globalized French-speaking space, the proliferation of content generated entirely by algorithms threatens the integrity of public debate and trust in educational institutions.
Higher education finds itself on the front lines of this challenge. While large language models make writing and research easier, they also excel at producing texts that appear highly scientific but are completely devoid of factual basis. For researchers and educators in French-speaking Ontario, the challenge is not only to denounce this phenomenon, but also to design analytical methods and digital environments capable of restoring information traceability.
The Error Mechanism: Why Does AI Invent Things?
To understand how to counter algorithmic disinformation, we must analyze the internal workings of language models. Contrary to popular belief, an AI model does not consult a database of verified facts when answering a question. It is a probabilistic engine that predicts the most plausible next word based on billions of parameters acquired during its training phase. This behaviour regularly leads models to hallucinate: to fabricate citations, legal references, or statistical data that do not exist, all with perfect syntactic confidence.
The scientific countermeasure to this issue lies in an architecture called Retrieval-Augmented Generation, or RAG. Instead of letting the model draw freely from its chaotic internal memory, the system starts by searching for relevant document segments within a closed, verified corpus. These excerpts are converted into embeddings, which are vector representations that measure the semantic proximity of concepts. The model is then forced to formulate its response by relying exclusively on these source documents provided in real time, acting like a student taking an open-book exam.
Transparent Architecture as a Counter-Model
The research project at the Université de l'Ontario français illustrates the need for transparent tools in academia. It is precisely at this intersection that the ProductivIA application suite is positioned. Rather than relying on unsupervised AI generations, the platform offers a structured approach built around the Document Library application. Thanks to RAG technology, users submit their own reference documents, such as textbooks, official reports, and academic theses, within a secure, airtight environment. The AI invents nothing: it extracts, summarizes, and systematically links back to the exact source stored in the system.
This traceability is reinforced by the Cloud application, which guarantees full visibility over the physical location and structure of the organization's data. Unlike opaque commercial solutions where users do not know where their files are processed, the ProductivIA ecosystem makes it possible to locate every document used to feed the AI. For public institutions subject to Law 25 in Quebec or strict privacy regulations in Ontario, this security is a non-negotiable prerequisite.
Finally, the critical evaluation of technological bias is directly addressed by the AI Comparator application. This tool allows users to query multiple models simultaneously, from commercial giants to sovereign, local solutions like Matania, which is hosted in Quebec. By comparing responses side by side, students and researchers can empirically analyze how each model interprets a prompt, thereby revealing the ideological distortions and hallucinatory tendencies unique to each architecture.
Toward Sustainable Information Hygiene
The initiative at the Université de l'Ontario français serves as a reminder that the fight against disinformation cannot be solved simply by calling for individual vigilance. It requires software infrastructure designed for verifiability. By prioritizing AI systems grounded in controlled, locally hosted document collections, education and public administration sectors are laying the groundwork for digital sovereignty. The challenge of tomorrow will not be to produce more content, but to ensure that every paragraph displayed on a screen can be traced back to its original human source.