Can AI become culturally competent? Canada’s AI strategy may depend on it
Photo by Sandro Schuh on Unsplash
As countries race to develop national artificial intelligence strategies, public debate has largely centred on productivity, fairness, innovation, economic competitiveness and the risks of misinformation. Yet in a multicultural society like Canada, what is at stake is whether AI models and systems can actually become culturally competent enough to be used safely in cross-cultural settings?
This question matters because AI is no longer a backend, behind-the-scenes technology; it is actively used in communication in schools, hospitals, immigration services, workplaces, customer service interactions and social media platforms. Increasingly, it functions as a mediator on how people encounter one another across cultural differences.
But the problem is that much of what tech companies and policymakers describe as “culturally aware AI” may not amount to genuine cultural competence at all.
In a recent interdisciplinary paper co-authored with researchers from the National Research Council Canada, Toronto Metropolitan University, and Mila–Quebec AI Institute, we argue that conversations about “cultural capability” in AI often rely on vague and interchangeable terminology that obscures important distinctions. Systems that appear culturally informed may still misunderstand users, reinforce stereotypes or flatten cultural complexity.
Three levels of cultural capability
Our research proposes distinguishing between three levels of cultural capability in AI systems: cultural awareness, cultural sensitivity and cultural competence.
The first, cultural awareness, refers to factual knowledge about cultures. For example, an AI chatbot may know that Japanese workplaces tend to emphasize hierarchy and formality, or that certain gestures carry different meanings in different cultures. Much current AI evaluation operates at this level. But knowing facts about cultures is not the same as interacting appropriately across cultural differences.
The second level, cultural sensitivity, concerns how AI models frame cultural differences. Does the model speak respectfully? Does it avoid stereotyping or ethnocentric assumptions? Does it present cultural practices as complex and contextual, rather than exotic or backward? While more advanced, even this level of cultural understating is not enough.
The third level, cultural competence, involves the ability to adapt communication dynamically as an interaction unfolds. A culturally competent AI system would not merely retrieve cultural information; it would adjust its tone, explanations and assumptions as users reveal new context about their identities, values or circumstances.
This distinction matters because public discussions about AI often collapse all three levels into one. A system that performs well on multicultural trivia or translation tasks may still fail in a real intercultural interaction.
Why this matters in Canada
The stakes are especially high in a country like Canada, where public institutions often serve populations shaped by migration, linguistic diversity and differing cultural understandings of health, education, authority and communication itself.
Consider healthcare. An AI system assisting communication between a doctor and a newcomer patient may accurately recognize cultural information while still framing that patient’s beliefs in ways that feel dismissive or paternalistic. Or consider education: a chatbot may give technically accurate answers while missing the cultural nuances that shape how students interpret authority, disagreement or emotional expression.
These are not simply technical glitches. They reflect deeper limitations in how AI models are built and work.
Most large AI models are trained primarily on English-language and Western-centric data. This creates structural imbalances in how cultural norms, communication styles and worldviews are represented: some cultures appear richly documented and “normal,” while others are reduced to stereotypes, fragmented data points or exoticized difference. As a result, AI systems can unintentionally reproduce existing global inequalities in cultural visibility and representation.
Written by Masoud Kianpour. Masoud Kianpour is a Senior Research Associate at the Global Migration Institute at Toronto Metropolitan University, whose research explores social psychology, intercultural communication, multiculturalism, identity and social inequalities.
LATEST NEWS
