Comparing Artificial Intelligence Curricula in Canadian and US Universities

Rose Niousha, Lexie Jingruo Guo, Rick Kaifeng Li, Narges Norouzi, Lisa ZhangAAAI 2025 · DOI

TL;DR

This paper compares the structure of Artificial Intelligence (AI), Machine Learning (ML), and Data Science (DS) courses in Canadian universities with US Research-1 institutions. It reveals that Canadian AI/ML courses generally have higher exposure levels (meaning they are taken later in the curriculum) and that DS courses are less prevalent compared to the US, suggesting a need for curriculum adjustments to improve early access to AI education.

What Problem Does This Paper Address?

The rapid growth of Artificial Intelligence (AI) has led to a surge in demand for AI education. To foster the next generation of AI professionals, it's crucial to make AI-related courses accessible and engaging early in university curricula. However, the prerequisites for these courses can create barriers, especially for students who don't have a strong prior background in computer science and mathematics. While previous research has explored AI curriculum structures in US universities, there's a need to understand how these structures compare internationally, particularly in countries with similar educational systems but distinct approaches, like Canada. This paper aims to bridge this gap by analyzing and comparing the prerequisite structures of AI, ML, and DS courses in Canadian universities with those in US Research-1 institutions.

Methods

The researchers employed a methodology similar to their previous work on US universities. They collected data on AI, ML, and DS courses from a sample of 30 Canadian universities. This involved:

  • Course Identification: Examining CS department course calendars and descriptions to identify and categorize AI, ML, and DS courses.
  • Prerequisite Analysis: Systematically identifying and coding the prerequisite sequences for each identified course. They updated a pre-existing codebook to accommodate Canadian-specific course titles like "Mathematical Proof" and "Introduction to Math."
  • Exposure Level Calculation: Computing an "exposure level" for each course, representing the earliest term a student could take it based on its prerequisite chain.
  • Visualization: Using Sankey diagrams to visualize common prerequisite paths.
  • Clustering: Applying K-means clustering to group courses based on their prerequisite structures, course type, and level.
  • Cross-country Comparison: Directly comparing the calculated exposure levels, Sankey diagrams, and cluster analyses with their prior findings for US universities.
  • Early Exposure Analysis: Identifying and analyzing "low exposure" AI and ML courses (those with minimal or no prerequisites) to understand strategies for making AI education more accessible.

Datasets

The study utilized publicly accessible course information from:

  • Canadian Universities: A sample of 30 universities from a list of 104 Canadian universities, selected based on having a CS department and offering AI courses.
  • US Universities: Data from a previous study (Niousha et al., 2024) which sampled 50 US Research-1 institutions.

No new benchmarks were introduced; instead, the study focused on comparing existing curriculum structures.

Key Insights and Ideas

  • Canadian AI/ML courses are generally harder to access early: Canadian AI and ML courses tend to have higher exposure levels, meaning students typically take them later in their academic careers compared to their US counterparts. This could delay students' entry into advanced AI studies and research.
  • DS courses are less prevalent in Canada: Canadian CS departments appear to offer fewer Data Science (DS) courses compared to US institutions.
  • DS courses offer a potential entry point: Similar to the US, DS courses in Canada often have fewer prerequisites and can serve as an accessible entry point for students interested in AI and ML.
  • Innovative approaches exist for early exposure: Both Canadian and US institutions are employing strategies like offering introductory AI courses with minimal prerequisites or interdisciplinary courses that cover AI concepts without deep technical demands.
  • Parallelizing prerequisites is key: Structuring prerequisites in a parallel fashion (allowing students to take multiple foundational courses concurrently) can significantly reduce the time to access AI and ML courses.
  • Academic advising is crucial: Effective academic advising is essential to help students navigate prerequisite pathways and plan their courses to ensure timely access to AI and ML education.

Key Results

  • Exposure Levels: In Canada, most AI courses had an exposure level of 3 (requiring 3 semesters), while ML courses generally had even higher levels. DS courses showed more variability, with many accessible in the first year. In contrast, US AI and ML courses more commonly had exposure levels of 2.
  • Prerequisite Chains: Canadian AI and ML courses often involved longer prerequisite chains, with "Data Structures" being a common prerequisite for AI courses (72% in the Canadian sample). US institutions sometimes integrated prerequisite content into other courses, whereas Canadian universities more frequently required specific foundational courses like "Mathematical Proof."
  • DS Course Prevalence: The percentage of DS courses in the Canadian sample was lower than in the US.
  • Course Clustering: The study identified 8 clusters of Canadian AI courses based on their prerequisite structures, revealing distinct patterns in how courses are sequenced.

Limitations and Future Work

  • Data Accessibility: The study relied on publicly accessible course information, which may have been incomplete or outdated. Specific prerequisite enforcement and course offerings might not have been fully captured.
  • Departmental Scope: The analysis was limited to CS departments and did not include AI courses offered in other disciplines like Statistics or Mathematics, potentially underrepresenting DS course availability.
  • Regional Differences: Higher education in Quebec has a different structure, which might have led to an underestimation of exposure levels for AI courses in that region.
  • Private vs. Public Institutions: Due to a limited number of private universities in the Canadian sample, comparisons based on institution type were not fully replicated from the US study.
  • Future Work: The authors suggest offering DS, introductory AI, and applied ML courses with fewer prerequisites and parallelizing prerequisites to improve early access. They also emphasize the importance of curriculum revisions and early academic advising to enhance AI course accessibility.

Summary updated

BibTeX
@inproceedings{niousha2025comparing,
  author = {Rose Niousha and Lexie Jingruo Guo and Rick Kaifeng Li and Narges Norouzi and Lisa Zhang},
  title = {{Comparing Artificial Intelligence Curricula in Canadian and US Universities}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aaai.v39i28.35176},
}
Figure