AI could improve how Canada tracks its early learning and child care workforce – but human expertise remains essential
Early learning and child care (ELCC) in Canada has been expanding since the Canada-wide system was introduced in 2021, with the federal government aiming to create 250,000 new child care spaces by 2026. As demand for these services grows, so does the need for ELCC workers. Accurately understanding the workforce needed to support this expansion is not straightforward.
Existing occupational classifications can be too broad, making it difficult to identify the specific groups of workers that are most relevant to policy and workforce planning. A new study, published in the International Journal of Population Data Science (IJPDS), explores whether artificial intelligence (AI) could help develop a more detailed and policy-relevant way of identifying Canada’s ELCC workforce.
National Statistical Organisations (NSOs) are increasingly using AI to improve efficiency of producing official statistics, including automating the coding of open-ended responses. More recently, large language models (LLMs) have shown promise for classifying occupations with little, if any, labelled data. Building on these approaches, the researchers explored whether an LLM could combine information from the Canadian Census with subject-matter expertise to develop more detailed occupational categories for ELCC workers.
Canada’s ELCC workforce is typically defined by two occupational groups in the National Occupational Classification (NOC): early childhood educators and assistants, and home child care providers. However, these broad groups include people working in diverse roles and settings, making it difficult to distinguish the groups that policymakers need to understand. Additional census information, such as industry, self-employment status and place of work can help to provide a more detailed picture.
Using 57,135 occupational write-in responses from the 2021 Canadian Census, the researchers explored an LLM to classify workers into proposed occupational subgroups. These subgroups were defined, reviewed, and repeatedly refined by subject-matter experts. This approach allowed the LLM to take on the time-consuming task of processing large numbers of responses, while experts focused on checking and improving the classifications.
The findings suggest that AI could help National Statistics Organisations make better use of existing population data to respond to new and evolving policy needs. For ELCC in particular, more detailed information about the workforce could support better workforce planning and policy development. But the study also highlights an important limitation: AI can support expert judgement rather than replace it.
Click here to read the full article
Maria Ahmed, Department of Sociology, University of Western Ontario, Canada