A STEAM-PBL-AIoT course was associated with broader elementary AI-literacy gains and narrower measured gender gaps

Cheng and colleagues study both how elementary AI literacy is measured and how a project-based course may change it. In the first study, 504 Taiwanese fifth graders completed a questionnaire covering affective, behavioural, cognitive and ethical dimensions. The researchers examined reliability, validity and measurement equivalence across gender. In the second study, four classes completed 19-week courses: 55 students took an AIoT comparison course, while 54 took an integrated STEAM, project-based learning and AIoT course organized around authentic problems and AI-literacy goals.

Both groups improved from pre-test to post-test. After accounting for initial differences, the integrated course had significantly higher post-test scores for overall AI literacy and the affective, behavioural and cognitive dimensions, with reported effects in the medium range, approximately r = .30 to .44. The groups did not differ significantly on the ethics dimension. The result suggests that richer projects and explicit literacy goals may support several aspects of how learners perceive and approach AI, while ethical understanding may require different or more sustained pedagogy.

Gender patterns were a central question. Boys scored higher than girls across the measured dimensions at pre-test. By post-test, gender differences within both course groups were no longer statistically significant, and girls in the integrated course showed substantial gains. This is encouraging evidence of a narrowed measured gap in this sample. It is not proof that a course permanently eliminates gender inequity or that every learner experienced the same classroom opportunities.

The design limits causal claims. Classes were not randomly allocated, so teacher, class or selection differences may contribute to the outcomes. The intervention comparison involved only 109 students in one Taiwanese fifth-grade context. Most outcomes came from a self-report questionnaire rather than externally scored projects, observed practices or transfer tasks. There was no long-term follow-up to determine whether gains or narrower gender differences persisted. The lack of an additional ethics effect also warns against treating ethical literacy as an automatic by-product of technical projects.

For curriculum design, the study supports making AI literacy multidimensional. A unit can ask learners to build or investigate a useful system, explain how it works, evaluate data and limitations, collaborate in varied roles and discuss who may benefit or be excluded. Teachers can monitor who handles hardware, coding, explanation and leadership so participation does not reproduce stereotypes. Ethics can be assessed through age-appropriate cases and decisions rather than agreement with broad statements alone.

For Hong Kong primary education, replication should include Cantonese and English materials, local curriculum aims, objective performance tasks and classroom observations. Schools can disaggregate participation and outcomes while avoiding deficit assumptions about any group.

The paper offers a promising course model and a validated measurement effort, but its strongest conclusion is bounded: in this quasi-experiment, an integrated course was associated with larger gains in several self-reported AI-literacy dimensions and with gender gaps that were no longer statistically detectable at post-test.

Future studies can strengthen the evidence by randomly assigning classes where feasible, scoring student projects blind to condition and checking retention months later. Qualitative interviews can also show whether a smaller measured gender gap reflects genuinely broader participation.