Design of a Science-Integrated Secondary School AI Literacy Curriculum: A Youth & AI Expert Guided Design-Based Research Approach

Synthesis: Moore, Rabinowitz, Ali, Weckel, Lee, Gupta, and Chaffee (2026) present a mixed-methods analysis of the iterative design of an informal science-integrated machine learning curriculum for high school youth enrolled in a four-week summer program. Each step of the two-year design-based research (DBR) process was informed by input from an advisory board consisting of alumni and industry experts. Participants in both cohorts gained understanding in ML knowledge, with gains in Cohort 2 (M2-M1=0.175, p < 0.001, n = 42) exceeding those of Cohort 1 (M2-M1=0.076, p < 0.001, n = 35). Participants who identified as female and non-White tended to show greater learning gains than their White male counterparts.

Key Findings

• A two-year DBR process designed an informal science-integrated ML curriculum for high school youth, informed by a youth and AI-expert advisory board.
• Participants in both cohorts gained ML knowledge, with Cohort 2 gains (M2-M1=0.175) exceeding Cohort 1 (M2-M1=0.076).
• Participants who identified as female and non-White tended to show greater learning gains than their White male counterparts.
• The project exemplifies a participatory curriculum design process that centers youth voices in an advisory capacity.
• Findings have implications for educational designers seeking to integrate AI/ML into existing curricula.

What this means for practice

• Educators. Teach machine learning through authentic science datasets and tools rather than as standalone computing content, so learners see how the technique serves a disciplinary question.
• Educators. Use unplugged and online simulations alongside media such as Coded Bias to make ML's limitations and societal bias concrete: youth reported these activities as the clearest route to understanding what models get wrong.
• Curriculum designers. Run explicit design-based research cycles with an advisory board of program alumni and industry experts, and revise the curriculum from interview, item-distractor, and survey evidence at each cycle.
• Curriculum designers. Analyze wrong answers item by item to find the misconception behind each distractor, then rebuild the activity around it — the practice that drove the second iteration here.
• Researchers. Report equity outcomes by subgroup rather than only as aggregate pre-post gains: the Cycle 2 revisions narrowed a persistent gender gap in ML general knowledge.

Limitations

• Both cohorts are small and non-random within one informal four-week summer program with no control group: Cohort 1 n = 35 and Cohort 2 n = 42, so gains (Cohort 2 M2−M1 = 0.175) cannot be attributed to the curriculum rather than to the program or to cohort composition.
• Learning is indexed by pre- and post-institute survey responses rather than a standardized ML assessment, and the authors acknowledge the small sample may have reduced test power (for example, a gender comparison at p = 0.318).
• Subgroup findings rest on self-reported gender and ethnicity, and interview data are retrospective accounts collected at the end of the program.
• The authors name two unresolved limits: the long-term influence of the curriculum on students' AI knowledge, skills, attitudes, and career interests is unexamined, and the materials are not yet scaffolded for science educators in varied contexts.

Citation

Moore, K. S., Rabinowitz, G., Ali, S., Weckel, M., Lee, I., Gupta, P., & Chaffee, R. (2026). Design of a science integrated secondary school AI literacy curriculum: A youth & AI expert guided design-based research approach. Computers and Education: Artificial Intelligence, 10, 100552.