Artificial Intelligence teaching and learning in K-12 from 2019 to 2022: A systematic literature review

Computing education research centre and foundation

Cambridge, UK

Authors screening, extracting, and synthesizing empirical AI education studies

• In-school (K–12)
• Informal learning

PRISMA-style systematic search across five databases (2019–2022)

Publication window 2019–2022

28 included empirical studies

Diverse tools across primary studies (robots, unplugged, IDEs, etc.)

• Rapid post-2022 expansion not covered
• Heterogeneity limits meta-analysis
• Publication bias toward positive results possible
• Outcome measures often inconsistent

K-12 learners across included studies

Varies widely

Varies by intervention level

• Inventory empirical evidence on K-12 AI learning outcomes
• Map pedagogical and theoretical coverage
• Highlight research gaps and challenges

• Inform policy and curriculum with evidence-based patterns
• Encourage standardized constructs for AI learning measurement

• Not a meta-analysis of effect sizes
• Not teaching-with-AI-only reviews
• Not longitudinal primary data

Heterogeneous across corpus (robots, visual ML tools, block languages, etc.)

• Tutor
• Co-creator

English-language search; studies may be multilingual

• Varies by study design in corpus

• Primary studies should report ethics and data practices clearly
• Equity in who gets rigorous AI interventions
• Avoid overstating evidence from small pilots

• Define inclusion criteria for empirical outcome studies
• Screen thousands of records to 28 papers
• Content-analyze pedagogy, theory, topics, outcomes
• Synthesize limitations and future research agenda

• Reviewers judge study quality; AI tools optional in future updates of method

• Learner-centred and context-aware designs recommended from patterns

• Limited empirical base relative to hype
• Need consistent outcome instruments
• More work on transfer and retention beyond immediate post-tests

• Tight focus on empirical learning outcomes distinguishes from descriptive program reviews

• Most reviewed work reports positive cognitive and/or affective signals

• Evidence supports feasibility but not uniformity of rigor

Calls for more learner-centred, context-aware pedagogy and better measurement constructs.

• Ethical reporting of child studies in underlying corpus
• Data privacy in interventions using student models
• Transparency in selective reporting of successful pilots

• Activity documentation
• Practitioner observation

• Registered systematic reviews updating annually
• Shared outcome banks for AI education RCTs

• K-12 AI education
• Systematic review methodology
• Learning sciences

• Completed

K-12

International corpus

Teach AI / ML

Varied (reviewed)

Low (synthesis)

N/A