Systematic review of research on artificial intelligence in K-12 education (2017–2022)

US university education / learning technology research

North Carolina, USA

Research team systematic search, screening, and coding

• In-school (K–12)

Database search with PRISMA screening and thematic coding

Publication window 2017–2022 in included corpus

66 included studies

Spans tutors, chatbots, ML model tools, games, robots, VR as reported in primary studies

• Database scope excludes some venues
• Rapid post-2022 GenAI shift may date some application categories
• Heterogeneity limits meta-analytic effect pooling
• Equity and safety need ongoing primary research

K-12 (corpus-wide)

Varies widely by grade and community

Varies; many studies target introductory ML or tool use

• Map publication trends and research themes for AI in compulsory schooling
• Classify AI methods and technology modalities used in K-12
• Contrast typical teacher-facing vs student-facing uses

• Surface ethics, equity, and safety as persistent themes
• Guide practitioners selecting evidence-informed implementations

• Not a meta-analysis of standardized effect sizes across all tools
• Not limited to teaching-AI-only (includes AI for learning and operations)

Heterogeneous: ML builders, ITS, chatbots, games, robots, VR

• Tutor
• Automation tool
• Evaluator

English-language review; primary studies multilingual

• Teachers demonstrate models and predict performance/behavior in many studies
• Students engage in discovery learning and data-informed decisions in others

• Student data sensitivity with predictive analytics
• Bias auditing when models inform decisions
• Transparency to parents when AI supports instruction or discipline signals

• Define databases and screening criteria
• PRISMA flow from initial hits to full-text inclusion
• Code themes, methods, and application types
• Synthesize implications for K-12 personnel and researchers

• Review maps how humans delegate prediction/tutoring to AI in schools
• Primary studies retain local IRB and policy constraints

• Varies by included study; review highlights diversity rather than one best design

• Ethics, equity, and safety recur as concerns alongside innovation
• Teacher uses often emphasize demonstration and prediction
• Student uses emphasize experience and inquiry—implementation quality varies

• K-12-specific scope versus broader AIEd reviews
• Explicit taxonomy of ML paradigms and technology modalities

• Shows multifaceted uptake of AI across teaching, learning, and operations

• Documents diverse intended outcomes from performance prediction to engagement

Concludes with implications for practitioners implementing AI and for researchers addressing corpus gaps.

• Sensitive student data in predictive models require governance
• Equitable access to AI-enhanced learning environments
• Safety and oversight for chatbots and immersive tech with minors
• Algorithmic bias and transparency obligations for schools

• Activity documentation
• Practitioner observation

• Longitudinal studies on equity impacts of school AI predictors
• Design-based research linking ML demos to durable conceptual understanding

• AI in K-12
• Learning analytics ethics
• Educational technology implementation science

• Completed

K-12

Schools (multi-use)

Teach AI / learn with AI / operations

Corpus-dependent

Medium–High (prediction, data)

High in many primary studies