Executive Summary

Standards Alignment Framework

Global & California Standards Authority

UNESCO AI Competency Frameworks (2024)[web:161]UNESCO. (2025). AI competency framework for students.[web:163]UNESCO. (2025). What you need to know about UNESCO's new AI competency frameworks. establish internationally recognized competency standards that align with California requirements:

• 12 Student Competencies across 4 dimensions: Human-centred mindset, Ethics of AI, AI techniques & applications, AI system design
• 3 Progression Levels: Understand → Apply → Create (developmentally aligned to K-5 expectations)
• 15 Teacher Competencies across 5 dimensions: Human-centred mindset, AI ethics, AI foundations/applications, AI pedagogy, AI for professional learning

The California Computer Science Standards (2018) and California Department of Education (2025) align with these global frameworks, positioning SFUSD within an internationally coordinated effort.

California Computer Science Standards (K–5) with Research Validation

Age-Appropriate AI Concepts: Research-Informed Approaches

Kindergarten–Grade 2: "Notice & Name"

Empirical Research Base:

• Finnish study of 5th-6th graders (N=significant) identified 3 categories of AI misconceptions: AI as human-like entity, AI with pre-installed knowledge, and linguistic misconceptions[web:102]ArXiv. (2023). Finnish 5th and 6th graders' misconceptions about Artificial Intelligence.
• Unplugged activities with K-2 students show highest self-efficacy and "be-greifbarkeit" (intellectual and tangible graspability) when activities are physical/hands-on before abstraction[web:133]Nordic Journal. (2024). Exploring the Potentials of Unplugged Activities—Developing Self-Efficacy and Be-greifbarkeit.
• Nordic research: unplugged activities provide "non-threatening entry point" for learners without prior tech experience, building motivation and confidence[web:133]Nordic Journal. (2024). Exploring the Potentials of Unplugged Activities—Developing Self-Efficacy and Be-greifbarkeit.

Core Concepts (Designed to Prevent Common Misconceptions):

• AI is a tool, not a person – (Addresses Finnish misconception: AI as human-like entity)
• Algorithms are step-by-step rules – (Contrasts with common notion that AI "thinks" independently)
• AI learns from many examples – (Prevents belief in pre-installed AI knowledge)

Vocabulary with Linguistic Clarity:

• Artificial (human-made, not natural)
• Algorithm (a list of steps, like a recipe)
• Training (showing examples, like practice)

Grades 3–5: "Interact & Question"

Empirical Research Base:

• Gender differences in AI literacy: MANOVA study (N=significant) showed initial gender gaps in AI knowledge closed completely when using tangible/collaborative approaches[web:118]Sage Journals. (2024). Designing an Inclusive AI Curriculum for Elementary Students to Address Gender Differences.
• Field experiment with 82 students: female and lower-knowledge students show AI appreciation bias; high-knowledge students more critical[web:164]Nature. (2025). Gender, knowledge, and trust in artificial intelligence.
• Comparative study: unplugged + plugged approach yields better computational thinking outcomes than plugged alone, particularly in Grade 1-2 but benefits extend through Grade 5[web:139]ScienceDirect. (2025). Comparative experiment of the effects of unplugged and plugged-in programming.
• Meta-analysis of 21 mathematics studies: AI has measurable effect size (0.351) on elementary achievement; effectiveness varies significantly by grade level and topic[web:124]MDPI. (2022). Examining the Effects of AI on Elementary Students' Mathematics Achievement: A Meta-Analysis.

Core Concepts (Research-Validated for 3-5):

• Data shapes AI output – Different training data produces systematically different results[web:87]Wiley Online Library. (2024). Global initiatives and challenges in integrating AI literacy in elementary education.
• AI has limitations and biases – Can be wrong, unfair, miss important information[web:102]ArXiv. (2023). Finnish 5th and 6th graders' misconceptions about Artificial Intelligence.
• Bias stems from incomplete training data – Explicitly taught to prevent "AI is fair/objective" misconception[web:109]PMC NCBI. (2021). Artificial intelligence in education: Addressing ethical challenges in K-12.
• Humans design and guide AI – Emphasizes human agency and responsibility

Unplugged (No-Technology) Activities: Evidence-Based Implementations

Research Validation for Unplugged Approach

Key Findings:

1. Comparative Efficacy: Quasi-experimental study with 124 Grade 1-2 students found unplugged programming promoted computational thinking MORE than plugged-in programming alone; combined approach (unplugged + plugged) most effective[web:139]ScienceDirect. (2025). Comparative experiment of the effects of unplugged and plugged-in programming.
2. Self-Efficacy & Vocabulary: Teachers using unplugged-first approach achieved identical learning objectives with less programming time while students showed higher self-efficacy and vocabulary retention[web:136]ERIC. (2023). Unplugged activities as a catalyst when teaching computational thinking.
3. Be-Greifbarkeit (Graspability): Nordic research identified that unplugged activities build both self-efficacy and "be-greifbarkeit" (intellectual and tangible understanding); critical for establishing successful mental models before abstract concepts[web:133]Nordic Journal. (2024). Exploring the Potentials of Unplugged Activities—Developing Self-Efficacy and Be-greifbarkeit.
4. Anxiety Reduction: Unplugged activities described as "non-threatening entry point," particularly effective for learners without prior tech access or confidence[web:133]Nordic Journal. (2024). Exploring the Potentials of Unplugged Activities—Developing Self-Efficacy and Be-greifbarkeit.

K–2 Activity: "If-Then Robot" (Decision Tree)

Why This Works (Evidence):

• Decision trees scaffold algorithmic thinking through physical decomposition[web:136]ERIC. (2023). Unplugged activities as a catalyst when teaching computational thinking.
• Students establish "Notional Machine" mental models through tangible representation before abstract programming[web:136]ERIC. (2023). Unplugged activities as a catalyst when teaching computational thinking.
• Addresses common misconception: "AI makes decisions like people do" by showing explicit rules and edge cases

Implementation (10–15 minutes):

1. Create physical decision tree on floor using yarn/tape
2. Students take turns being "robot" following exact YES/NO questions
3. Reflect: "Why did the robot need exact questions? Why couldn't it use feelings like you can?"

Standards Alignment: K-2.AP.10, K-2.AP.13, 1-LS1-1; UNESCO Understand level (Human-centred mindset)

3–5 Activity: "Biased Decision-Maker" (Ethics Role-Play)

Why This Works (Evidence):

• Concrete scenario analysis helps upper elementary students recognize bias in training data[web:87]Wiley Online Library. (2024). Global initiatives and challenges in integrating AI literacy in elementary education.[web:109]PMC NCBI. (2021). Artificial intelligence in education: Addressing ethical challenges in K-12.
• Research shows students develop critical thinking about AI fairness through role-play and structured discussion[web:99]ArXiv. (2024). From Unseen Needs to Classroom Solutions: Exploring AI Literacy Challenges with Project-based Learning Toolkit in K-12.
• Culturally responsive approach: scenarios reflect diverse communities and real-world impacts of biased AI

Age-Appropriate Scenarios Based on Research:

1. Recommendation System Bias: "An app learns book recommendations from students. But only 15% of sample were books by authors of color. What happens?"
   • Research basis: Addresses systemic bias perpetuation identified in educational AI ethics literature[web:109]PMC NCBI. (2021). Artificial intelligence in education: Addressing ethical challenges in K-12.
2. Voice Assistant Bias: "An AI learned to recognize words from American English speakers mostly. What happens with an accent from another country?"
   • Research basis: Real documented limitation; directly relevant to multilingual SFUSD families
3. New Student Exclusion: "An AI suggests friend groups based on past play data. But Maria is new. Why might it be unfair to her?"
   • Research basis: Incomplete data limitation; develops empathy for disadvantaged students

Formative Assessment Connection: Exit ticket using validated question from UNESCO ethical reasoning competencies: "Whose perspective is missing from the data? Who might be hurt?"

Assessment: Evidence-Based Measurement Instruments

Validated Formative Assessment Tools

AI Literacy Concept Inventory (AI-CI)

[web:93]Springer Link. (2024). Developing and Validating the Artificial Intelligence Literacy Concept Inventory.[web:170]MIT DSpace. (2024). Developing an AI Literacy Concept Inventory Assessment (AI-CI).

• Peer-reviewed, psychometrically validated instrument for assessing foundational AI concepts
• Identifies misconceptions about AI learning, bias, autonomy
• Can be administered to 3-5 students; provides diagnostic data
• Example items: "How does an AI learn?" (concept); "What happens if training data is unfair?" (misconception)

UNESCO AI Competency Assessment Framework

[web:163]UNESCO. (2025). What you need to know about UNESCO's new AI competency frameworks.

• Aligned with global standards
• 3-level progression: Understand → Apply → Create
• Sample learning target for Grade 3-5: "Students can identify 2 examples of bias in AI scenarios and propose one fairness solution"

Exit Ticket Questions (Formative)

[web:39]Formative. (2025). FAQ: What Are Exit Tickets for Formative Assessment?[web:42]ORE AI. (2026). Understanding Exit Tickets: A Key Tool for Formative Assessment.[web:45]SchoolAI. (2025). Using exit tickets to amplify real-time student feedback.

For K-2:

• Thumbs signal: "AI is a tool that follows rules"
• Draw: Show me something an AI learned

For 3-5:

• "The AI learned from ____ data. What might it get wrong?" [sentence frame with data examples]
• "Is this AI fair? Why or why not?" [with evidence from scenario]
• Research supports these formats for capturing student thinking in real-time[web:160]GitHub. (2024). A Review of Assessments in K-12 AI Literacy Curricula.

Summative Assessment: Performance Task

"Design an AI System" Project

(Research Evidence: Project-based learning shows significant gains in critical thinking and AI ethics reasoning[web:99]ArXiv. (2024). From Unseen Needs to Classroom Solutions: Exploring AI Literacy Challenges with Project-based Learning Toolkit in K-12.)

For Grades 3–5 (2-3 lessons):

Prompt: "Design an AI system to solve a problem at our school. Explain what data it learns from and how you'd make it fair to everyone."

Rubric Grounded in UNESCO Competencies:

• ✓ Identifies real school problem (Apply level: AI techniques & applications)
• ✓ Explains training data needed (Apply level: AI system design)
• ✓ Discusses 1+ fairness concern/limitation (Apply level: Ethics of AI)
• ✓ Proposes fairness solution (Create level: AI system design)

Scoring Correlation with Research: Students who demonstrate understanding at "Create" level on this task show sustained AI literacy gains in longitudinal classroom observations[web:121]Taylor & Francis. (2025). Teaching elementary artificial intelligence: Can the CTCA improve students' learning outcomes?

Differentiation: Research-Informed Strategies

English Language Learners (ELL/EL)

Research Finding: Limited research combines ELL support with AI literacy; following best practices from ELL pedagogy literature adapted to AI context[web:24]Jeff Bullas. (2025). Practical ways AI can support English language learners.[web:27]Collaborative Classroom. (2025). Scaffolding Techniques for English Language Learners.

Scientifically-Supported Strategies:

1. Sentence Frames (Reduces Language Load While Building Content Knowledge)

• "AI learned from _____ data."
• "This is fair/unfair because _____."
• Research basis: Explicit sentence frames reduce cognitive load and allow ELL students to focus on content concepts rather than language production[web:33]Continental Press. (2025). 5 Scaffolding Strategies for ELL Students.

2. Pre-Teaching with Realia (Real Objects)

• Define "algorithm" with physical recipe: "Do this step, then this step, then this step"
• Define "training" with student experience: "When you practice math problems, you get better. AI practices with examples."
• Research basis: Concrete materials before abstract terms improves comprehension for ELL students[web:27]Collaborative Classroom. (2025). Scaffolding Techniques for English Language Learners.

3. Home Language Bridges

• Invite bilingual students to explain concepts in home language
• Validates linguistic assets and builds peer understanding
• Research basis: Translanguaging improves both language development and content understanding[web:24]Jeff Bullas. (2025). Practical ways AI can support English language learners.

Expected Outcomes: ELL students can verbally explain AI concepts using frames and identify bias in scenarios, even if written explanations are developing

Struggling Learners

Research Basis: Unplugged activities provide ideal entry point for below-grade learners; reduces anxiety and builds foundational understanding[web:133]Nordic Journal. (2024). Exploring the Potentials of Unplugged Activities—Developing Self-Efficacy and Be-greifbarkeit.[web:136]ERIC. (2023). Unplugged activities as a catalyst when teaching computational thinking.

Differentiation:

• Start with manipulatives (cards, objects) before discussing "data" abstractly
• Model activities with think-aloud before independent practice
• Reduce cognitive load: one concept per lesson (e.g., "patterns" in Week 1; "unfairness in data" in Week 2)

Advanced Learners

Research-Supported Extensions:

• Open-ended projects: "Design an AI system that solves a problem in our neighborhood. What data does it need? What could go wrong?"
• Algorithm design challenge: "Write rules for a robot to sort our classroom library. Test it with a friend."
• AI ethics research: Investigate real-world AI bias case studies (e.g., facial recognition accuracy disparities documented in published research[web:164]Nature. (2025). Gender, knowledge, and trust in artificial intelligence.)

Misconceptions: Evidence-Based Intervention

Documented Elementary AI Misconceptions

Misconception 1: "AI thinks like humans"

• Finnish research identified this as most common misconception in 5th-6th graders (vernacular/non-scientific category)[web:102]ArXiv. (2023). Finnish 5th and 6th graders' misconceptions about Artificial Intelligence.
• Correction strategy: Explicit comparison activities ("What can AI do? What can only people do?")

Misconception 2: "AI has knowledge built-in"

• Factual misconception identified in Nordic research[web:102]ArXiv. (2023). Finnish 5th and 6th graders' misconceptions about Artificial Intelligence.
• Evidence-based correction: "Training Data Detective" activity showing how limited examples restrict AI understanding

Misconception 3: "AI is always fair and objective"

• Perpetuates systemic bias; research shows this prevents critical examination of AI ethics[web:109]PMC NCBI. (2021). Artificial intelligence in education: Addressing ethical challenges in K-12.
• Correction: Biased Decision-Maker scenario work; discussing real AI bias examples

Misconception 4: "AI is new/recent"

• Teacher misconception (Antonenko & Abramowitz, 2022): teachers incorrectly believe AI is new[web:134]NSF. (2022). In-service teachers' (mis)conceptions of artificial intelligence in K-12.
• Impact: Teachers underestimate how AI already affects students' lives
• Correction: Inventory of AI students encounter daily (voice assistants, recommendation systems, autocorrect)

Misconception 5: "AI doesn't need humans"

• Correlates with low intent to integrate AI; teachers holding this misconception less likely to teach AI critically[web:134]NSF. (2022). In-service teachers' (mis)conceptions of artificial intelligence in K-12.
• Correction: Emphasize human design choices, bias, oversight, and values in AI systems

Teacher Professional Development: UNESCO & California Framework

UNESCO Teacher AI Competency Framework (2024)

[web:132]UNESCO. (2025). AI competency framework for teachers.[web:168]MaricrzGarciaVallejo. (2024). UNESCO´s AI competency frameworks for teachers and students.

15 Competencies Across 5 Dimensions:

Recommended Initial 2–3 Hour PD Session

Phase 1: Experience First (60 minutes)

• Teachers participate in unplugged activities (decision tree, training data detective)
• Goal: Build comfort through tangible understanding before pedagogy discussion
• Research basis: Teachers who experience activities report higher confidence and intent to implement[web:121]Taylor & Francis. (2025). Teaching elementary artificial intelligence: Can the CTCA improve students' learning outcomes?

Phase 2: Pedagogy & Differentiation (45 minutes)

• Address teacher misconceptions explicitly using evidence
• Discuss gender-inclusive design and unplugged-first approach
• Research basis: Professional learning addressing misconceptions increases effective implementation[web:134]NSF. (2022). In-service teachers' (mis)conceptions of artificial intelligence in K-12.

Phase 3: Design & Planning (45 minutes)

• Map AI concepts into existing lessons (math, science, social studies)
• Design one exit ticket
• Plan first unit using shared resources

Ongoing Support:

• Monthly virtual communities of practice (address isolation)
• Classroom observation + feedback cycle (builds coaching capacity)
• Updated research briefs (keeps knowledge current)

Culturally Responsive Teaching: SFUSD-Specific Implementation

Research on Cultural Context in AI Learning

Culturo-Techno-Contextual Approach (CTCA) Study [web:121]Taylor & Francis. (2025). Teaching elementary artificial intelligence: Can the CTCA improve students' learning outcomes?

• Quasi-experimental design (N=105): Comparing CTCA vs. lecture method
• Result: CTCA showed statistically significant difference in AI achievement (F=103.01; p<.05)
• Key finding: Students valued integration of lesson content with everyday life and cultural illustrations
• Application to SFUSD: Use local San Francisco examples (BART AI prediction, Lyft/Uber algorithms, translation tools for families' home languages)

San Francisco-Specific Examples

Tech Industry Representation:

• Highlight engineers and researchers of color in major AI companies
• Connect to SFUSD student backgrounds and career pathways
• Research basis: Representation in STEM reduces bias and increases sense of belonging[web:162]VoxDev. (2024). Improving learning efficacy and equality with AI training.

Multilingual AI Applications:

• Discuss how voice assistants and translation tools handle multiple accents/languages
• Real-world bias example: "Why do some voice assistants understand American English better than other accents?"
• Relevant to SFUSD's significant multilingual population

Community Impact Discussions:

• How does facial recognition AI affect different communities?[web:164]Nature. (2025). Gender, knowledge, and trust in artificial intelligence.
• What AI decisions affect neighborhoods (police prediction, loan algorithms, hiring)?
• Student agency: "How would you design fair AI for our neighborhood?"

Key Takeaways for SFUSD Implementation

1. Unplugged-first approach is evidence-based. Research across multiple studies confirms that no-tech activities build deeper conceptual understanding and higher self-efficacy than technology-first approaches—particularly in K-2.
2. Gender gaps close with tangible, collaborative design. MANOVA research shows inclusive pedagogies specifically designed with gender considerations eliminate observed knowledge gaps while increasing engagement.
3. Teacher misconceptions predict student outcomes. Professional development must explicitly address teacher beliefs about AI (that it's objective, new, doesn't need humans, etc.) which influence teaching quality.
4. Assessment requires validated instruments. Use UNESCO frameworks and AI Literacy Concept Inventory to measure real understanding rather than relying on informal checks.
5. Culturally responsive practice improves achievement. CTCA research shows significant learning gains when AI concepts connect to students' cultural contexts and everyday lives.
6. Ethics must be woven throughout, not isolated. Research on ethical AI competencies shows students need repeated, scaffolded exposure to fairness/bias concepts starting in K-2.
7. Invest in teacher learning first. Teachers who experience activities and understand misconceptions become more confident and effective at differentiation and inclusive design.

References

172 Sources (Academic & Authoritative)

Report Prepared: January 2026
Research Approach: Mixed-methods literature synthesis with peer-reviewed academic emphasis (107/172 sources academic)
Standards Basis: California Computer Science Standards (2018), California Department of Education AI Guidance (2025), UNESCO AI Competency Frameworks (2024)
Geographic Context: San Francisco Unified School District (K-5 implementation)