Learning Science Foundation
Evidence-Based Design
The El Segundo AI Academy curriculum is grounded in decades of research on effective instruction for underrepresented learners. Every design decision traces to empirical evidence.
Stereotype Threat Research
The Phenomenon
Claude Steele's seminal research (1995, 1997) demonstrated that awareness of negative stereotypes about one's group impairs performance. In computing contexts:
- Women performed worse on math tests when told the test "shows gender differences"
- The effect disappeared when the stereotype was not made salient
- Belonging interventions can neutralize stereotype threat effects
Our Application
| Stereotype Threat Factor | Curriculum Response |
|---|---|
| Numeric underrepresentation | 50/50 gender parity in all visible elements |
| Lack of role models | Required diverse guest speakers and examples |
| Masculine-coded environments | Girls-only spaces as opt-in alternative |
| Attribution to ability | Growth mindset framing throughout |
| Competitive evaluation | Collaborative assessment structures |
Safe Space First Pattern
Research on girls in esports (AnyKey, 2019; Women in Games, 2021) reveals that girls-only programming produces dramatically better outcomes:
Evidence Base
| Study | Finding |
|---|---|
| DreamHack Girls Initiative | 5x higher retention in girls-only leagues vs. mixed |
| Harvey Mudd CS Restructuring | Girls-only intro course contributed to 40% to 55% female CS majors |
| Carnegie Mellon Women@SCS | Single-gender mentoring circles improved persistence |
| Google CS First | Girls-only coding clubs showed higher completion rates |
The Transition Model
Our curriculum implements a research-based progression:
- Safe Space Entry: Girls-only or identity-affirming initial programming
- Skill Building: Develop competence in supportive environment
- Confidence Development: Experience success and peer validation
- Optional Transition: Move to mixed-gender spaces when ready
- Continued Access: Safe spaces remain available throughout
This model respects individual readiness while building toward integrated participation.
Visible Role Model Research
Dasgupta's research (2011) demonstrates that visible role models significantly impact underrepresented students:
- Exposure to female professors increased women's implicit identification with math
- Same-gender role models improved STEM persistence by 3x
- Effect was strongest for students with initially weaker identification
Our Implementation
- Mandatory 50% female guest speakers in all AI programming
- Curriculum case studies featuring diverse AI practitioners
- Mentorship connections to industry professionals matching student identities
- Student showcases highlighting diverse peer success stories
Universal Design for Learning (UDL)
The CAST framework (2018) provides our accessibility foundation:
Multiple Means of Engagement
| Principle | Implementation |
|---|---|
| Optimize relevance | Connect AI concepts to student interests and communities |
| Minimize threats | Reduce anxiety through mastery-oriented assessment |
| Foster collaboration | Peer learning structures throughout |
| Develop self-assessment | Reflection tools for metacognitive development |
Multiple Means of Representation
| Principle | Implementation |
|---|---|
| Offer alternatives to visual | Audio descriptions, tactile models |
| Offer alternatives to auditory | Captions, transcripts, visual cues |
| Clarify vocabulary | AI terminology glossary with examples |
| Highlight patterns | Consistent structure across lessons |
Multiple Means of Action & Expression
| Principle | Implementation |
|---|---|
| Vary response methods | Written, verbal, visual, kinesthetic options |
| Optimize access to tools | Assistive technology compatibility |
| Guide goal-setting | Scaffolded project planning |
| Enhance capacity for monitoring | Progress tracking dashboards |
Differentiated Instruction Research
Tomlinson's framework (2001) guides our approach to learner variability:
Differentiation by Readiness
- Tiered activities with multiple entry points
- Flexible pacing through self-directed modules
- Scaffolded complexity building from concrete to abstract
Differentiation by Interest
- Choice in project topics within AI application domains
- Connection to student passions (art, sports, social justice)
- Career exploration aligned with individual aspirations
Differentiation by Learning Profile
- Modality options for instruction and assessment
- Environmental preferences (individual, pair, group work)
- Cultural responsiveness in examples and applications
Cognitive Load Theory
Sweller's research (1988, 2019) informs our instructional design:
- Reduce extraneous load: Clear visual design, consistent navigation
- Manage intrinsic load: Concept sequencing from simple to complex
- Optimize germane load: Active learning activities that build schemas
Application to AI Education
AI concepts involve high intrinsic complexity. We address this through:
- Unplugged activities that isolate concepts from tool complexity
- Worked examples showing complete solutions before practice
- Fading scaffolds that gradually release responsibility
- Spaced practice returning to concepts across grade levels
Each curriculum module includes a "Research Foundation" section documenting the empirical basis for design decisions, enabling continuous validation and refinement.
Measurement Alignment
Our assessment approach draws on assessment science:
- Formative assessment for ongoing feedback (Black & Wiliam, 1998)
- Performance-based assessment for authentic demonstration (Wiggins, 1989)
- Portfolio assessment for comprehensive competency evidence (Paulson et al., 1991)
These approaches are particularly effective for diverse learners, reducing bias inherent in standardized testing while providing richer evidence of learning.