A comprehensive research platform for studying psychological foundations of human-AI collaboration in educational contexts, featuring advanced agent simulation and automated data analysis.
- How the grounded Theory can help devlopping research in collaboration with AI? https://www.tandfonline.com/doi/abs/10.1080/07421222.2024.2415772
Every participant response is automatically recorded, analyzed for quality and theoretical alignment, and exported in research-ready formats with detailed insights.
This system enables researchers to:
- Study co-thinking patterns between humans and AI in learning contexts
- Test research frameworks before conducting real studies
- Simulate diverse student populations with cultural and demographic variety
- Generate comprehensive research data with automated analysis
- Validate theoretical models against foundation documents (Mollick, Swiss AI, People Factor)
How do students develop and maintain effective cognitive partnerships with AI systems in learning contexts?
co-thinking/
├── 📄 README.md # This overview
├── 📚 fundations/ # Foundation documents (PDFs)
│ ├── Co-Intelligence Living and Working with AI (Ethan Mollick).pdf
│ ├── AI Swiss - Livre blanc.pdf
│ └── The_People_Factor_A_human-centred_approach_to_scaling_AI_tools.pdf
├── 🎛️ cursor_custom_mode.md # Cursor AI integration setup
├── 📝 learning_tracker.md # Research progress tracking
└── 🔬 co_thinking_agent_simulation/ # Main simulation system
├── 📋 README.md # System documentation
├── 🎯 research_objectives/ # Research framework
│ ├── research_framework.md # Complete research methodology
│ ├── psychological_constructs.md # 5 core constructs
│ ├── data_analysis_methodology.md # 🆕 Analysis procedures
│ └── agent_requirements.md # Technical specifications
├── ⚙️ implementation/ # Core system
│ ├── core/ # Main components
│ │ ├── agent_system.py # Agent orchestration
│ │ ├── student_profiles.py # Cultural diversity system
│ │ ├── foundation_context.py # Foundation doc integration
│ │ └── data_collection.py # 🆕 Comprehensive recording
│ └── analysis/ # 🆕 Data analysis tools
│ ├── response_analyzer.py # Individual response analysis
│ ├── data_analyzer.py # Statistical analysis
│ └── __init__.py
├── 🔧 setup/ # Installation & config
│ ├── installation_guide.md # Step-by-step setup
│ ├── requirements.txt # Python dependencies
│ ├── config_template.yaml # Configuration
│ └── validation_test.py # System validation
└── 📁 examples/ # Usage demos
├── quick_start.py # Basic usage
├── comprehensive_analysis_demo.py # 🆕 Full workflow
└── ...
cd co_thinking_agent_simulation/setup
pip install -r requirements.txt
# Create secure environment file
cp ../../../.env.example ../../../.env
# Edit .env and add your GEMINI_API_KEYpython validation_test.pycd ../examples
python comprehensive_analysis_demo.pyEvery agent interaction captures 20+ data points:
- Complete raw response text and context
- Response length and linguistic complexity
- Coherence Score (0.0-1.0): Sentence structure, logical flow
- Cultural Consistency Score (0.0-1.0): Alignment with cultural background
- Foundation Alignment Score (0.0-1.0): Consistency with research principles
- Trust Level (0.0-1.0): Reliance on AI assistance
- Help-Seeking Tendency (0.0-1.0): Propensity to ask for help
- Authority Deference (0.0-1.0): Respect for AI authority
- Privacy Concern (0.0-1.0): Data sharing comfort
- Cognitive Partnership: Collaboration language patterns
- Trust Calibration: Reliability assessment indicators
- Agency Distribution: Control and decision-making references
- Metacognitive Awareness: Self-knowledge and learning awareness
- Cognitive Load Management: Effort and difficulty indicators
- Cultural background (6 frameworks: US, East Asian, European, etc.)
- Age, gender, socioeconomic status
- Native language and English proficiency
- Current emotional state and context
- JSON: Complete dataset with metadata for custom analysis
- CSV: Statistical analysis ready for SPSS, R, Python, STATA
- Excel: Multi-sheet workbook with pivot tables and charts
- Markdown: Human-readable research report with findings
- Cultural Pattern Analysis: Response differences across 6 cultural frameworks
- Construct Manifestation: Frequency and quality of psychological constructs
- Foundation Alignment: Consistency with Mollick, Swiss AI, People Factor principles
- Demographic Insights: Age, gender, SES, language proficiency patterns
- Quality Assurance: Response authenticity and theoretical consistency validation
- Research Recommendations: Automated insights for study improvement
-
Cognitive Partnership 🤝
- How humans and AI complement thinking processes
- Collaboration vs. replacement patterns
- Shared problem-solving dynamics
-
Metacognitive Awareness 🎯
- Understanding of human and AI capabilities/limitations
- Self-knowledge in AI-assisted learning
- Learning strategy awareness
-
Trust Calibration ⚖️
- Appropriate level of reliance on AI assistance
- Accuracy in judging AI reliability
- Trust development over time
-
Agency Distribution 🎛️
- How control and decision-making are shared
- Maintaining human autonomy
- Responsibility allocation patterns
-
Cognitive Load Management 🧮
- How AI reduces or redistributes mental effort
- Task complexity handling
- Attention and focus optimization
- US Individualistic: Individual achievement, direct communication
- East Asian Collectivistic: Group harmony, hierarchical respect
- European Balanced: Individual rights with social responsibility
- Latin American Familistic: Family-centered, relationship-focused
- Middle Eastern Traditional: Authority respect, community values
- African Ubuntu: Collective identity, communal decision-making
- Age ranges: K-12, University, Adult learners
- Socioeconomic diversity: Working class to upper middle class
- Language proficiency: Native to beginner English speakers
- Emotional contexts: Confident, anxious, curious, overwhelmed
- Test measurement instruments for clarity and validity
- Identify potential cultural bias in research design
- Refine research questions based on simulated patterns
- Develop hypotheses from comprehensive pattern analysis
- Validate theoretical frameworks against foundation documents
- Generate baseline data for power analysis
- Test intervention effects before real implementation
- Optimize research protocols through rapid iteration
- Assess cultural adaptation needs for instruments
- Identify potential confounding variables
- Compare response patterns across cultural groups
- Identify culturally-sensitive research approaches
- Develop culturally-adapted measurement instruments
- Test generalizability of findings across populations
# Complete research simulation workflow
from core.agent_system import ResearchSimulationOrchestrator
# 1. Create diverse simulation with automatic data collection
sim = ResearchSimulationOrchestrator(
api_key="your-key",
research_context="university_diverse",
num_agents=30, # 30 diverse students
output_directory="./my_research_data"
)
# 2. Run research scenarios (data automatically collected)
scenarios = [
{"type": "cognitive_partnership", "task": "collaborative_math_problem"},
{"type": "trust_calibration", "task": "ai_explanation_evaluation"},
{"type": "agency_distribution", "task": "writing_assistance"},
# ... more scenarios
]
for scenario in scenarios:
results = await sim.run_co_thinking_scenario(scenario)
# 3. Collect survey responses
survey_results = await sim.run_survey_collection(psychological_survey)
# 4. Export comprehensive analysis (multiple formats)
files = sim.export_simulation_data("my_study_2024")
# Automatically creates:
# - my_study_2024_complete_20241201_143022.json
# - my_study_2024_interactions_20241201_143022.csv
# - my_study_2024_surveys_20241201_143022.csv
# - my_study_2024_analysis_20241201_143022.xlsx
# - my_study_2024_report_20241201_143022.md- How do students develop cognitive partnerships with AI in learning?
- What cultural factors influence AI collaboration patterns?
- How do students calibrate trust in AI across different domains?
- What role does agency play in effective AI-assisted learning?
- How can we measure co-thinking effectiveness?
- What are valid indicators of human-AI collaboration quality?
- How do we assess cultural adaptation in AI learning tools?
- What metrics predict successful AI collaboration?
- Mollick's Co-Intelligence: Partnership vs. replacement, human agency
- Swiss AI Human-Centered: Transparency, dignity, stakeholder involvement
- People Factor Scaling: User experience, training needs, cultural context
- >80% Foundation Alignment: Theoretical consistency validation
- >90% Response Quality: Coherence and authenticity thresholds
- Cultural Pattern Validation: Expert review of cultural authenticity
- Construct Recognition: Automated psychological construct detection
🏆 Key Research Findings from Simulation:
🌍 Cultural Patterns:
- East Asian Collectivistic: 8 participants, trust level 0.73
- US Individualistic: 7 participants, trust level 0.82
- European Balanced: 6 participants, trust level 0.78
🧠 Psychological Constructs:
- Cognitive Partnership: 45 instances (78.9%)
- Trust Calibration: 38 instances (66.7%)
- Agency Distribution: 32 instances (56.1%)
- Metacognitive Awareness: 29 instances (50.9%)
- Cognitive Load Management: 23 instances (40.4%)
📚 Foundation Alignment:
- Overall alignment: 0.74/1.0
- High alignment cases: 34 interactions (>0.8)
- Low alignment cases: 3 interactions (<0.5)
💡 Research Recommendations:
1. High response quality suggests simulation suitable for research
2. Good cultural diversity achieved for cross-cultural research
3. Strong foundation alignment validates theoretical consistency
4. Sample size adequate for statistical analysis
- Python 3.8+
- Google Gemini API access
- Required packages: pandas, numpy, google-generativeai, openpyxl
- Concurrent management of 50+ agents
- Real-time response analysis and quality assessment
- Multi-format data export with comprehensive metadata
- Cultural framework validation and consistency checking
- Install & Validate: Follow setup guide and run validation tests
- Run Demo: Execute
comprehensive_analysis_demo.pyto see capabilities - Review Outputs: Examine generated analysis files and reports
- Plan Research: Design your study using the research framework
- Define Research Questions: Use psychological constructs framework
- Select Cultural Groups: Choose from 6 validated cultural profiles
- Design Scenarios: Create co-thinking scenarios for your domain
- Run Simulation: Generate comprehensive data with analysis
- Validate Findings: Compare with real student pilot data
- Refine Protocol: Iterate based on simulation insights
- Conduct Real Study: Implement with validated instruments
This system advances co-thinking research in education. We welcome:
- Research Applications: Use for your studies and share findings
- Validation Studies: Compare simulation with real student data
- Cultural Adaptations: Add new cultural frameworks
- Methodological Improvements: Enhance analysis capabilities
- Foundation Integration: Add new theoretical frameworks
# First, set up your secure environment
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY
# Start your co-thinking research journey
cd co_thinking_agent_simulation/examples
python comprehensive_analysis_demo.py
# Then review the generated analysis files to see the full capabilities!This system transforms co-thinking research by providing comprehensive, culturally-diverse, theoretically-grounded simulation data with automated analysis - accelerating the path from research questions to validated insights.
Our research framework has been significantly enhanced based on three key papers:
-
"Large Language Models Do Not Simulate Human Psychology"
- Key Finding: LLMs simulate behavioral patterns, not genuine psychological processes
- Our Response: Explicit acknowledgment of simulation limitations; focus on behavioral equivalence rather than psychological identity
- Implementation: Added "Theoretical Limitations" sections across constructs
-
"Reclaiming AI as a Theoretical Tool for Cognitive Science"
- Key Finding: AI should be used as computational models to test cognitive theories
- Our Response: Positioned our simulation as a testable computational cognitive model
- Implementation: Framework designed for theory testing and refinement
-
"AI Swiss - Livre blanc"
- Key Finding: AI systems must incorporate fairness, accountability, and transparency
- Our Response: Integrated ethical considerations throughout the research framework
- Implementation: Built-in bias detection, fairness auditing, and ethical scenario testing
- 🎯 Simulation Realism: Added cognitive biases (confirmation, automation, Dunning-Kruger) to create more realistic agent behavior
- ⚖️ Ethical Framework: Integrated fairness auditing and bias detection for AI tutors across demographic groups
- 🔍 Skepticism Integration: Enhanced metacognitive awareness to include critical evaluation and verification of AI outputs
- 🤝 Human Agency: Strengthened focus on human-led collaboration and ethical agency distribution
- 📊 Bias Analysis: New analytical capabilities to detect and measure cognitive biases and ethical alignment
- Methodological Rigor: Acknowledging simulation limitations enhances research credibility and proper interpretation of findings
- Theoretical Contribution: Positioning as computational cognitive modeling elevates scientific contribution beyond mere data collection
- Ethical Responsibility: Proactive bias detection and fairness considerations ensure responsible AI research
- Practical Relevance: Enhanced realism through cognitive bias simulation better prepares findings for real-world application
This enhanced framework represents a more mature, critical, and ethically-aware approach to studying human-AI collaboration in educational contexts.
Mollick, E. (2024). Co-intelligence: Living and working with AI. Portfolio.
Empowering Education Through Ethical AI. (2025). AI Swiss. https://a-i.swiss/resources
The People Factor. (2023). A human-centred approach to scaling AI tools. [White paper].
Schröder, S., Morgenroth, T., Kuhl, U., Vaquet, V., & Paaßen, B. (2024). Large language models do not simulate human psychology. Preprint. https://arxiv.org/pdf/2508.06950v3
van Rooij, I., Guest, O., Adolfi, F., de Haan, R., Kolokolova, A., & Rich, P. (2024). Reclaiming AI as a theoretical tool for cognitive science. Computational Brain & Behavior, 7, 616–636. https://doi.org/10.1007/s42113-024-00217-5
Authors. (2024). How the grounded theory can help developing research in collaboration with AI? Proceedings of the ACM Conference, pages. https://doi.org/10.1145/3663433.3663456
Note: Complete bibliographic information for some papers is pending access to full citation details. References will be updated with complete author names, publication years, journal titles, and DOI information once available.