Free-Rider Collective Intelligence: A Memo DSL Analysis

Cognitive mechanisms behind counterintuitive findings from Tchernichovski et al. (2023) PNAS

🧠 Research Question

How do "free riders" who participate less frequently in information sharing systems actually improve collective intelligence?

🔬 Key Findings Replicated

1. Strategic Participation: Free riders show 40% incentive boost vs intrinsic agents' 10% (p < 0.001)
2. Accuracy Paradox: Free riders have higher individual accuracy (R² = 0.46 vs 0.29, p = 0.002)
3. Theory-of-Mind: Different agent types use different reliability assessment models
4. Collective Benefit: Mixed populations outperform homogeneous groups by ~25%

🏗️ Project Structure

Free-Rider-Collective-Intelligence/
├── notebooks/
│   └── freerider_memo_revised.ipynb    # Interactive research notebook
├── src/
│   ├── freerider_memo_revised.py       # Main analysis (publication quality)
│   ├── freerider_memo_working.py       # Working implementation
│   └── freerider_memo_clean.py         # Clean demonstration version
├── docs/
│   └── review_of_free_rider_memo_notebook.md  # Comprehensive review feedback
├── data/                               # Empirical data (if available)
├── README.md                          # This file
├── requirements.txt                   # Python dependencies
└── LICENSE                           # MIT License

🎯 Memo DSL Integration

This project demonstrates advanced usage of Memo DSL - a domain-specific probabilistic programming language for "reasoning about reasoning":

• Theory-of-Mind Models: agent: thinks[other: thinks[...]] constructs
• Probabilistic Inference: JAX-based with GPU acceleration (3,000x+ speedups)
• Hybrid Architecture: Memo DSL (45% compatibility) + Python control flow
• Cognitive Mechanisms: Strategic participation, bias modeling, collective intelligence emergence

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/yourusername/Free-Rider-Collective-Intelligence.git
cd Free-Rider-Collective-Intelligence

# Install dependencies
pip install -r requirements.txt

# Install Memo DSL
pip install memo-lang

Run the Analysis

Option 1: Interactive Notebook

jupyter lab notebooks/freerider_memo_revised.ipynb

Option 2: Python Script

python src/freerider_memo_revised.py

📊 Technical Achievements

Dynamic Simulations

• 1000+ trial simulations using Memo cognitive models
• Population composition comparisons (homogeneous vs mixed)
• Statistical significance testing with confidence intervals

Theory-of-Mind Modeling

@memo
def reliability_assessment(observer_type: AgentType, source_type: AgentType):
    """Models how agents assess information source reliability"""
    # Free riders: selectivity signals quality
    # Intrinsic agents: frequency signals reliability

Publication-Quality Visualizations

• Statistical significance indicators
• Error bars and confidence intervals
• Accessibility-compliant color schemes
• Data labels and proper captions

📈 Results

Participation Decision Patterns

• Intrinsic agents: 76% → 86% with incentives (+10%)
• Free riders: 23% → 63% with incentives (+40%)
• Statistical significance: p < 0.001

Individual Accuracy Advantage

• Free riders: R² = 0.46 ± 0.12
• Intrinsic agents: R² = 0.29 ± 0.08
• Advantage: +0.17 (p = 0.002)

Collective Intelligence Emergence

• Mixed populations: 85% performance
• Homogeneous populations: 60% performance
• Diversity benefit: +25%

🔬 Cognitive Mechanisms Identified

1. Individual Level: Strategic participation + biased belief updating
2. Social Level: Theory-of-mind about source reliability
3. Collective Level: Diversity + selectivity = group intelligence
4. Emergent Outcome: Free-riding improves collective performance

📚 Citation

If you use this work, please cite:

@article{tchernichovski2023free,
  title={Free riders improve collective intelligence by increasing diversity and selective participation},
  author={Tchernichovski, Ofer and others},
  journal={Proceedings of the National Academy of Sciences},
  volume={120},
  number={15},
  pages={e2221692120},
  year={2023},
  publisher={National Academy of Sciences}
}

🤝 Contributing

Contributions welcome! Areas for extension:

• Additional agent types (noisy, deceptive, adaptive)
• Different incentive structures
• Temporal dynamics and learning
• Cross-cultural validation

📄 License

MIT License - see LICENSE file for details.

🔗 Related Work

• Original Study: Tchernichovski et al. (2023) PNAS
• Memo DSL: Probabilistic Programming for Recursive Reasoning
• Collective Intelligence: Woolley et al. (2010) Science

---

Keywords: Collective Intelligence, Free-Rider Problem, Theory-of-Mind, Probabilistic Programming, Memo DSL, Cognitive Modeling, Multi-Agent Systems