Skip to content

Programmable Human Behaviour Through Evidence-Based Agent Coordination - Research with Strong Ethical Framework

License

Notifications You must be signed in to change notification settings

devvyn/behaviour-lab

Repository files navigation

Behaviour Lab 🧠

Programmable Human Behaviour Through Evidence-Based Agent Coordination

A comprehensive research and development project exploring how AI agents and intelligent document staging can ethically prime and nudge human behavior toward self-chosen goals.


⚠️ ETHICAL USE ONLY - READ THIS FIRST ⚠️

This research describes powerful behavior influence techniques.

By accessing this work, you acknowledge and agree to:

Beneficence: Use ONLY to genuinely benefit users, NOT for manipulation or profit at user expense ✅ Informed Consent: Obtain explicit, informed consent for any behavior change attempts ✅ Autonomy: Respect user autonomy, provide easy opt-out, preserve choice ✅ Transparency: Be honest about intentions and methods when asked ✅ Non-Maleficence: Do no harm, including subtle psychological harm ✅ Reversibility: Allow users to undo any system suggestions or changes

❌ PROHIBITED USES:

🚫 Commercial manipulation without informed consent 🚫 Exploitation of vulnerabilities or cognitive limitations 🚫 Dark patterns, deception, or hidden persuasion 🚫 Coercion or removal of meaningful choice 🚫 Use on populations unable to consent (children without guardian approval, etc.) 🚫 Any application that prioritizes system/company benefit over user welfare

🔴 WARNING:

Misuse of these techniques is unethical and may be illegal in your jurisdiction.

This research is published to:

  • Enable defensive awareness (recognize manipulation)
  • Establish ethical standards for the field
  • Support beneficial applications (education, therapy, personal development)
  • Advance scientific understanding

If you observe misuse, please report to: [Contact information TBD]

📜 License:

This work is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International with additional ethical use requirements (see LICENSE.md).

Academic/Research Use: Encouraged with proper attribution Therapeutic/Educational Use: Encouraged with informed consent Commercial Use: Requires explicit permission and ethical review


🎯 Mission

Create a behaviour engineering system that amplifies human agency through:

  • Evidence-based priming and nudging techniques
  • Multi-agent coordination and consistency
  • Intelligent information architecture
  • Transparent, ethical influence
  • Continuous measurement and adaptation

Core Principle: Technology should enhance human flourishing, not diminish autonomy.


📚 Project Structure

behaviour-lab/
├── README.md                          # This file
├── docs/
│   └── system-design-comprehensive.md # Complete system architecture
├── research/
│   ├── priming-foundation.md          # Priming: theory, mechanisms, evidence
│   ├── nudging-techniques-catalog.md  # Complete nudging techniques library
│   ├── human-vs-agent-priming.md      # Critical comparison + integration
│   └── narrative-priming-identity.md  # AI narratives & identity (philosophical foundations)
├── src/
│   └── priming_analyzer.py            # Core analysis tool (Python)
├── experiments/                        # Experimental designs (TBD)
├── data/                              # Measurement data (TBD)
└── docs/                              # Additional documentation


🔬 Research Foundation

1. Priming Research (priming-foundation.md)

Key Findings:

  • Cognitive priming is REAL: Perceptual, semantic, and repetition priming have robust evidence
  • Social priming is DEAD: Failed replication, only 2 JESP studies in 2024
  • ⏱️ Time scales matter: Effects range from milliseconds (perceptual) to months (repetition)
  • 🧠 Neural mechanisms: fMRI and EEG studies show representational sharpening, spreading activation

Evidence Tiers:

  • HIGH (⭐⭐⭐⭐⭐): Repetition priming (d=0.8, 95% replication rate)
  • HIGH (⭐⭐⭐⭐): Semantic priming (d=0.6, 90% replication rate, short-term)
  • HIGH (⭐⭐⭐⭐): Perceptual priming (d=0.7, 92% replication rate)
  • MODERATE (⭐⭐⭐): Conceptual priming (d=0.4, 70% replication rate)

Critical Insight: Only use priming with neural mechanism support and successful replications.


2. Nudging Techniques (nudging-techniques-catalog.md)

Most Effective Techniques:

Technique Evidence Effect Size Duration Use Cases
Defaults ⭐⭐⭐⭐⭐ Large (0.8) Long Opt-in/out decisions
Framing ⭐⭐⭐⭐ Medium (0.6) Short Messaging, risk communication
Anchoring ⭐⭐⭐⭐ Large (0.7) Medium Pricing, negotiations
Social Proof ⭐⭐⭐⭐ Medium (0.5) Medium Behavior normalization
Commitment ⭐⭐⭐⭐ Medium (0.6) Long Goal achievement
Simplification ⭐⭐⭐⭐ Large (0.7) Long Process design
Feedback ⭐⭐⭐⭐ Medium (0.4) Ongoing Habit formation

Quick Reference Frameworks:

  • MINDSPACE: Messenger, Incentives, Norms, Defaults, Salience, Priming, Affect, Commitments, Ego
  • EAST: Easy, Attractive, Social, Timely

3. Human vs. Agent Priming (human-vs-agent-priming.md)

Key Discoveries:

Humans:

  • Neurobiological priming with unconscious effects
  • Limited working memory (7±2 items)
  • Exponential decay functions
  • Embodied, emotional, social

AI Agents:

  • Attention-based "priming" via context
  • Vast context windows (4K-200K+ tokens)
  • No decay (within context)
  • Extremely sensitive to:
    • Instruction phrasing (40% accuracy swings)
    • Example ordering (>40% differences)
    • Format changes (76 accuracy points)

The Bridge:

Agent outputs can prime humans through:

  • Semantic content (word associations)
  • Visual formatting (perceptual priming)
  • Conceptual frameworks (mental models)
  • Behavioral demonstrations (observational learning)
  • Sequential staging (cascade priming)

Critical Asymmetry: Agents are MORE sensitive to surface features than humans, but LESS sensitive to embodied/emotional cues.


🏗️ System Architecture

High-Level Components

┌──────────────────────────────────────────┐
│         HUMAN OPERATOR(S)                │
│  Perception → Cognition → Decision       │
└─────────┬────────────────────┬───────────┘
          │ Priming            │ Feedback
          ↓                    ↑
┌─────────────────────────────────────────┐
│    BEHAVIOUR ENGINEERING LAYER          │
│                                         │
│  • Document Staging Engine              │
│  • Multi-Agent Orchestrator             │
│  • Priming Choreography                 │
│  • Nudge Execution                      │
│  • Measurement & Adaptation             │
│  • User Model (cognitive profile)       │
└─────────┬───────────────────┬───────────┘
          │                   │
          ↓                   ↑
┌─────────────────────────────────────────┐
│         AGENT ECOSYSTEM                  │
│  Research │ Planning │ Execution         │
│  Review │ Teaching │ Support             │
└─────────────────────────────────────────┘

Core Capabilities

1. User Modeling

  • Cognitive style profiling
  • Priming susceptibility estimation
  • Learning rate tracking
  • Behavioral history analysis

2. Document Staging

  • Semantic chunking (respects working memory limits)
  • Temporal staging (spacing effects)
  • Progressive disclosure
  • Adaptive complexity
  • Multi-modal presentation

3. Multi-Agent Orchestration

  • Specialized agent roles (research, planning, execution, review, teaching, support)
  • Consistency enforcement across agents
  • Collaborative problem-solving
  • Behavioral modeling through agent interactions

4. Priming Choreography

  • Multi-stage priming sequences
  • Cross-modal priming (visual + semantic + behavioral)
  • Semantic network activation
  • Adaptive intensity control
  • Repetition with variation (spacing effect)

5. Nudge Execution

  • Evidence-based nudge library
  • Context-appropriate selection
  • Combination effects
  • A/B testing framework

6. Measurement & Adaptation

  • Real-time cognitive load estimation
  • Behavioral outcome tracking
  • Longitudinal persistence measurement
  • Predictive modeling
  • Continuous adaptation

🧪 Demo: Priming Analyzer

Run the core priming analyzer to see the knowledge base in action:

cd ~/Documents/GitHub/behaviour-lab
python3 src/priming_analyzer.py

Output includes:

  • Knowledge base summary (4 priming effects)
  • Detailed effect profiles with confidence scores
  • Paradigm recommendations for specific use cases
  • Temporal dynamics analysis (effect strength over time)

Example output:

Direct Repetition Priming (repetition)
  Evidence: HIGH (confidence: 0.95)
  Effect size: d=0.80
  Replications: 95/100
  Mechanism: Neuronal Sharpening & Response Learning
  RT Advantage: ~50ms

🎯 Use Cases

1. Learning & Education

  • Spaced repetition for long-term retention
  • Progressive disclosure for complex topics
  • Multi-agent teaching with consistency
  • Behavioral modeling of problem-solving

2. Habit Formation

  • Identity priming before behavior priming
  • Environmental design guided by agents
  • Commitment devices with social accountability
  • Feedback loops for reinforcement

3. Decision Support

  • Framing optimization for risk communication
  • Default selection for optimal choices
  • Anchoring for reference point establishment
  • Simplification to reduce cognitive load

4. Productivity & Workflow

  • Implementation intentions (if-then planning)
  • Salience for important information
  • Feedback on progress and patterns
  • Agent demonstrations of best practices

5. Knowledge Work

  • Semantic priming before complex concepts
  • Document staging for optimal comprehension
  • Cognitive load management during information delivery
  • Multi-agent support for different work phases

🛡️ Ethical Framework

Core Principles

  1. Beneficence: System must genuinely benefit user
  2. Autonomy: User retains final authority
  3. Transparency: Explainable intentions
  4. Non-maleficence: Do no harm (including subtle psychological harm)
  5. Justice: Fair treatment, no exploitation
  6. Privacy: Minimal data, strong protection
  7. Reversibility: User can undo any suggestion
  8. Consent: Informed consent for behavior change attempts

Safeguards

Ethical Evaluation: Every intervention evaluated against principles ✅ Consent Management: Explicit, informed consent with periodic refresh ✅ Continuous Monitoring: Real-time detection of ethical violations ✅ Dark Pattern Detection: Automatically flags manipulative design ✅ User Override: User can reject, query, or reverse any intervention ✅ Audit Trail: Complete log of all interventions and outcomes

Red Flags (Automatic Rejection)

🚩 System benefits more than user 🚩 Deceptive framing or hidden manipulation 🚩 Exploitation of vulnerabilities 🚩 Bypasses conscious awareness (unless user-consented research) 🚩 Difficult opt-out or reversal 🚩 Based on discredited science


📊 Current Status

✅ Completed

  • Comprehensive research foundation (3 major documents)
  • Evidence-based priming taxonomy
  • Complete nudging techniques catalog
  • Human vs. agent priming analysis
  • System architecture design
  • Ethical framework specification
  • Priming analyzer prototype (Python)
  • Implementation roadmap

🚧 In Progress

  • Multi-persona research review (NEXT STEP)
  • MVP implementation (Phase 1)
  • Initial user pilot study
  • Measurement validation

📋 Planned

  • Multi-agent coordination implementation
  • Document staging engine
  • User modeling system
  • Production deployment
  • Research publication

🚀 Getting Started

Prerequisites

python3 -m pip install numpy scipy networkx matplotlib pandas

Quick Start

  1. Explore the research:

    cd ~/Documents/GitHub/behaviour-lab/research
    cat priming-foundation.md | head -100
  2. Run the priming analyzer:

    python3 src/priming_analyzer.py
  3. Review the system design:

    open docs/system-design-comprehensive.md
  4. Understand the roadmap:

    • Phase 1 (Months 1-2): Foundation
    • Phase 2 (Months 3-4): Multi-agent coordination
    • Phase 3 (Months 5-6): Advanced features
    • Phase 4 (Months 7-12): Deployment & iteration

🤝 Contributing

This project is currently in the research and design phase. Contributions welcome in:

  • Research review: Validate findings against latest literature
  • Ethical analysis: Strengthen ethical framework
  • Implementation: Build core components
  • Experimentation: Design and run studies
  • Documentation: Improve clarity and accessibility

Next Priority: Multi-Persona Research Review

Before implementing, we need diverse expert perspectives to validate research and identify blind spots.


📖 Key Documents

Document Purpose Length Status
priming-foundation.md Priming theory, evidence, mechanisms 25 pages ✅ Complete
nudging-techniques-catalog.md All nudging methods with evidence 30 pages ✅ Complete
human-vs-agent-priming.md Comparative analysis + integration 35 pages ✅ Complete
system-design-comprehensive.md Complete system architecture 50+ pages ✅ Complete

Total Documentation: ~140 pages of research-backed design


🎓 Research Sources

Priming

Nudging & Choice Architecture

AI & Prompting


🌟 Vision

Near-term (2025-2026):

  • Working MVP with single-user effectiveness
  • Published research on agent-human priming
  • Open-source core components
  • Small-scale pilots (n=100)

Mid-term (2026-2028):

  • Multi-agent coordination at scale
  • Personalized priming schedules
  • Cross-cultural adaptation
  • Large-scale deployments (n=10,000+)

Long-term (2028+):

  • Seamless human-AI collaborative cognition
  • Evidence-based behavioral health tools
  • Educational transformation
  • Enhanced human agency and autonomy

Ultimate Goal: A world where technology genuinely amplifies human capacity while respecting and enhancing individual autonomy.


⚠️ Important Notes

What This Is

✅ Research-based behavior engineering ✅ Transparent, consensual influence ✅ User-goal aligned interventions ✅ Evidence-tier classification ✅ Continuous ethical monitoring

What This Is NOT

❌ Manipulation for commercial gain ❌ Hidden persuasion techniques ❌ Exploitation of vulnerabilities ❌ Social engineering attacks ❌ Dark patterns or deception

Critical Distinctions

Nudging ≠ Manipulation

  • Nudging preserves choice and autonomy
  • Manipulation restricts options or deceives

Priming ≠ Mind Control

  • Priming facilitates processing
  • Effects are subtle and context-dependent
  • User always retains agency

Behavior Engineering ≠ Coercion

  • Engineering creates environments that support goals
  • Coercion removes choice

📞 Contact & Collaboration

Project Lead: Behaviour Lab Research Team Repository: ~/Documents/GitHub/behaviour-lab/ Status: Research & Design Phase Next Milestone: Multi-Persona Review

Collaboration Opportunities

🤝 Research Partners: Psychology, HCI, AI ethics experts 🤝 Implementation: Software engineers, UX designers 🤝 Ethical Review: Ethicists, privacy advocates 🤝 User Studies: Research participants, pilot users


📄 License

License to be determined - likely open-source with ethical use restrictions

Core principles:

  • Open research and findings
  • Ethical use requirements
  • No manipulation for profit
  • User welfare priority

🙏 Acknowledgments

Research Foundations:

  • Daniel Kahneman, Amos Tversky (heuristics & biases)
  • Richard Thaler, Cass Sunstein (nudge theory)
  • John Bargh (priming research - and subsequent replication crisis lessons)
  • UK Behavioural Insights Team (practical applications)
  • Replication Index (critical evaluation)

This work stands on the shoulders of giants and learns from both their successes and failures.


Last Updated: 2025-12-02 Version: 1.0.0 Next Review: After multi-persona review


🎯 Next Steps

  1. Multi-Persona Research Review ← YOU ARE HERE
  2. Human sign-off on research and design
  3. Begin Phase 1 implementation (MVP)
  4. Initial user pilot study
  5. Iterate based on evidence

Ready to transform human-AI interaction through evidence-based behavioral engineering.

About

Programmable Human Behaviour Through Evidence-Based Agent Coordination - Research with Strong Ethical Framework

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •  

Languages