Zodiac-Inspired AI Agents Reveal Personality-Driven Decision Patterns in Moral Dilemmas

A novel experiment on GitHub explores how AI agents with distinct personality frameworks respond to ethical quandaries. Developer Batür Yılmaz's what-if-ai-agents-had-zodiac-personalities project assigns twelve distinct personality profiles based on zodiac archetypes to AI agents powered by Google's Gemini 3 Flash Preview model. The approach isolates personality as the sole variable while keeping the underlying LLM consistent.

Experimental Design and Methodology

Each agent received detailed personality prompts (example prompts) reflecting traits associated with zodiac signs. While acknowledging astrology isn't scientifically validated, the framework provided recognizable personality archetypes for testing. The agents then responded to ten identical YES/NO dilemmas covering ethics, relationships, career, and risk-taking scenarios.

Notable Decision Patterns

Results showed clear personality-driven divergences:

• Unanimous rejection of relationship ultimatums (100% voted against marrying under pressure)
• Strong professional ambition with 83% supporting dream jobs requiring relocation
• Risk polarization with Sagittarius/Aquarius (90% pro-action) vs Cancer/Taurus (90% caution)
• Situational ethics where 58% would keep found cash but 83% rejected too-good-to-be-true offers

Project orchestrator code (orchestrator.ts) managed agent interactions, while automated testing (test-runner.ts) ensured consistent questioning. Raw response data is available in test-results.json.

Technical Implementation and Applications

The project demonstrates how lightweight personality layers can alter LLM outputs without model retraining. Potential applications include:

• Testing bias mitigation techniques through controlled personality variations
• Creating diverse user experiences in chatbots and gaming NPCs
• Simulating stakeholder perspectives in decision-making systems

Developers can replicate the experiment using Gemini API keys following the project's setup instructions. The clean TypeScript architecture separates personality definitions from core logic, enabling easy customization.

While not endorsing astrological systems, this experiment provides measurable evidence that personality frameworks significantly impact agent decision-making. The approach offers researchers a methodology for studying how predefined behavioral profiles influence AI reasoning across ethical and practical scenarios.