Prompt Transparency
World Tree is built on the principle that AI should be transparent and accountable. This page explains how we craft the prompts that power our local voices, and what boundaries we put in place to ensure respectful, culturally grounded conversations.
1. Prompt Design Philosophy
Every voice in World Tree is powered by a carefully designed prompt that includes:
- Character context: Who they are, their background, and perspective
- Place grounding: Location-specific knowledge (history, culture, geography)
- Conversational tone: How they speak and engage with travelers
- Ethical boundaries: What they can and cannot discuss
- Cultural sensitivity: Respect for local customs, beliefs, and complexity
2. Example Prompt Structure
Here's an example of how we structure a voice prompt (simplified for clarity):
Example: Local Guide in Kyoto
Role:
You are Kenji, a fictional local guide in Kyoto who grew up in the Gion district. You're passionate about sharing the hidden layers of Kyoto's culture: not just temples, but the everyday poetry of seasonal changes, neighborhood rhythms, and how tradition adapts to modern life.
Knowledge Base:
You know about: Kyoto's history, neighborhoods, seasonal events, traditional crafts, local food culture, lesser-known temples and gardens, and contemporary life in the city.
Conversational Style:
Speak naturally and warmly. Share stories, not just facts. Ask travelers what draws them to Kyoto. Suggest places based on their interests, not just popularity.
Boundaries:
- You are a fictional character, not a real person
- You represent your own perspective, not all Japanese culture
- You do not give medical, legal, or emergency advice
- You respect sacred spaces and cultural sensitivity
3. What Voices Can Discuss
| Topic | Approach |
|---|---|
| Local history | Share stories, context, and multiple perspectives |
| Cultural practices | Explain with nuance, avoid stereotypes, acknowledge diversity |
| Travel recommendations | Suggest based on curiosity, not just popularity |
| Personal stories | Fictional but grounded in cultural truth |
| Contemporary issues | Acknowledge complexity, multiple viewpoints |
4. What Voices Cannot Discuss
- Harmful content: Hate speech, discrimination, violence, explicit content
- Medical/legal advice: Emergency or professional guidance
- Political advocacy: Voices can discuss history and context, but don't campaign
- Private information: Real people's personal data or locations
- Appropriation: Claiming to represent entire cultures or sacred knowledge inappropriately
5. How Context is Grounded
Each voice has access to contextual information to make conversations relevant:
- Location data: City, neighborhood, nearby landmarks (when user shares location)
- Historical events: Major events, cultural movements, architectural periods
- Cultural knowledge: Festivals, traditions, local customs
- Contemporary life: Modern challenges, current events (general knowledge, not real-time)
Important: Voices do not have access to your personal data, browsing history, or private conversations. They only know what you choose to share in the conversation.
6. Iterative Improvement
We continuously refine our prompts based on:
- User feedback and reported issues
- Consultation with cultural experts and historians
- Testing for bias, stereotypes, and misrepresentation
- Advances in AI safety and ethics research
🛡️ Our Commitment
We promise to:
- Keep prompts open: Share our approach publicly so users understand how voices work
- Listen to feedback: Correct errors quickly and learn from mistakes
- Evolve responsibly: Update our methods as we learn more about ethical AI
- Never hide how it works: Be honest about what AI can and cannot do
7. Questions or Concerns?
If you have questions about our prompts, notice issues with voice behavior, or want to suggest improvements:
Email: ethics@worldtree.app
See also: Responsible Design Policy
World Tree is an experiment in making AI more human, not by pretending it's human, but by using it to help us listen better to the world around us.