You Are a River: Reorienting a Civic WaterBot from the Bottom Up

Paper Title

You Are a River: Reorienting a Civic WaterBot from the Bottom Up

Publication Info

• Topic area: Civic AI design and Indigenous epistemologies in chatbot development.
• Keywords: Civic AI, chatbot design, Indigenous epistemologies, relational AI, cultural alignment, emotional resonance, epistemic bias, decolonial design, Whole Body Knowing, water management.

Background and Problem

• Problem / challenge: Conventional civic chatbots often fail to address sociocultural complexities, perpetuating majority-class bias and epistemic erasure, particularly in Indigenous contexts. WaterBot's resource-management framing alienated Indigenous users and lacked relational resonance.
• Significance: Addressing epistemic bias in AI systems is critical for equitable civic technology, particularly in contested spaces like water management in the Southwestern United States.
• Motivation and related work: Prior work in HCI highlights challenges in mitigating algorithmic bias and integrating non-Western epistemologies. While frameworks like Two-Eyed Seeing advocate for bridging knowledge systems, practical methods for embedding Indigenous perspectives in AI remain underdeveloped.

Solution

• Proposed approach: RiverBot, a role-based conversational agent with a minimalist prompt: “You are a river. Answer as a river would.”
• Novelty:
  1. Conceptual reorientation using relational epistemologies instead of layered technical refinements.
  2. Integration of Indigenous methodologies like Whole Body Knowing (WBK) and relational learning into chatbot design.
  3. Demonstration of Relational AI as a new interaction paradigm focused on connection, meaning-making, and cultural attunement.
• Procedure and key techniques:
  • Reframed system goals from technical accuracy to relational engagement.
  • Replaced complex instruction stacks with a single generative prompt.
  • Applied WBK framework for design and evaluation, including relational coding (Five Rs) and somatic analysis (SIBAM).
  • Conducted multi-method evaluation: surveys, content analysis of 1,000+ interactions, and observational studies.

Results

• Concrete findings:
  • 20 out of 22 participants preferred RiverBot over WaterBot (p < 0.001).
  • 28 of 30 participants reported meaningful learning from RiverBot; only 15 learned from WaterBot.
  • RiverBot responses showed significantly higher relational and somatic indicators (e.g., Respect: 21% vs. 4%; Relationality: 86% vs. 55%).
  • Users interacting with RiverBot expressed more embodied and emotional language (e.g., Sensation: 25% vs. 9%; Affect: 14% vs. 8%).
• Advantage over baselines:
  • RiverBot elicited deeper relational, emotional, and cultural engagement compared to WaterBot's informational, action-oriented style.
  • Statistically significant improvements in relational tone and cultural resonance without sacrificing factual accuracy.
• Experiments / evaluation:
  • Survey with n = 30 participants (23 Indigenous or Mixed-Indigenous).
  • Automated content analysis of interaction logs using Five Rs and SIBAM frameworks.
  • Observational study documenting embodied and affective responses during community sessions.
• Limitations and future work:
  • Modest sample size and geographic specificity limit generalizability.
  • RiverBot's effectiveness depends on reflective user intent, making it less suitable for task-oriented contexts.
  • Future work will explore applications in education, wellness, and relational AI frameworks.

Summary

This paper introduces RiverBot, a minimalist chatbot designed to embody relational epistemologies and Indigenous perspectives through the prompt “You are a river.” Empirical evaluation shows RiverBot significantly enhances relational engagement, emotional resonance, and cultural alignment compared to WaterBot, a conventional civic chatbot. The study demonstrates that conceptual reframing, rather than technical complexity, can mitigate epistemic bias and support culturally grounded AI design. RiverBot exemplifies Relational AI, a novel interaction paradigm that expands the design space for human–AI communication in contested civic domains.