• Existing literature prioritizes technical fixes for scaling up deliberation—such as handling volume and structure—over the nuanced requirements of high-quality deliberation:
    • Homogenous test groups: Current design relies on WEIRD
    • Limitations of automated tools: automated tools manage data volume, but lack the discretion and nuanced intervention capabilities of human facilitators, such as mirroring perspectives or sensitively enforcing norms. These tools often impose expert-driven ideals regarding what constitutes a “valid” argument, excluding those who rely on storytelling or personal experience.
    • Exclusionary structures: technical designs overlook how pre-existing social inequalities, like difference in cognitive abilities, literacy, ore self-efficacy, affect participation.
  • Process recommendations
    • Utilize “Solution Generation” for Synthesis: Rather than forcing users to read all comments, use AI to synthesize scattered discussions into a cohesive list of proposed success metrics or answers, reducing the cognitive load of decision-making.
    • Design for “Perspective-Taking”: Address the “unawareness of others’ backgrounds” by explicitly prompting users to consider the viewpoints of diverse stakeholders (e.g., marginalized groups affected by the public service) before they finalize their voting or suggestions.
    • Prioritize “Argumentative Quality” Metrics: When evaluating your tool’s success, move beyond engagement counts (quantity) and measure “deliberativeness,” “persuasiveness,” and “constructiveness” to ensure the tool is actually producing useful policy decisions.