How might we introduce audio recording into democratic assemblies in a way which transparently communicates how and why the data will be used?

Overview

Inspired by the proven model of citizen assemblies worldwide, the MIT Center for Constructive Communication and partner DemocracyNext, designed, and implemented a tech-enhanced and student-focused version of a citizen assembly in January of 2025, building on a 2024 experimentation.

Since citizen assemblies involve days of breakout conversations, we were curious recording these conversations and using artificial intelligence (AI) to help discover topics with high consensus, under heard perspectives, and areas needing more input, might improve the quality of assemblies, and public trust that these assemblies sufficiently deliberated.

To help on-board delegates to audio recording and AI, I designed and developed three high-engagement interventions to transparently explain how we intend to use their data, and why electing to record has positive benefits to the collective goal of creating a more effective assembly.

The interventions were:

  1. Data bar

  2. Data comics

  3. Mic Pods

Partnership: MIT Center for Constructive Communication, DemocracyNext

Role: Lead designer, fabrication lead, researcher

Time: January 2025

Background

A citizen assembly is a gathering of people selected by lottery who broadly represent a community. Together, they spend multiple days learning, discussing and finding common ground so that they can write recommendations for policy makers that reflect the shared values of their community.

Audio recordings and artificial intelligence are natural fits for tech-enhancing democractic , since they give organizers and participants a way to understand the hundreds of hours of conversations that make up a democractic assembly.

However, gathering consent to record conversation audio during research is a point of tension for both participants and organizers in these settings. Our past work has found that:

  • Delegates care about how and why their data is being used, but legal consent forms (current solution) do not adequately communicate the value of long-form recording and AI

  • The dry nature of consent forms lowers how much delegates actually read or pay attention to them

  • Data processing with AI is difficult to understand and conceptualize without expert knowledge

  • Consent forms not inspire specific organizer-trust, since they are generic

DESIGN QUESTION

How might we design an experiential protocol for citizen assemblies which incorporates audio recording in a way that is trustworthy, easy, and empowering.

Data Bar

Data Comics

Mic Pods

Next Steps:

  • Spring 2025: controlled online experiment to compare engagement and trust using data comics compared to consent forms

  • Spring 2025: Release overview of qualitative findings captured during the experience