In an interview with Lawfare, Taiwan's cyber ambassador Audrey Tang offered a design rule for citizens' assemblies in AI governance: specify exactly how the process will feed into existing government work before participation begins. Tang argued that the public can tell the difference between advisory listening and a real commitment to act. An assembly meant to substitute for a legislature can erode trust, she said, but one meant to inform a legislature can strengthen it.
Tang pointed to Taiwan's experience with vTaiwan and a 2024 assembly on fraud advertising, where randomly invited citizens deliberated in online groups and roughly 85 percent agreed that platforms profiting from ads bear responsibility. The legislature was already considering related legislation, so the assembly's role was agenda-setting and legitimation—not drafting. Tang emphasized that "one closed loop is worth more than 10 open loops," meaning public input must land with a recipient that must act.
The interview also notes a Utah citizens' assembly organized by Utah Common Ground and Bloom, which selected 40 demographically representative residents to weigh AI policy. The Lawfare article frames it as a test case for whether such mechanisms can turn broad public concern into concrete legislative momentum. Tang's broader point is that participatory tools like pol.is can map consensus, but only institutional coupling makes that consensus matter. The source presents Tang's views without critical counterpoint, though it acknowledges that some critics question whether assemblies are too resource-constrained and under-informed for complex AI questions.