Agent parliament explores how LLM agent populations might anticipate collective responses to policy decisions and support more participatory forms of democratic engagement.
The first tool generates personas at scale. LLMs behave quite differently depending on the persona they're given, accessing different slices of their training distribution. I've tested this by generating 343 agents representing one million people each, with high fidelity demographic details.
The second tool runs dialogues between these agents. It surfaces key segments, overlooked opinions, and sensitive positions requiring careful handling helping users make sense of where a population might agree, diverge, or need bridging.
One research direction: how such a tool could help people imagine collective perspectives, empathize with viewpoints that aren't default to them, or stress-test proposals against a representative sample before deployment on high-impact civic or policy questions.
A second direction for controlled study: What if participants had LLM representatives? This addresses a core bottleneck of direct democracy: people have finite cognitive bandwidth and lack either the knowledge across every issue or the time to engage. Participants could operate in automatic mode (agent votes), manual (agent recommends, person chooses), or delegated in a liquid democracy fashion (passing certain domains to other participants they trust). Crucially, participants would retain control over which domains they vote on or delegate. This research could open new avenues on how LLMs might help coordinate large groups, study failure modes, model sensitivities, or explore how we may verify the legitimacy of delegation in agent-based systems.
<aside> 💡
To improve dialogue quality and representation accuracy, each persona could act as an independent LLM interacting with others
</aside>
https://chatgpt.com/share/681b92e7-f9d0-8013-88a6-37635905ce73
**PART 1 – “Town‑Hall‑Style” Discussion (excerpted highlights)**Moderator: “Welcome, everyone. President Trump is considering folding USAID into the State Department and slashing most foreign‑aid programs. Speak freely; he wants to hear why he should —or shouldn’t—pull the trigger.”
01 White Gen Z Male (21.1 M | 6.3 %) “I’m a senior at a state university. I get the ‘America First’ pitch, but I also see climate disasters, pandemics, and refugee flows on TikTok every day. If we walk away, China will fill the vacuum. That hurts U.S. jobs long‑term.”
02 White Gen Z Female (21.1 M) “Foreign aid is barely 1 % of the budget. Cutting it won’t lower my student‑loan bill, but it will raise the number of girls abroad who can’t go to school. That feels wrong.”
03 White Millennial Male (22.2 M) “I’m a veteran‑turned‑contractor. USAID grants kept local allies employed in Afghanistan; that kept me safer. Roll it into State if you want efficiency, but don’t gut the programs.”
04 White Millennial Female (22.2 M) “As a small‑business owner exporting baby‑formula tech, USAID’s Trade Hubs opened markets for me in East Africa. Kill the hubs, and you hurt my bottom line.”
05 White Gen X Male (21.1 M) “Tax burden’s real. The Pentagon has plenty of security‑cooperation tools; why duplicate? I’m fine merging USAID—just make sure career diplomats don’t slow things down even more.”
06 White Gen X Female (21.1 M) “I’m PTA president in rural Ohio. Folks here want potholes fixed, not wells in Liberia. Show me domestic wins first; then maybe I’ll back overseas charity.”
07 White Boomer Male (25.5 M | largest single block) “Vietnam vet, Social Security recipient. We spent trillions abroad and got chaos. Shut the spigot. If aid matters to national security, let State run a lean version—but slash at least half.”
08 White Boomer Female (25.5 M) “Church mission trips rely on USAID seed money. I’m fiscally conservative, but my faith calls me to help the poor. Reform, yes; dismantle, no.”
09 White Silent+ Male (8.9 M) “At 80 I remember the Marshall Plan. Aid built allies that bought Buicks from Detroit. Don’t repeat the 1930s isolation mistake.”
10 White Silent+ Female (8.9 M) “Hospitals in Africa named after American nurses are goodwill we can’t buy with bombs. Keep that brand alive.”
11 Hispanic Gen Z Male (10.2 M) “Family WhatsApp lights up when hurricanes hit Puerto Rico and Honduras. USAID shows up first. Abandoning them looks like we don’t care about Latinos—inside or outside our borders.”
12 Hispanic Gen Z Female (10.2 M) “Migration pressures rise when aid drops. You want fewer people at the border? Keep development dollars flowing.”
13 Hispanic Millennial Male (8.6 M) “I send remittances home. USAID training helped my cousins start a coffee co‑op so they didn’t have to trek north. Economic aid is border security.”
14 Hispanic Millennial Female (8.6 M) “I vote in Arizona. Cuts feel anti‑Latino. Republicans need our swing vote.”
15–20 Hispanic Gen X through Silent+ echo variants of the above: skeptical of waste but see aid as migration‑control and soft‑power tool; older cohorts emphasize anti‑communism in Latin America.
21 Black Gen Z Male (5.9 M) “Flint still doesn’t have clean water. Fix home first, but keep HIV/AIDS funding in Africa—it’s a bipartisan success.”
22 Black Gen Z Female (5.9 M) “Cuts usually hit women’s health. That’s a non‑starter.”
23–30 Black Millennial to Silent+ stress moral leadership, link foreign and domestic justice, recall USAID work in Ebola and Haiti. Older Boomers tie aid to Cold‑War civil‑rights victories; some Silent‑era voices warn against ‘charity when our own neighborhoods crumble.’