Talk to the City

Aligning decision-makers with the interests of their populations is a cornerstone of democracy. While polling can collect opinions from large publics, and small focus groups can elicit in-depth perspectives, few options exist for nuanced qualitative analysis at scale.

Talk to the City (T3C) is an open-source AI tool that enhances collective decision-making by analyzing democratic input while preserving the diversity and nuance of individual opinions. It equips decision-makers to accurately understand and act upon public concerns.

T3C stands out in its ability to distill large-scale public input while preserving the diversity of views represented. It analyzes public concerns while ensuring each respondent's specific claims are acknowledged. The novel report structure mitigates LLM inaccuracies, by connecting high-level summaries of broad themes and discourse to the details of individual opinions.

The power of T3C lies in its ability to resolve the trade-off between in-depth discussion and large-scale data collection. The tool's successful deployment in national-level consultations demonstrates capabilities that go beyond traditional polling or existing commercial solutions.

T3C is live at talktothecity.org, where instances from real-world implementations can be explored.

Project Overview

“Back in 2014, it was impossible to interview a mini public of people and aggregate their ideas while preserving the full nuance. But now, with Talk to the City’s help, that cost has been reduced to essentially zero. It’s broad-listening, and it can change the nature of this recursive public.”

— Audrey Tang, Taiwan’s 1st Digital Minister and co-author of Plurality.net

What makes Talk to the City unique is its combination of features

  • It leverages the power of frontier LLMs to analyze large datasets:

    • Processes input from unstructured text, interviews, and in-person/remote meetings, as well as structured data from digital tools (e.g., Polis, Remesh, Twitter).

    • Clusters, labels, and organizes key themes and points of difference discussed.

    • Provides an interactive interface for exploring the diversity of a population's opinions, at both group and individual scales.

    • Creates simulations of discourse with a specific respondent group.

  • It is designed to summarize human opinions, not just factual data

  • It automatically prepares summaries, visualizations and reports 

  • It is developed as a non-profit initiative 

Talk to the City’s data processing pipeline starts by processing a variety of data types, then uses LLMs to extract key arguments, and finally arranges similar arguments into clusters and subclusters. Users can navigate through a map of opinions and drill down to the subclusters they find most interesting.

Motivation

The fast pace of technological development in the AI space is both a source of concern and a source of hope for democracy. Pessimists worry our existing democratic processes and institutions are too slow, and too inefficient, to address the many crises humanity is facing. The more we lose trust in the efficacy of our democratic processes, the more we lose the broad participation and agreement on legitimacy of outcomes that make democracy functional, which in turn may further damage trust in democratic institutions. 

But there is hope that new technologies may help us escape this spiral of declining trust by allowing us to consult the public in much larger numbers, at a much faster pace, and in much more inclusive and transparent ways that capture diversity and nuance of opinion.

Making this a reality won’t be an easy task, and building sufficient public trust in AI will take time, but now that a growing number of AI pioneers have been popularizing this idea (e.g. here and here),  we believe that with the appropriate AI safety precautions, it is time to accelerate the design and development of open-source prototypes and begin testing them in the wild.

Read more in our recent paper, How can AI be used to inform policymaking?

Case Studies

National consultations in Taiwan

We collaborated with Taiwan's Ministry of Digital Affairs (moda) to use T3C for analyzing large response datasets on a variety of topics, including AI policy in the 2023 AI Assemblies, same-sex marriage, and party platforms in the 2024 election (Kuomintang, Democratic Progressive Party). T3C appealed for its preservation of the nuance in different respondents' opinions, compared to moda's existing Polis-based process. moda now hosts its own instance of T3C, and we're expanding the collaboration to include producing reports for the Ministry of Education's Let's Talk initiative. Full case study: Amplifying Voices: Talk to the City in Taiwan

Democratic inputs with Chatham House

AOI partnered with Chatham House and vTaiwan on their Recursive Public project, funded by OpenAI's Democratic Inputs grant, to use an experimental alternate T3C interface for semantic clustering of large-scale survey responses. Inspired by past vTaiwan work, the project involved 1000 participants in a new deliberation process, to identify their priorities for AI governance and how they approached collective decision-making on specific issues.


Broad Listening for the Tokyo Gubernatorial Election

Japanese researchers used T3C to poll Tokyo residents about their policy needs in advance of the 2024 Tokyo gubernatorial election, in a series of reports with 1000 respondents. The pilot introduced ‘Broad Listening’ to the 


Union Leaders in the United States

We’ve partnered with leaders of a California healthcare workers union to poll members on their priorities for a series of three negotiations with the Department of Veterans Affairs.

Further research questions

We're building this tool as a proof of concept and testing it with a variety of datasets, including data extracted from twitter, blog posts, pol.is consultations, and video interviews.

Our primary goal with these first experimental deployments has been to improve our understanding of the risks and benefits of using LLMs in the context of democratic consultations, and to identify where LLMs are helpful in creating intuitive interfaces for complex datasets. In particular, we wanted to answer the following questions: 

  • Are modern LLMs already reliable enough to prove helpful? 

  • Which interfaces best help users understand large corpora of opinion data?

  • How can we mitigate natural safety and quality concerns inherent in the use of current LLMs? 

  • How can we collaborate with and bring value to the broader researcher community?

If you’re interested in exploring similar questions, or want to explore using Talk to the City at your organization, get in touch: hello@objective.is