What is Augmented Deliberation?

Thanks to television and the internet, it has become easy for one person's words to reach millions. But the reverse — one person listening to millions of voices and understanding what they say — is still very hard. Augmented Deliberation is the effort to make careful discussion among large numbers of people possible with the help of technology. "Deliberation" here means talking things through carefully: putting forward views and hearing each other's reasons.

In short

Augmented Deliberation means using digital technology to extend people's ability to talk and think together at scale.

The book ⿻ Plurality frames the problem like this: "while methods of broadcast (allowing many to hear a single statement) have dramatically improved, broad listening (allowing one person to thoughtfully digest a range of perspectives) remains extremely costly and time consuming" [1][2].

For example

Community Notes on X (formerly Twitter). Participants write "notes" that add context to posts that may be misleading, and rate each other's notes. Only notes rated helpful by people who usually disagree with each other are shown. The book puts this system, to which "hundreds of millions of people are currently exposed each week", at the very start of chapter 5-4 as an example of Augmented Deliberation [3].

vTaiwan and Polis in Taiwan. With the tool Polis, participants do not reply to other people's statements; they only click "agree" or "disagree". From the pattern of clicks, opinion groups emerge, along with the statements supported across groups. In Taiwan it was used in discussions such as how to regulate ride-hailing services [4].

The 2024 Tokyo governor's election. The candidate Takahiro Anno published his manifesto and gathered opinions through X, online forms, an AI that answered phone calls, and GitHub (a site for editing programs together). The voices gathered were made visible with a tool called "Talk to the City", and policies were revised on GitHub. The book writes that the almost unknown candidate won more than 150,000 votes. Afterwards, Talk to the City was also used for a call for public input on Tokyo's long-term plan toward 2050, and "as of December 13, over 10,000 opinions have been collected" [5].

How it works

Depth and breadth are hard to combine

Face-to-face conversation is rich, but reaches only a few people. Writing travels far, but the exchange is thin. The book calls this "the trade-off between diversity and bandwidth" (a trade-off: gaining one means giving up the other) [1].

The Japanese book Broad Listening explains this with a city of one million people. Have in-depth one-hour conversations with 10 residents and you understand each of them deeply, but that does not tell you what the whole city thinks. Ask all one million "Do we need a park? Yes/No" and you get the overall picture, but not the reasons [6].

The political scientist Hélène Landemore also treats the difficulty of combining high-quality discussion with mass participation as an unsolved problem in democratic theory. Landemore says the quality of discussion drops once it goes beyond hundreds or thousands of people, and asks how far AI can close that gap [7].

Discussion tools we have today

The book describes approaches such as recording and summarizing small-group conversations and sharing them widely. It says that facilitation — the skill of making a gathering's conversation meaningful — "is a science unto itself" [8].

Social media, meanwhile, widened who can take part, but also brought problems: "the lack of context", the spread of misinformation and disinformation, "the dominance of well-resourced entities", and "echo chambers" in which people see only views like their own [8].

Conversation tomorrow: three directions

The book sorts future discussion tools into three directions [4]:

1. Systems that gather many people's responses and bridge between them. Tools like Polis. The AI company Anthropic worked with a research group to steer its AI's behavior with a "constitution" gathered from citizens through Polis. OpenAI also gave grants for experiments bringing citizens' voices into AI.
2. Systems that center face-to-face conversation and network what is learned there.
3. Systems that organize existing debates and articles.

Listening to many with AI

Talk to the City, built by the AI Objectives Institute, shows "how GFMs can be used to replace a list of statements characterizing a group's views with an interactive agent one can talk to" [9]. (GFMs, generative foundation models, are the technology behind today's AI chatbots.) The book writes that participants will be able to go beyond short statements and up-and-down votes and express themselves fully, while AI condenses the conversation and helps "look for areas of rough consensus not simply based on common votes but on a natural language understanding" [9].

According to the book, in a large-scale experiment, AI gave each participant in a discussion "real-time, evidence-based suggestions aimed at refining the quality of political discourse", and "the results indicated a noticeable improvement in the overall quality of conversations" [5]. A 2024 paper by Landemore and others also sets out the hopes that AI could step in on rude remarks and suggest rephrasing, summarize opinions and find common ground [10].

Not only consensus but new conflicts

The book also writes that bridging systems are not only for building consensus. They can find opinion groups that are "not a deterministic function of historical assumptions or identities", and create "a new conflict that does not reinforce existing divisions" [11]. It also suggests new forms of "representation", such as choosing "the maximally diverse committees for a deliberation", and giving a voice to natural features like rivers and forests, or to future generations [11].

What it feels like in practice

In a 2025 conversation, Audrey Tang (Taiwan's former Minister of Digital Affairs) said, in substance, that people may not care about the fine details of a solution, but they are experts on their own feelings. When Taiwan discussed how to counter scam advertising that uses AI-made fake videos (deepfakes), people put forward ideas in 45 online rooms of 10 people each (the Taiwanese government's official page says 447 people in 44 groups), and 85% agreed with the final package. The remaining 15% were asked whether they could live with it even if they were not fully convinced, and almost all said they could. Tang said this check matters more than the 85% agreement [12].

What is debated

• It is hard to see inside. The book writes that ways of speeding up discussion with AI have a limit: it is hard to explain how opinions are turned into conclusions. So, it says, "it might be very hard to achieve common understanding and legitimacy on how this transformation takes place" in the way that voting and markets often achieve [13].
• Can humans check the AI? The book raises the question of whether humans can really audit more capable AI, and whether AI, following instructions blindly, might "censor some perspectives" [13]. It says AI "must offer a diverse array of reasonable responses" and be "accurately calibrated to the nuances of specific populations" [13].
• Idealizing the "common will". The book notes that "deliberation is sometimes idealized as helping overcome divisions and reach a true 'common will'", and writes that equal attention must go to "the regeneration of diversity and productive conflict" [13].
• The dangers of AI-run discussion. The paper by Landemore and others also lists risks: political bias in AI, automated persuasion, and outputs that exaggerate stereotypes and so deepen conflict [10].

Where it fits in Plurality

Part 5 of the book lists seven types of collaboration between different people, and Augmented Deliberation is the fourth [2]. It comes after the more intimate types, and before ⿻ Voting and Social Markets.

The book writes that words "are as dust compared to the richness of human sensory experience", so that "however far deliberation advances, it cannot substitute for the richer forms of collaboration" [13]. And because "talk takes time", the book says the systems of administration, voting and markets in the following chapters are also needed [13].

The book says that in making the book itself, points of disagreement were settled with the tools of this chapter [14]. For how the book was run, see also the article on the Plural Management Protocol.

In his review, Vitalik Buterin, the founder of Ethereum, wrote from his own experience that whether an organization's governance works is decided about 20% by its formal decision rules and 80% by how discussion happens beforehand. In Buterin's view, this is the context in which to understand why Weyl and Tang devote so many pages to technology for large-scale conversation [15].

Further reading

1. Plurality, chapter 5-4 "Augmented Deliberation" (English, book, with extensive footnotes) — https://www.plurality.net/v/chapters/5-4/eng/
2. Broad Listening, chapter 1 "What Is Broad Listening?" (Japanese and English, book, free to read) — https://broadlisteningbook.com/
3. "Using AI to Support Public Deliberation: A Conversation with Audrey Tang", with Danielle Allen and others (English, conversation) — https://rebootdemocracy.ai/blog/deliberating-with-the-public
4. Summerfield and others, "How will advanced AI systems impact democracy?" (English, paper) — https://arxiv.org/abs/2409.06729
5. Aviv Ovadya, "'Generative CI' through Collective Response Systems" (English, paper) — https://arxiv.org/abs/2302.00672
6. Plurality, chapter 5-4 "拡張熟議" (in Japanese, book) — https://www.plurality.net/v/chapters/5-4/jpn/

Sources

1. Plurality, 5-4 "Augmented Deliberation", first half of "Conversation today" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
2. Plurality, chapter summaries (Chapter 5-4) — https://github.com/pluralitybook/plurality/blob/86158859464aee75633acd854c656928121a7fd8/contents/english/8-blurb.md ［CC0］
3. Plurality, 5-4, opening — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
4. Plurality, 5-4, "Conversation tomorrow" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
5. Plurality, 5-4, "Frontiers of augmented deliberation", the paragraph on Takahiro Anno and what follows — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
6. Broad Listening, chapter 1 "What Is Broad Listening?" (in Japanese; English translation in the same repository) — https://github.com/digitaldemocracy2030/broad-listening-book/blob/f69c60c838a91b91668abf109d7093b240610139/en/01_what_is_broad_listening.md ［CC BY-NC 4.0］
7. Hélène Landemore, "Can AI Bring Deliberation to the Masses?" (Stanford HAI, 2022) — https://hai.stanford.edu/events/helene-landemore-can-ai-bring-deliberation-masses
8. Plurality, 5-4, second half of "Conversation today" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
9. Plurality, 5-4, opening of "Frontiers of augmented deliberation" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
10. Summerfield and others, "How will advanced AI systems impact democracy?" (2024) — https://arxiv.org/abs/2409.06729 ［CC BY 4.0］
11. Plurality, 5-4, second half of "Frontiers of augmented deliberation" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
12. "Using AI to Support Public Deliberation: A Conversation with Audrey Tang" (Reboot Democracy, 2025-12-19, automatic transcript on the host site) — https://rebootdemocracy.ai/blog/deliberating-with-the-public
13. Plurality, 5-4, "Limits of augmented deliberation" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
14. Plurality, 5-0 "Collaborative Technology and Democracy", opening — https://www.plurality.net/v/chapters/5-0/eng/ ［CC0］
15. Vitalik Buterin, "Plurality philosophy in an incredibly oversized nutshell" (2024-08-21) — https://vitalik.eth.limo/general/2024/08/21/plurality.html