What is the Collective Intelligence Project (CIP)?

For now, how AI should behave is decided mainly by the companies that build it. Yet it is each of us who is affected by AI. Could the direction of AI be decided through discussions among ordinary people? The Collective Intelligence Project (CIP) is an organization trying to answer this question with working mechanisms.

In short

The Collective Intelligence Project (CIP; the Japanese edition of Plurality keeps the English name [1]) is a nonprofit. It researches and puts into practice mechanisms that let how to build and use technologies that transform society, such as AI, be decided through discussions among ordinary people, not only by a few companies or experts [2][3].

Divya Siddarth and Saffron Huang started it at the end of 2022 [2][3].

For example

Alignment Assemblies (2023 onward). An effort to have ordinary people gather and discuss how AI should behave, and to feed the results into the decisions of AI companies and governments. As of November 2023, CIP was running assemblies with or cooperating with the UK government's AI taskforce, OpenAI, Anthropic, Taiwan's Ministry of Digital Affairs and others [4]. Taiwan's Ministry of Digital Affairs became an official partner of CIP in May 2023 [5]. This is covered in more detail in the article on Alignment Assemblies.

Collective Constitutional AI (2023, with Anthropic). About 1,000 people chosen to represent US adults were asked, using the opinion-organizing tool Polis, how they wanted an AI chatbot to behave. An experimental AI was then trained on the "public constitution" (a set of rules for AI behavior) drawn up from their answers [6].

CIP reports that the AI trained on this public constitution was less biased on many kinds of bias, and that its abilities, such as math and language understanding, were no different from the standard AI [6].

Global Dialogues and Weval (2025 onward). Every two months, CIP asks people chosen to represent the world about life with AI. For 2025 it compiled seven rounds covering more than 6,000 people in 70 countries. Results included that 58% of people trust AI chatbots more than their elected representatives, and that 67% use AI for emotional support at least once a month [7].

CIP has also built "Weval", a place where people create and share their own methods for evaluating AI by region or field [8].

In more detail

The name and the start

In a TED talk, Siddarth described how it began. The gist: together with co-founder Huang, Siddarth left jobs at Google DeepMind and Microsoft to launch a nonprofit to build new models for democratically governing transformative technology, and chose the name to express continually growing the collective intelligence needed for everyone to flourish [3]. As of 2025, Siddarth serves as executive director [2].

A "fourth way" that meets all three at once

CIP's founding document (its white paper) sorts out the views on how to handle technologies that transform society. Every camp, it says, tends to sacrifice one of three things: "progress", "participation" and "safety" [9].

• Sacrificing safety: capitalist acceleration, which just pushes ahead as fast as possible.
• Sacrificing participation: technocratic rule, in which a few experts or powerful people decide.
• Sacrificing progress: stagnation, in which everyone stops together.

CIP set out to research and build collective intelligence mechanisms that meet all three, as a "fourth way" that is none of these [9]. Collective intelligence here means decision-making technologies, processes and institutions that expand a group's capacity to set shared goals and cooperate toward them. The white paper counts markets and national democracy as kinds of it too [9].

Two pillars

The white paper set two areas to work on first [9].

1. Eliciting values — surfacing and combining the conflicting values of overlapping groups.
2. Rebuilding the organizations of technology — building organizations that can actually act on the values gathered, such as prototypes of companies with capped profits.

Its initial plans included trying to decide the uses of AI through citizens' assemblies, as well as supporting closely related efforts such as Polis, Talk to the City (a tool that uses AI to organize many people's opinions) and RadicalxChange [9].

Starting from groups, not individuals

In a 2025 podcast, CIP named as one of its principles moving interaction with technology from the individual to the collective. It also described trials of "community models", in which a community discusses and adjusts an AI to fit its own values [2].

Research on economics

A paper by Siddarth, Matthew Prewitt and E. Glen Weyl is published on CIP's website. It argues that things that become cheaper and more effective when provided to many people together rather than one at a time ("supermodular" goods) cannot be handled by the two-way split into public and private, and need mechanisms for providing and governing them collectively [10]. This is covered in more detail in the article on supermodularity.

What is debated

• A gap between the book and the primary sources. Chapter 5-4 of Plurality writes that, "working closely with the ⿻ NGO the Collective Intelligence Project (CIP), Anthropic's recently released Claude3 model ... sourced the constitution used to steer model behavior using Polis" [11]. CIP's own primary sources, however, say that the model trained on the public constitution was an experimental AI for comparison, and do not say it was used for Claude 3 [6][12]. Audrey Tang said in an April 2024 interview that Claude 3 Opus had been tuned with a constitution made by many people [13]. Whether it was used for Claude 3 needs to be checked against Anthropic's own announcements.

Where it fits in Plurality

Chapter 5-4 of Plurality, "Augmented Deliberation", calls CIP "the ⿻ NGO" [11]. Besides the cooperation with Anthropic above, it writes that OpenAI "also worked closely with CIP to run a grant program on 'democratic inputs to AI'", and that on this basis OpenAI is now forming a "Collective Alignment Team" [11].

Chapter 7-0, "Policy", lists "alignment assemblies", with a link to CIP's page, as one of the means for the public itself to take part in decisions about technology [14]. Chapters 3-2, 4-3 and 5-0 cite the paper on CIP's website in explaining supermodularity [15][16].

CIP's blog also carried "We need network societies, not network states" by Siddarth, Weyl and Anne-Marie Slaughter [17]. It is a criticism of the network state.

Tang describes herself as a Senior Research Fellow at CIP [18]. The glossary of Tang and Caroline Green's 6-Pack of Care lists the Alignment Assembly as a tool whose term and format "the Collective Intelligence Project developed", and presents CIP's Weval as "the working example" of shared, community-authored evaluations [19][20]. This thread is also covered in the article on civic AI.

Further reading

1. CIP white paper (English, document) — https://www.cip.org/whitepaper
2. Divya Siddarth, "How AI and democracy can fix each other" (English, talk, TED) — https://www.ted.com/talks/divya_siddarth_how_ai_and_democracy_can_fix_each_other
3. Plurality, chapter 5-4 "Augmented Deliberation" (English, book) — https://www.plurality.net/v/chapters/5-4/eng/
4. CIP, "CIP and Anthropic launch Collective Constitutional AI" (English, article) — https://www.cip.org/blog/ccai
5. Collective Constitutional AI (FAccT 2024, English, paper) — https://arxiv.org/abs/2406.07814
6. CIP, "How the World Lives with AI" (English, report) — https://blog.cip.org/p/2025gdindex

Sources

1. PLURALITY, Japanese edition, 5-4 "拡張熟議" (Augmented Deliberation) (in Japanese) — https://www.plurality.net/v/chapters/5-4/jpn/ ［CC0］
2. Computer Says Maybe, "The Collective Intelligence Project" (2025-05-30, podcast) — https://share.transistor.fm/s/bf0d5209
3. Divya Siddarth, "How AI and democracy can fix each other" (TED, 2023) — https://www.ted.com/talks/divya_siddarth_how_ai_and_democracy_can_fix_each_other ［CC BY-NC-ND 4.0］
4. CIP, "Alignment Assemblies: Nine Months In" (2023-11-16) — https://www.cip.org/blog/alignment-assemblies-nine-months-in
5. Ministry of Digital Affairs, "對齊大會" (Alignment Assemblies) (in Chinese) — https://moda.gov.tw/major-policies/alignment-assemblies/1453
6. CIP, "CIP and Anthropic launch Collective Constitutional AI" (2023-10-17) — https://www.cip.org/blog/ccai
7. CIP, "How the World Lives with AI" (Global Dialogues 2025, 2026-01-21) — https://blog.cip.org/p/2025gdindex
8. CIP, "Notes on building collective intelligence into evals" (2025-09-29) — https://blog.cip.org/p/notes-on-building-collective-intelligence
9. CIP white paper (2023) — https://www.cip.org/whitepaper
10. Divya Siddarth, Matthew Prewitt and Glen Weyl, "Beyond Public and Private: Collective Provision Under Conditions of Supermodularity" — https://cip.org/supermodular
11. Plurality, 5-4 "Augmented Deliberation" — https://www.plurality.net/v/chapters/5-4/eng/ ［CC0］
12. Collective Constitutional AI (FAccT 2024 paper) — https://arxiv.org/abs/2406.07814 ［CC BY 4.0］
13. Business Weekly April column, interview with Minister Audrey Tang (2024-04-09, in Chinese) — https://sayit.archive.tw/2024-04-09-%E5%95%86%E5%91%A8%E5%9B%9B%E6%9C%88%E5%B0%88%E6%AC%84%E8%A8%AA%E8%AB%87%E5%94%90%E9%B3%B3%E9%83%A8%E9%95%B7%E9%80%90%E5%AD%97%E7%A8%BF ［CC0］
14. Plurality, 7-0 "Policy" — https://www.plurality.net/v/chapters/7-0/eng/ ［CC0］
15. Plurality, 5-0 "Collaborative Technology and Democracy" — https://www.plurality.net/v/chapters/5-0/eng/ ［CC0］
16. Plurality, 4-3 "Commerce and Trust" — https://www.plurality.net/v/chapters/4-3/eng/ ［CC0］
17. Divya Siddarth, Glen Weyl and Anne-Marie Slaughter, "We need network societies, not network states" (2024-05-20) — https://www.cip.org/blog/network-societies
18. audreyt.org — https://audreyt.org/
19. civic.ai, Glossary — https://civic.ai/glossary/ ［CC0］
20. civic.ai, Measures — https://civic.ai/measures/ ［CC0］