The Collective Intelligence Project

The Collective Intelligence Project (CIP) is a San Francisco–based nonprofit research organization founded in August 2022 by Divya Siddarth and Saffron Huang as an incubator for new governance models for transformative technology. CIP occupies the corpus's US anchor slot for deliberative-AI infrastructure — the organization that ran the OpenAI Democratic Inputs to AI program and co-led the Anthropic Collective Constitutional AI project, establishing the principal practitioner body for the strategy of routing AI development decisions through structured public deliberation rather than closed expert or corporate processes. Siddarth and Huang were both named TIME 100 AI 2024 honorees in recognition of this work.

Founding and theory of change

CIP was launched in August 2022 by Divya Siddarth and Saffron Huang. Siddarth arrived from a trajectory spanning Stanford computational decision analysis, the Ostrom Workshop, Harvard Safra Center, the Oxford Ethics in AI Institute, Microsoft's Office of the Chief Technology Officer (where she co-led the Social Technology team), and the UK AI Safety Institute, where she led the AI and Democracy program establishing public-comment infrastructure for AI risk assessment. Huang moved through Harvard, DeepMind's language-model and human-AI-interaction research teams, and Anthropic's societal impacts research, co-founding Kernel Magazine alongside her AI research work.

CIP's founding whitepaper articulates what it calls a transformative technology governance trilemma: existing governance approaches force societies to choose between progress, participation, and safety when managing AI and similarly powerful technologies, and CIP's proposition is that collective intelligence tools — deliberative processes, consensus platforms, participatory design mechanisms — can dissolve the trade-off by embedding public preferences into development decisions before the fact rather than correcting harms after. The organizational form is itself designed to demonstrate the model: CIP explicitly declines commercial and contractual funding, operating exclusively through philanthropic donations to preserve independence from the companies whose development decisions it aims to shape.

Alignment Assemblies

CIP's signature program is the Alignment Assembly — a structured, facilitated deliberation convening representative publics to surface values and preferences for AI systems, with the deliberated output fed directly into developer decision-making. The first Alignment Assemblies ran beginning in early 2023 in partnership with OpenAI, as CIP's delivery mechanism for OpenAI's Democratic Inputs to AI program — OpenAI's grant initiative offering ten $100,000 awards for experiments in democratic processes for determining what rules AI systems should follow. CIP's assemblies brought random and regionally representative samples of the public through structured deliberation sessions using the Polis platform — the consensus-mapping tool developed in the context of Taiwan's g0v civic-tech ecosystem and vTaiwan platform — to identify where broad consensus and genuine division exist.

The most significant Alignment Assembly output to date is the Collective Constitutional AI collaboration with Anthropic, in which approximately 1,000 representative Americans deliberated to produce a 275-statement constitution for an AI language model — a set of behavioral principles that Anthropic then used to train a model, producing the first publicly documented instance of members of the public collectively directing the behavior of a large language model through an online deliberation process. The assembly's constitution differed substantively from Anthropic's in-house 58-principle constitution, with the public version placing notably greater emphasis on avoiding paternalism, protecting privacy, and providing balanced perspectives on contested topics. The result, published as a joint CIP–Anthropic research paper in 2024, is the corpus's primary documented case of a deliberative democracy process reaching all the way into AI training decisions rather than stopping at policy consultation.

Global Dialogues and Community Models

CIP's 2024 programming extended the Alignment Assembly model toward two further tracks. The Global Dialogues initiative runs monthly public-input surveys across a growing international panel; the 2024 pilot engaged 1,200 participants across 68 countries in 8 languages, producing cross-national data on how diverse publics reason about AI values, governance, and risk. Early findings show strong global preference for international AI coordination and consistent concern about single-actor dominance in AI development decisions — the first systematic multinational public-opinion dataset specifically on AI governance questions. The Community Models platform inverts the earlier assembly architecture: rather than sampling from the general public to advise corporations, it enables specific communities — Grandmothers' Collective, ITS Rio, Equiano Institute (Africa), Civis (India) among the 2024 cohort — to build AI systems shaped by their own deliberated values, directly addressing the historic exclusion of non-Western and non-technical communities from AI development decisions.

Team and senior fellows

CIP's core team as of 2025 includes Zarinah Agnew as Research Director (a neuroscientist with fifteen years of academic background and former president of Irrational Labs and the Social Science Observatory), Evan Hadfield as Projects Lead, Faisal M. Lalani as Head of Global Partnerships, and Joal Stein leading Communications and Operations. The organization's most prominent senior fellow is Audrey Tang — Taiwan's inaugural Minister of Digital Affairs (2022–2024) and the civic hacker who, at g0v, built the vTaiwan online deliberation platform that established the methodological lineage CIP's Alignment Assemblies draw from. Tang's hire connects CIP to the deepest practitioner tradition in digital democratic infrastructure and introduces into the organization's work the East Asian civic-tech genealogy — Polis, vTaiwan, broad listening — that shaped the deliberative AI tools CIP now deploys at scale. CIP has also hired Audrey Tang's concept of "broad listening" as a conceptual anchor for its work, which situates the organization inside a global intellectual lineage running from Elinor Ostrom's common-pool resource governance through the e-democracy tools of the early 2000s to the AI alignment work of the current decade.

Divya Siddarth's concurrent roles as Research Director of Metagov — the research consortium working on governance infrastructure for digital platforms and online communities — and as Researcher in Residence at RadicalXChange Foundation connect CIP's deliberative AI work to the broader field of pluralist, participatory institutional design, of which AI governance is CIP's specific focus.

Governance, funders, and positioning

CIP achieved 501(c)(3) nonprofit status in 2024 and opened its San Francisco office the same year, having operated previously in distributed form. Its funders include the Ford Foundation, Omidyar Network, Survival and Flourishing Fund, Robert Wood Johnson Foundation, Google.org, Future of Life Institute, One Project, and Amaranth Foundation. The explicit refusal of commercial relationships with the AI companies whose processes CIP seeks to shape is the organization's structural claim to methodological independence — the same argument that makes it a credible counterpart to Anthropic and OpenAI in published joint research.

CIP's government advisory work expanded significantly in 2024, with active engagements advising AI governance strategy in Bhutan and Montenegro and contributing to Taiwan's Ministry of Digital Affairs' Alignment Assemblies project. The British Frontier AI Taskforce also listed CIP as a collaborator.

Posture in the movement

Within the corpus's frame, CIP is the primary US anchor for the strategy of embedding public deliberation inside AI development processes — a practitioner organization that has moved the strategy from theoretical advocacy to concrete deliverables (a trained model on a publicly-written constitution; a seventy-country multinational opinion dataset on AI governance). Its structural position is unusual: it operates simultaneously inside AI company research programs (the Anthropic and OpenAI collaborations) and as an independent nonprofit whose funders are explicitly not those companies, which gives it a different leverage point than either pure advocacy organisations or pure researcher-contractors. CIP's distinctive thesis — that the absence of democratic infrastructure for AI development is itself a governance failure as significant as any individual harm — is the thesis the Alignment Assembly form is designed to operationalize, and the Collective Constitutional AI publication is the corpus's first documented empirical test of that thesis at scale.