CIP Q3: TIME100, Global AI Dialogues, Community Models, and more!

We live in an age of endless AI policy reports, whitepapers, agendas, and convenings. A few are even doing the hard and important work of getting at the complex relationships between frontier technology, rights, power, decentralization, agency, and democracy—we’ve been sharing some of our favorites in our weekly Collections.

That being said, we’ve been thinking deeply about a core principle of our organization: prototypes over papers, and how best to avoid the white paper trap. Our team has been heads-down building scalable platforms that enable the collective input into, and governance of, AI.

Reach out at hi@cip.org if you have thoughts, ideas for collaboration, or just want to say hello.

Global AI Dialogues

Along with Remesh and Prolific, we conducted a pilot, global-scale public input process to answer the following question: what do people around the world want from the future of artificial intelligence? 

We constructed a set of forty questions, and gathered input via the Remesh platform from 1243 participants around the world, selected by Prolific as a representative sample. These questions asked participants to reflect on their lives, their fears and hopes for the future, their culture, and their views on transformative technology. Our process was targeted to elicit insights that could inform decisions along the entirety of the AI stack, from model development and evaluation to policy and governance.

While this project was designed to produce a set of normative outcomes that AI developers can use to evaluate their impact, we also see it as a stepping stone towards more global democratic processes for AI governance. We’d love to talk with researchers, organizations, and policymakers that have ideas on how to use this data for policy and more robust sociotechnical evaluations.

In the weeks ahead we'll be sharing how we ran the experiment, snapshots of our findings, and taking action on what we’ve learned, so stay tuned!

Introducing Community Models

In our last quarterly update we shared news about our Open-Source Collective Constitutional AI work, which will allow any group of people to align an AI to their own preferences. 

Lately we've been calling this project Community Models, which reflects our focus on working with groups that have clear use cases in which they need easier, lower-cost ways to collectively align and have input into AI. 

The lived realities of communities involve a mixture of local knowledge, context, and culture. Just as there is no one-size-fits-all solution, there is no such thing as a perfectly aligned model. We need productive diversity; a healthy, pluralistic ecosystem of models that serve different needs at different scales. The world needs more tools that allow people to shape the AI they use.

We’ve been working alongside a few leading organizations from across the globe to prototype this approach. There’s an opportunity to combine innovations in deliberative technology with constitutional AI alignment to make it possible for AI to be built by communities, for communities.

We’re looking for early design and prototype partners who want to work with the CIP team to identify ways that they can use Community Models, or want to run a collective model alignment process.

Welcome to James Padolsey

CIP is growing! We’d like to introduce James Padolsey as our new founding engineer. James is a software engineer with many years’ experience at companies like Facebook, Twitter and Stripe. He’ll be leading our work on building a full-stack approach to democratic AI, and has already done incredible work on Community Models. He enjoys hacking away on experiments, exploring the potential of AI in spaces like education, disability support, and mental health. Plus, he’s a big plant enthusiast, and the CIP office has benefitted.

Time100/AI

In recognition for our work at CIP, Divya Siddarth and Saffron Huang were included on Time’s 100 AI list for 2024. We’re thankful to Billy Perrigo for capturing our mission, and want to congratulate all the fellow honorees.

Recent Publications

A Vision of Democratic AI
What would The Federalist Papers look like if they were written today? The Digitalist Papers, a project from the Stanford Digital Economy Lab, gathered the world’s leading thinkers in technology, economy, and democracy to answer this question. CIP offered “A Vision of Democratic AI,” detailing how institutions can better govern transformative technologies. The NYTimes gave us a nice mention in their feature about the collection.

AI Morality
CIP has a chapter in the upcoming book, AI Morality, edited by David Edmonds, titled “Collective Intelligence over Artificial Intelligence.”

Personhood Credentials
When AI is able to mimic both human language and voice, how do we preserve privacy and prove who is real online? Divya was a co-author on a seminal new study that explores this problem.

How large language models can reshape collective intelligence
What are the social impacts of LLMs through the lens of collective intelligence? What impacts do they have on how we manage and act on information? Divya contributed to this joint paper, led by Jason W Burton, which was published in Nature.

CIP in the world

Technodiversity at Bellagio AI
CIP’s work helped shape the agenda of the Technodiversity Convening, led by ITS-Rio, at Rockefeller Foundation’s Bellagio Center. The convening was a week-long residency to develop ideas and reframe the global narrative of technology through the lens of technodiversity.

Rise25 Mozilla
Audrey and Divya were both honored as part of Mozilla’s Rise25 list, which recognizes leaders that are building a more responsible AI ecosystem.

Bhutan
Zarinah was in Bhutan for the Bhutan Innovation Forum, a high-level gathering to chart the future of the country’s technological development and explore ideas for their Gelephu Mindfulness City. While there, she also met with the Bhutanese AI society to discuss future collaborations.

National Cooperative Business Association
Evan attended the National Cooperative Business Association’s IMPACT conference to share our newest Community Models work, and held a fireside on how democratically-aligned AI can enhance participation, transparency and decision-making within cooperatives.

Montenegro
Divya was a keynote speaker at Montenegro’s AI Strategy meeting hosted by the Central Bank of Montenegro and the World Bank, sharing examples on how countries can establish democratic processes for AI.