Enhancing Democracy with AI

A DMDU Reading List

A digital rendering of the U.S. capital

Photo by Jarmo Piironen/Getty Images

Interactions with algorithms can have a profoundly adverse and significant impact on the functioning of democratic societies. But what would it look like to design AI and the associated socio-technical systems to improve democracy?

A collection of RAND researchers, graduate students, and outside contributors collaborated to develop a reading list on the topic. This list served as the foundation for lively discussions about how AI and DMDU could be used to steer democracy toward a brighter, more sustainable future.

The collaborators broke these questions into their component parts, and examined them over a period of four weeks.

Jump To

Week 1 - What do we mean by democracy?

Our readings included three pieces suggesting how AI might enhance democracy, one by Bruce Schneier, a computer security expert who writes widely on AI and democracy; another by Helene Landemore, a political science professor known for her advocacy of direct democracy; and a third by Aviv Ovadya, an expert on AI governance. Our readings then included two commentaries on Taiwan’s experiments in digital democracy – one positive and the other negative. Finally, to engage with the question – what do we mean by democracy? – we consulted the provocative introductory chapter of the book Justice by Means of Democracy by classicist, political philosopher, and Washington Post columnist Danielle Allen.

Discussions and readings for week one were organized by Rob Lempert.

Readings

Week 2 - Expanding democracy beyond traditional representative bodies

Our topic during the second week was on new approaches that expand democracy beyond traditional representative bodies (e.g citizen assemblies, liquid democracy, deliberative polling) and how AI might facilitate such new approaches.

Our readings this week included three pieces focused on what we might call Democracy 2.0: a magazine article on citizen’s assemblies by Claudia Chwalisz, founder and CEO of the NGO DemocracyNext and previously lead author of a major 2020 OECD report on Deliberative Democracy; another magazine article on deliberative democracy by journalist James Traub, with a focus on the deliberative polling done by James Fishkin and used in Stanford’s America In One Room deliberative engagements; and a blog post on liquid/delegative democracy by Bryan Ford, a computer scientist who leads the Decentralized/Distributed Systems (DEDIS) lab at the Swiss Federal Institute of Technology in Lausanne (EPFL).

Our readings then included two pieces that discuss AI technologies in support of Democracy 2.0: a white paper from the AI Objectives Institute, a lab building AI tools for coordination and deliberation, and a chapter by Fishkin and colleagues from The Routledge Handbook of Collective Intelligence for Democracy and Governance about the self-moderating platform for online deliberation used in American in One Room. Finally, we offer links for two software platforms that implement some of these ideas to explore and discuss.

Discussions and readings for week two were organized by Leah Dion and Rob Lempert.

Readings

Tools

Week 3 - How AI might impact human agency, and how humanity might steer AI development

In this session, we explored the extent to which alternative technology development paradigms might steer AI technology in ways that might be more beneficial for democracy.

The first set of readings explores how algorithm design, development, and use might impact human agency. The first piece in this set is the introduction to a longer Pew piece that interviewed various experts on this topic. The second argues that oversight by individual humans is not very effective at providing democratic control over algorithms used by government. The third explores different models of AI development as it relates to generating solutions for climate change.

The second set of readings advocates for a greater government role in steering the directions of AI technology development. The first piece argues for a government funded and developed public AI. The second uses the example of self-driving cars to offer a framework for thinking about such a government role. The third reviews 49 recent AI policy documents and finds they have high expectations for public engagement in guiding the technology towards beneficial uses. The final reading is the Biden Administration’s AI R&D plan, which promotes a government role.

Discussions and readings for week two were organized by Swaptik Chowdhury and Rob Lempert.

Readings

Human Agency and Algorithmic Automatism

Government steering AI innovation

Week 4 - The transformative potential of AI to enhance democratic participation policymaking

Our readings during week four focused on some of the roles that AI agents might play as stand-ins for and facilitators of society, and what that might mean for citizens, policymakers, and institutions.

The group Existential Hope sponsored a World Building contest to showcase innovative and positive futures enabled by advanced AI. One of the winning entries envisions a world governed by V.O.I.C.E. (Voice for Open source Information and Community Engagement), a sophisticated AI-driven tool that collects, analyzes, and synthesizes public opinions on a grand scale. This tool champions direct democracy by engaging communities to express preferences and propose balanced solutions, capable of scaling to accommodate the input of billions globally. This allows for informed decision-making across all levels of society, from local community decisions to international treaties.

While the world is far from deploying V.O.I.C.E., many precursors already exist. There exist AI agents that can serve as proxies for individuals, responding to surveys and polls, participating in focus groups, or even voting. There also exist prototype AI agents and assistants that facilitate human-to-human interactions, including a Habermas Machine that supports deliberation.

Finally, given its general relevance to the reading group, we wanted to pass on the AI4 democracy site; it also includes the How Can AI Be Used to Inform Policymaking? reading). It may prove fruitful for future discussions and meetings of our group.

Discussions and readings for week two were organized by James Syme and Rob Lempert.

Readings

AI as Stand-ins for Society

AI as Facilitators of Human Decision Making