My Work in AI Has Always Been About Democracy
People sometimes describe me as someone who "worked in AI." That is true, but people are also often misled by this. I’ve been accused of working for Meta or Big Tech, and that’s not true. My work, however, has been noticed by the tech giants because it’s rare and rigorous. For almost decade my work has been about one question: how does a democracy make good decisions when the issues are too big, too technical, and too contested for any one person to hold in their head? How can we seek truth in earnest at the scale of the internet?
AI turned out to be both a tool we could use to answer that question while also being one of the hardest questions a democracy now has to answer. So my path ran through the biggest AI labs in the world, but it never started or ended there. As you will see, Big Tech has taken notice, but my decade long-career in charity has always put me squarely on Team Humanity.
The Society Library: a library of society's reasoning
After I spent time in humanitarian and environmental work, I founded the Society Library as a tech nonprofit to do something libraries have never quite done: catalog arguments. On any contested public question, we collect the positions, the reasons behind them, the claims those reasons depend on, and the evidence for and against each claim, then organize all of it into a structured map with every node linked to its source. We build large indexes of critical information about policy issues, so we can understand the plurality of people’s points of views, the evidence for and against them, and we would shape that data to figure out what policies can be both democratic and truth-seeking.
Our goal was to modernize democratic decision-making, because representation has a hidden weakness. Citizens can speak, vote, and petition, but the actual weighing of arguments happens inside a representative's head, or behind closed doors. We call it the black-box problem. Our maps and decision-making models pull that weighing into the open, so a council member, a voter, or a journalist can see which arguments are accepted, which are rejected, and on what evidence.
In the middle of all of this, I co-launched Democracy's Library at the Internet Archive to open up U.S. government records. Same idea: people cannot govern themselves on information they cannot find. AI simply made it possible to build this kind of infrastructure at the scale democracy needs, which is why we need more data to help us reason well. This is why the Future of Life Foundation named me an AI for Human Reasoning fellow.
Three labs, one subject
So what was my relationship with the big labs?
OpenAI. In 2023, OpenAI launched Democratic Inputs to AI on a premise I share: no single company should decide how AI behaves. The Society Library responded by mapping OpenAI's own policy questions. When OpenAI later invited me to present internally, that is what I brought: a way to map the full deliberation around a policy choice, so that governing a system like ChatGPT can happen in the open, on the record, with the public's reasoning visible, rather than through a one-line poll.
xAI. xAI describes its mission as building a maximally truth-seeking AI. When I was invited to present there, my subject was the epistemology behind that slogan: what are the practical methods that AI could execute in order to seek veritable information? The Society Library's answer is procedural: enumerate the positions instead of picking one early, trace every claim to a quoted source, separate what is asserted from what is supported, and keep all views in the map, clearly labeled. A citizenry cannot deliberate if its information systems quietly resolve disputes on its behalf.
Anthropic. In June 2023, researchers from Anthropic and the Computational Democracy Project published a paper on using language models to run large-scale public deliberations without replacing human voice. One risk they flag is framing: the seed statements that open a deliberation can steer it. Their proposed fix, in Section 3.4.1, is to draw seed content from existing archives of public debate, and the archive they name is the Society Library, as a way to give the public more power over the terms of its own deliberations. I was not an author, and the paper endorses no one. It simply treats our maps as the public resource my team and I built them to be.
What ties it together
Three labs, three conversations, one subject: how a diverse public governs a technology it did not build, what a truth-seeking system owes the people who rely on it, and how to keep deliberation in the public's hands as AI enters our economic marketplace and marketplace of ideas. The years I spent pouring into this work resulted in quality contributions that large labs recognized. So no, I never worked for any labs. I have always worked to benefit truth-seeking and democracy.
Naturally, this would mean seeking truth on one of the biggest challenges our democracy will face: how to govern AI. So our work turned to mapping the debate about AI itself, so that anyone can see the main points of view and the evidence behind each. This is what allowed me to befriend extremes on all sides of the debate. Me being able to meet people at the level of their deepest concerns and show I understand allows me to bridge ideological camps, ease tensions, and strive for optimal outcomes amidst ideological battles. My legislative approach to AI is based on this deep, balanced understanding of the plurality of views - technical, social, economic, geopolitical, etc. To me, this is no simple debate, but my expertise is literally in reasoning through complexity at the scale of a civilization, which is perhaps as rare a quality in Congress as is my devotion to democracy, skill in truth-seeking, and my deep understanding of the debate about AI itself - though I hope not.