OpenAI invited the 10 teams working on the Democratic Inputs to AI research project to meet at the OpenAI headquarters in San Francisco. The goal? To learn more about the various projects, and meet the people behind them.
In retrospect, it was a masterclass in how to kickstart a community.
- A clear mission (democratic inputs to AI)
- Enabling incentives (the grant)
- Frequent check in’s (weekly calls)
- Ever present, friendly and energised support (thanks Teddy, Tyna and Nathalie!)
- Meeting in person at least once
- Food!
- A chance to unwind and make shared memories (thanks Karl!)
- Embracing uncertainty
- Making plans for the future
The Journey
We travelled to San Francisco with a team of 7: Pepijn & Lei (BMD) who work on the app development, Rolf (Drostan) who brings expertise on social deliberation, Rich (Sortition foundation) who is an expert in representative citizens sampling, Bram (JADS) who focusses on the Ethics, and Evelien & Jorim (Dembrane) representing the core of the consortium.
On Thursday at 3pm, we headed to the OpenAI HQ to check-in for the dinner & mixer. We walked through the sunny mission district of San Fransisco. Block after block, suddenly it looked like a nondescript concrete office building.
I thought: Is this it?
We got there a bit early. It looked calm, like a five star hotel. Friendly. And then we started to see the faces of people we had only seen online before.
I started talking to a fellow grantee. A guy from Kenya, we talked about his PhD research in language models. Nice! We dived into the deep end almost immediately. The second woman I talked to had been doing her post-doc research on DAO’s (Decentralised Autonomous Organisations) for years. You can recognise the expertise!
Before long, the influx of new faces was too fast to keep track of. It was like the calm before the storm. Being there with such a large group, it felt like we belonged here.
The next couple hours flew by, filled with interesting conversations with all the different people coming from four different continents. Professional peacekeepers, successful entrepreneurs, academics from the most prestigious universities, democratic activists, journalists from a Nobel peace prize winning group news outlet, multiple OpenAI employees including one of the founders, and many more extremely knowledgeable, interesting, and kind people were present. All were connected via at least one common interest: exploring novel democratic processes in order to align AI and ensure a better AI-included future.
Everyone was so friendly, interested, eager and open to talk. We felt quite at ease and at home very quickly. After a delicious dinner, the jet lag forced us home fairly early with full heads. Jorim, Pepijn, Lei and Bram stayed up a little later, working on the presentation for the Frontier model forum.
The next day started at 8am with a nice and calm breakfast, which was a good foundation for the upcoming packed day. From 9am to 12am there was a presentation session where all teams presented their work in 10 minutes, after which there was time for a couple questions for each presentation. Outside of all the teams, multiple interested OpenAI employees joined in the presentation, and interested employees from Google, Anthropic, and Microsoft attended online. It was enlightening to hear the other teams present their work, and these presentations added much to our understanding of how the different projects could benefit each other and be combined into bigger democratic processes. I think Jorim nailed our presentation!
After the lunch, we started diving into the technical details. Teams, slipt up and re-merged. This was the time for people interested in similar dimensions of the democratic processes to dive deeper into the respective topics. There were presentations about the process and thinking behind Pol.is, X’s Community Notes, and vTaiwan’s inclusive consultation process, and a discussion session on the foundational values and goals underlying this project.
During the lunch I had a conversation with someone who was working on an AI to talk to animals! I had a similar idea years ago but I thought it would be a decade before it might be practical, but the future is now I guess?!
A beautiful question came up during the workshops: Why are we doing this? Why are we building democratic processes. Andrew from Remesh had a wonderful answer. Paraphrasing slightly:
“I want humans to have agency in the future. How do you define that? How do you get there? Processes are needed. We have some already but we aren’t there yet. Not all humans have control right now. We would all agree that we would rather not have the problems such as homelessness or climate change, but we seem unable to fix it. We need better processes and tools!”
After hours of intricate discussions it was time to blow of some steam, which we did by some wonderful sailing in the bay. One of the groups took a self driving car to the pier: that was a strange sight! The waves made the boat ride quite exciting, Karl the Fog obscured the view of the bridge, but that almost made it more epic!
In practice, however, the in-depth discussions continued during sailing, and even continued at one of the evaluation team’s home till far in the evening simply because there was so much to talk about.
The many conversations have deepened our understanding of the subtle details of the tackled problem space, and provided overview of what is currently worked on or has already been done - in short - a lot!). We also connected with many smart, knowledgeable, and kind people that can help both our collective democratic projects. And finally, it has fueled us with plenty of inspiration to continue working on novel democratic processes in order to ensure an AI-aligned future. To misquote Abraham Lincoln: Aligned AI is artificial intelligence of the people, by the people and for the people.
Also, San Francisco is an amazing city, but that is something for another blog post. 🏝️
