In 2023 OpenAI announced grant funding for experiments in setting up democratic process for deciding what rules AI systems should follow. I teamed up with Chris Knouss (CPO at JP Morgan, former Apple) and a diverse team of advisors to design a process for gathering input into AI development from affinity groups worldwide.
.png)
We designed a process where people join communities, engage in debate, and have their values codified in a charter that can be used to steer and evaluate AI systems. Charters can be used by AI engineers to tune LLMs and by community leaders to tell when AI aligns with their values, creating virtuous cycles of accountability.
OpenAI did not award us the grant so we took matters into our own hands. I noticed that everywhere I went, people from all backgrounds had something to say about AI. So I founded a membership community called humanfriendly.ai, a human-friendly community for the age of AI, and tested the prototype with this community.
.png)
.png)
The results are fascinating. We used community discussions, LLMs, and the open source machine learning tool pol.is to reveal what our community believes in and where we have divergent opinions. We're now in talks with some potential partners to fund a 500-person pilot and explore how the process works at greater scale.
"Thank you so much for taking my crazy, dense, technical vision and turning it into something that makes sense. You have a natural gift and you make me appreciate that there are good people in this world."
Chris Knouss, Chief Product & Data Officer @ JP Morgan | former Apple engineer
