In a groundbreaking initiative to involve the global community in shaping the future of artificial intelligence (AI), OpenAI funded 10 diverse teams from around the world through its Democratic Inputs to AI grant program. The program, announced in May, awarded $100,000 to each of the 10 teams out of nearly 1,000 applicants. These teams were tasked with designing, building, and testing ideas using democratic methods to determine the rules governing AI systems.

The initiative reflects OpenAI’s commitment to aligning AI models with human values, recognizing the need for public input in shaping the behavior of increasingly advanced and widespread AI technologies. The teams faced challenges such as recruiting participants across the digital divide, ensuring transparency in their processes, and creating outputs that represented a diverse range of viewpoints.

The selected teams, comprised of experts from various fields including law, journalism, peace-building, machine learning, and social science research, have now successfully completed the grant program. OpenAI is set to build on this momentum, collaborating with the grant teams to design an end-to-end process that collects inputs from external stakeholders and utilizes them to train and shape AI models.

Key Innovations Unveiled by Grant Recipients:

1. Case Law for AI Policy:
   • Team: Quan Ze (Jim) Chen, Kevin Feng, Inyoung Cheong, Amy X. Zhang, King Xia
   • Methodology: Creating a robust case repository through democratic engagement to make case-law-inspired judgments.
2. Collective Dialogues for Democratic Policy Development:
   • Team: Andrew Konya, Lisa Schirch, Colin Irwin
   • Methodology: Developing policies reflecting informed public will through collective dialogues to find areas of consensus.
3. Deliberation at Scale: Socially Democratic Inputs to AI:
   • Team: Jorim Theuns, Evelien Nieuwenburg, Pepijn Verburg, Lei Nelissen, Brett Hennig
   • Methodology: Enabling democratic deliberation in small group conversations conducted via AI-facilitated video calls.
4. Democratic Fine-Tuning:
   • Team: Joe Edelman, Oliver Klingefjord, Ivan Vendrov
   • Methodology: Eliciting values from participants through chat dialogue to create a moral graph for fine-tuning AI models.
5. Energize AI: Aligned – a Platform for Alignment:
   • Team: Ethan Shaotran, Ido Pesok, Sam Jones
   • Methodology: Developing guidelines for aligning AI models with live, large-scale participation and a ‘community notes’ algorithm.
6. Generative Social Choice:
   • Team: Sara Fish, Paul Gölz, Ariel Procaccia
   • Methodology: Distilling free-text opinions into a concise slate using mathematical arguments from social choice theory.
7. Inclusive.AI: Engaging Underserved Populations in Democratic Decision-Making on AI:
   • Team: Yang Wang, Yun Huang, Tanusree Sharma, Dawn Song
   • Methodology: Facilitating decision-making processes related to AI using a platform with decentralized governance mechanisms.
8. Making AI Transparent and Accountable by Rappler:
   • Team: Gemma B. Mendoza, Gilian Uy, Don Kevin Hapal, Ogoy San Juan, Maria Ressa
   • Methodology: Enabling discussion and understanding of participants’ views on complex, polarizing topics via linked offline and online processes.
9. Ubuntu-AI: A Platform for Equitable and Inclusive Model Training:
   • Team: Ron Eglash, Joshua Mounsey, Micheal Nayebare
   • Methodology: Returning value to creators while facilitating model training and ensuring more inclusive knowledge of African creative work.
10. vTaiwan and Chatham House: Bridging the Recursive Public:
    • Team: Alex Krasodomski-Jones, Carl Miller, Flynn Devine, Jia-Wei (Peter) Cui, Shu Yang Lin
    • Methodology: Using an adapted vTaiwan methodology to create a recursive, connected participatory process for AI.

Learnings from the Grant Program:

1. Changing Public Opinions:
   • Public views on AI can change frequently, emphasizing the need for thorough and recurrent input-collecting processes.
2. Digital and Cultural Divides:
   • Bridging across the digital divide remains challenging, affecting the results and making outreach and tooling improvements necessary.
3. Finding Agreement in Polarized Groups:
   • Finding compromise within strongly opinionated small groups can be challenging, highlighting the importance of balanced representation.
4. Reaching Consensus vs. Representing Diversity:
   • Balancing consensus and diverse representation is crucial when aiming for a single outcome to represent a group.
5. Hopes and Anxieties about AI Governance:
   • Participants expressed both hope and caution about the role of AI in democratic processes, emphasizing the need for transparency and accountability.

The grant program illuminated crucial learnings, emphasizing the dynamic nature of public opinion, challenges in bridging the digital divide, difficulties in finding consensus within polarized groups, and the delicate balance between reaching consensus and representing diversity.

Participants expressed both hopes and anxieties about AI governance. Transparency in AI application in democratic processes was a common concern, and participants became more hopeful about the public’s role in guiding AI after deliberation sessions.

OpenAI’s Implementation Plans:

OpenAI’s commitment continues with the formation of a “Collective Alignment” team, comprising researchers and engineers. The team aims to:

• Implement a system for collecting and encoding public input on model behavior.
• Collaborate with external advisors and grant teams, incorporating grant prototypes into steering AI models.
• Recruit diverse research engineers to contribute to this pioneering work.

The journey toward democratizing AI governance is unfolding, fueled by innovation, public engagement, and a collective commitment to shaping a future where AI aligns with human values. OpenAI invites individuals passionate about this mission to join the movement.

Check out the official blog notification here.