Partnership on AI: Shaping Responsible AI Practices in a Transformative Year

2023 marked a year of seismic shifts in the artificial intelligence landscape. The Partnership on AI (PAI), a nonprofit consortium of academic, civil society, industry, and media organizations, played a crucial role in addressing the challenges and opportunities presented by these rapid advancements. As generative AI technologies made unprecedented leaps, public awareness grew, and policymakers increased their focus on AI, PAI's mission to bring diverse voices together became more important than ever. Rebecca Finlay, CEO of PAI, emphasized the organization's ability to meet this moment of profound change. PAI responded to shifting contexts with collectively defined guidelines for responsible practices, engaged policymakers to turn ideas into lasting change, and maintained a people-centered approach throughout. This work built upon years of intentional foresight, thoughtful collaboration, and sustained effort by PAI staff, board members, and partners representing 119 organizations across 17 countries.

In February 2023, PAI launched the Responsible Practices for Synthetic Media: A Framework for Collective Action. This timely initiative addressed the rapid proliferation of AI-generated media, providing ethical guidelines for development, creation, and sharing. The Framework, supported by 18 partner organizations, aims to improve consent, disclosure, and transparency in synthetic media creation and distribution. Key achievements: - 50+ organizations involved in framework development - 18 organizations signed on to implement the Framework - 10 implementation case studies developed The Framework's principles are being put into practice by major tech companies and media organizations, potentially impacting billions of users. PAI plans to publish in-depth case studies from Framework supporters in 2024, using these learnings to inform public policy.

PAI launched the Guidance for Safe Foundation Model Deployment, incorporating expertise from over 50 experts and model providers. This guidance addresses the critical need for responsible development and deployment of foundational AI models, focusing on promoting safety for society. Key achievements: - 50+ organizations involved in development - 10+ endorsements of the guidance - 3 citations by PAI Partners in their work The guidance has been instrumental in informing discussions on AI safety among global policymakers. It laid the groundwork for PAI's first-ever AI Policy Forum and contributed to time-sensitive policy conversations, such as the UK AI Safety Summit. PAI plans to refine recommendations for open access models and publish an updated version of the Guidance in 2024.

PAI published Guidelines for AI and Shared Prosperity, focusing on ensuring that AI deployment benefits all sectors of society. These guidelines provide actionable ways for AI-creating and AI-using organizations, labor organizers, and policymakers to assess the potential impact of AI on the labor market and job quality. Key achievements: - 3+ years of research and practice insights incorporated - 2 tools created for building a better future - 13 endorsements of the guidance The guidelines have already influenced practices in the industry, such as OpenAI's commitment to PAI's Data Enrichment Sourcing Guidelines for the development of ChatGPT4. PAI plans to work closely with labor organizers and industry partners in 2024 to increase adoption of the guidelines and develop standards of practice for fair treatment of data workers.

PAI extended its impact to various sectors facing AI-related challenges: 1. Local Newsrooms: Published 'AI Adoption for Newsrooms: A 10-Step Guide' to support newsrooms in navigating AI tool identification, procurement, and use. 2. Privacy: Collaborated with Apple to publish a white paper on advancing fair and ethical algorithmic decision-making while balancing data collection and privacy. 3. Public Policy: Facilitated coordination between policymakers, civil society, academia, and industry through events like the PAI Policy Forum and contributions to key policy conversations. 4. Inclusivity: Launched the Global Task Force for Inclusive AI, cited by the White House Office of Science and Technology Policy, to ensure AI benefits all people and society.

PAI's multi-stakeholder model has been crucial to its progress in 2023. The organization's reach and impact are reflected in its numbers: - 119 partners across 17 countries - 61 virtual convenings reaching 910 participants from 13 countries - 6 in-person convenings with 667 participants from 9 countries - 101 public speaking engagements - 44K+ social media followers - 15+ educational briefings for US and UK policymakers - 6 new resources with 6,600+ downloads PAI's work in 2023 demonstrated its ability to anticipate, prepare for, and respond to rapidly changing AI contexts. By bringing together diverse stakeholders, PAI continues to play a vital role in shaping a responsible and inclusive future for AI.