Prioritizing Workers in the US Global AI Research Agenda

Recognizing AI’s potential for both good and harm is essential to developing systems that benefit everyone. The United States is shaping the frontier of these AI systems and products and has emphasized its commitment to leading responsible AI development and deployment on a global scale. This commitment is demonstrated in the recently published Global AI Development Playbook, which was developed in response to the Biden administration’s Executive Order (EO) emphasizing Safe, Secure, and Trustworthy AI Development.

The Playbook, a key resource and deliverable of the Administration, underscores AI as a global opportunity with significant impact on humanity. Although AI can benefit society it poses risks to populations vulnerable to the adverse effects of AI systems. The Playbook marks a step forward in the United States’ commitment to leading the charge in creating global alignment in AI policy that is rooted in human rights and sustainable development. 

The lack of global policy and regulation consensus on AI research and development prompted the EO to direct the United States Agency for International Development (USAID) and the U.S. Department of State in collaboration with the Department of Energy and the National Science Foundation (NSF) to develop the Global AI Research Agenda. The Research Agenda aimed to ensure the safe and responsible global development of AI through international research collaborations, study AI human impacts, and address labor market shifts due to AI. 

PAI’s Recommendations

Partnership on AI has contributed to the Research Agenda by responding to the call for public comment to inform the Agenda and Playbook. The Agenda and Playbook includes many of PAI’s key recommendations, which include: 

1. Prioritize in-depth research of the data supply chain to map various types of employment models that data enrichment workers fall under and assess the prevalence of data enrichment employment models, analyze the diverse actors involved across the global supply chain, and investigate the impact of these models on workers. 
2. Pursue empirical studies examining whether and to what extent the circumstances and labor conditions surrounding dataset construction affect their effectiveness, reliability, and safety. 
3. Examine how to design markets that promote fairer outcomes for those involved in the data supply chain.
4. Study job quality impacts in addition to job availability and wage impacts in its focus on labor market impacts.
5.  Provide opportunities for non-academic participation and engagement, with a specific focus on the inclusion of perspectives of workers and impacted communities.

These recommendations highlight important but often overlooked areas in AI policy. They aim to ensure that AI development does not just advance technology but also contributes to fairer and safer outcomes for workers and people affected by it. A deeper understanding of data enrichment workers’ roles in the development of these systems can shed light on unfair labor conditions which can and do affect the quality and reliability of AI systems. We have spent the last five years, at PAI, dedicated to addressing these challenges, outlining a pathway to responsible data enrichment practices for worker representatives, policymakers, auditors, and other actors across the supply chain. 

Our recently released drafts of ​​the Vendor Engagement Guidance help AI-developing companies promote responsible practices with downstream actors, while the Transparency Template outlines what companies should be monitoring and reporting on with respect to their data enrichment practices. Our Data Enrichment Sourcing Guidelines influence how AI practitioners involve and work with data enrichment workers to ensure proper working conditions.

Our recommendations suggest that the Research Agenda follow a broad definition of “labor market impacts,” to include an examination of AI’s effect on job availability, wages, and job quality. 

The Agenda provides a unique opportunity to pursue underexplored research areas identified by our previous work, such as the impact of AI on informal labor markets, anticipated AI exposure in low and middle-income economies, and mechanisms to strengthen worker voice. Including data enrichment workers, the hundreds of millions of people who perform the essential task of cleaning and labeling datasets, in these conversations is crucial to improving business outcomes for workers and companies. Affected workers and communities in particular have expertise and real-world experience that can be invaluable in informing research efforts. See PAI’s brief for the United Nations Department of Economic and Social Affairs’ 2023 Multi-Stakeholder Forum on Science, Technology, and Innovation for the SDGs on the importance of a multistakeholder approach.

Integrating Multistakeholder Voices in AI

As policymakers at national and international levels work to govern AI development, deployment, and usage, it is imperative to bring ideas from across sectors and disciplines to the policy discussion, so that the solutions work for people, not just companies. We encouraged the USAID, U.S. Department of State, Department of Energy, and the NSF to work with organizations, such as ours, in the design and implementation of the Global Research Agenda. A multistakeholder approach to developing the Agenda will ensure innovative and novel approaches are reflected in the Agenda and its resulting deliverables. At the intersection of industry, civil society, government, and academia, Partnership on AI plays a vital role in shaping responsible global AI governance by bringing together experts from across sectors. PAI continues to work on addressing the most important and difficult questions concerning the future of AI. To stay up to date on our work in this space sign up for our newsletter.