Protecting AI’s Essential Workers: Introducing our Vendor Engagement Guidance & Transparency Template

Artificial intelligence relies heavily on enriched datasets which in turn rely heavily on data enrichment workers. These people work behind the scenes to label, categorize, annotate, or otherwise enrich datasets which are then used to train AI systems. Although this work is crucial to the development of AI technology, it is often overlooked and undervalued. Research has shown that these workers often face unfair working conditions, low wages, unclear expectations, and little support in their work. 

In addition to ensuring that data enrichment workers enabling AI are treated fairly, the conditions under which datasets are constructed shape the resulting AI models. The well-being of data enrichment workers directly impacts the quality, safety, and reliability of data-driven systems so it is imperative that we advocate for just pay and proper working conditions. 

At Partnership on AI, we have focused on this issue since 2020, including hosting several workshops and publishing PAI’s Data Enrichment Sourcing Guidelines. In June 2023, we formed a Community of Practice made up of individuals working at AI-developing companies, focused on addressing barriers to adopting responsible data enrichment practices within their organizations. 

Based on our research, insights from this Community of Practice, and conversations with our broader partner network, we outlined a five-step “Path for Developing Responsible AI Supply Chains”. These actions are initial steps to enable better accountability and governance across the data supply chain.

1. Adopt PAI’s Data Enrichment Guidelines 
2. Introduce Internal Governance
3. Promote Consistent Practices Across the Supply Chain
4. Publish Transparency Reports
5. Include Workers’ Voices when Evolving Best Practices

To help companies better achieve these actions, we are developing supplementary resources related to each of the five actions. Today we are sharing resources for public comment that correspond with steps 3 and 4: the Vendor Engagement Guidance to help AI-developing companies promote responsible practices with downstream actors, and the Transparency Template to outline what companies should be monitoring and reporting on with respect to their data enrichment practices. 

These resources can serve as a step towards building shared responsibility across the data supply chain and enable greater accountability for the actors that are directly shaping working conditions for data enrichment workers. To make these resources as effective as possible, we are seeking feedback from the experts across the PAI community. 

Promoting Consistent Practices Across the AI Supply Chain

Steps within the Path reflect insights from a 2023 workshop with AI-developing companies which focused on barriers to adoption. In particular, the third step, Promote Consistent Practices Across the Supply Chain, responds to the finding that advancing just labor conditions for data enrichment workers will require AI-developing companies to work closely with other actors across the supply chain, including downstream vendors. To facilitate this, we have developed Vendor Engagement Guidance. 

Read the Vendor Engagement Guide

This resource outlines key questions related to how workers are being treated that AI-developing companies should be discussing with their downstream vendors at different phases of their engagement: Vendor Vetting,  Contract Negotiation, Project Management/Monitoring, and Post-Project. Given the distributed decision-making across the supply chain, intentionally assessing the impact on workers sometimes gets overlooked in favor of discussions around cost and timeline.  We hope that by outlining the questions and topics that should be asked about worker experience throughout the vendor engagement process, impact on workers will be brought to the forefront. These questions are intended to get companies across the supply chain aligned on their policies, better monitor their practices, and continuously improve their practices.

Promoting Transparency around Data Enrichment Practices

The fourth step in the Path is Publish Transparency Reports. This responds to an insight from our June 2023 workshop that greater transparency and accountability are necessary to drive widespread adoption of responsible data enrichment practices across the value chain and AI industry. With this as the goal, we developed the Transparency Template. 

Explore the Transparency Template

This serves as a guide for what companies should be monitoring and reporting regarding their data enrichment policies, internal governance mechanisms, and processes for working with vendors. In order to enable a broader ecosystem of accountability for stakeholder’s actions around data enrichment practices, we believe that there is a need to increase transparency around data enrichment practices across global data supply chains. By creating a practice of monitoring and reporting out on current data enrichment practices, we hope that other key stakeholders will be better positioned to hold actors in the supply chain accountable for their impact on workers and help advance the dialogue on how to consistently push for better conditions for workers.

Seeking Feedback on these Resources

To ensure that these resources have the intended impact, we believe in the importance of gathering and incorporating multistakeholder input on these resources. We are seeking feedback from:

• Labor organizations, advocates, data enrichment workers and their representatives to comment on whether the content and scope of these resources align with what they are advocating for and will have the intended impact.
• Human rights experts to better align these resources with existing human rights guiding principles and due diligence frameworks.
• Supply chain and auditing experts who can share learnings from other supply chains on how to make these resources feasible to implement and maximize their potential impact.  
• Policymakers and civil society on whether this type of information would help them hold actors across the supply chain accountable for their actions.
• Different departments at companies that might be involved in data enrichment to share the feasibility of using these resources.

We welcome feedback from all sides of this multifaceted issue and we are excited to foster dialogue with diverse perspectives to develop resources informed by a multistakeholder community. If you have thoughts on how we can improve these resources, who we should be getting feedback from, or who we should be engaging in the development of future resources, please submit your feedback here!  We’re particularly interested in getting feedback on the content, feasibility, and impact of these resources.

Register for the Sept. 12 Public Feedback Session

In an effort to gather more feedback, we will be hosting targeted workshops and listening sessions. If you are interested in learning more about these resources, how they were created, and how we plan to gather input on them, please join our webinar on September 12th, 11 am ET. You can register here.  By gathering multistakeholder input on these resources, we are hopeful we can advance just labor conditions for data enrichment workers.