While many organizations are reasonably beginning to consider the environmental impacts of AI, especially with the growing energy demands of generative AI, greater focus is still needed on the human cost embedded in the development of AI models.
This International Workers’ Day, as we celebrate labor movements and workers’ rights across the globe, we are reminded of the ongoing efforts to uphold human dignity and foster equity in the workplace. Data enrichment workers who perform the essential role of training, validating, and fine-tuning AI models are one of the many groups of workers who are still overlooked, lack recognition, and lack protections.
It is clear that building responsible AI means building responsible AI data supply chains. While the specifics of the EU Corporate Sustainability Due Diligence Directive (CSDDD) get ironed out, the debate around this regulation underscores the need to leverage risk-based due diligence practices grounded in internationally accepted human rights principles. Specifically, it is important to strengthen human rights and environmental protections across global supply chains, to improve conditions for data enrichment workers globally.
The Responsible AI Movement Needs to Include Data Workers
AI relies on data, but before that data can power algorithms, it must first move through a vast global data supply chain built by data enrichment workers. These workers label, categorize, and moderate data and information to make it usable for AI systems. Yet despite their essential role, data enrichment workers remain largely invisible and undervalued. The AI data supply chain is opaque, fragmented, and unregulated, leaving workers vulnerable to precarious working conditions, low wages, and even mental health impacts from reviewing traumatic content.
“Responsible AI cannot be achieved without responsible treatment of the people whose labor makes AI possible.”
Through our work at PAI, leading workshops, producing case studies, developing a resource library and targeted resources, and proposing pathways for responsible data enrichment, we’ve learned that addressing these issues at scale requires ecosystem-wide accountability. Two central and practical challenges stand out, making it difficult for the industry to change practices:
Internal decision fragmentation: internal decisions that meaningfully impact workers’ conditions are currently distributed across multiple teams and functions.
External complexity: the complexity and scale of the supply chain means that multiple organizations across a complex data supply chain are shaping decisions that impact workers’ livelihoods.
While these challenges require formalization and standardization of practices, they are not impossible to overcome. Other industries, such as garments, minerals, manufacturing, have faced similar supply chain dilemmas. Their experience shows us that existing human rights principles and governance frameworks can be adapted to AI data supply chains as well. We don’t need to start from scratch. We can apply what we already know works.
Bridging Human Rights and Data Supply Chains
We’re encouraged to see momentum building with more human rights impact assessments now including the impact on data enrichment workers. Our Guidelines for Responsible Data Enrichment Practices have been incorporated into the OECD’s upcoming Due Diligence Guidance on Trustworthy AI, signaling momentum on a global scale. However, work is still needed to bridge the gap between responsible data supply chain practices and human rights principles.
Organizations, including PAI, the OECD, and United Nations are already making the connection between human rights due diligence frameworks and responsible data supply chains. To advance this dialogue, we hosted a webinar, Human Rights Due Diligence Frameworks & Ethical Data Enrichment Supply Chains, bringing together experts from the UN OHCHR B-Tech initiative, the OECD’s Responsible Business Conduct group, and industry leaders from Cisco, Intel, and Google DeepMind. This discussion brought together diverse perspectives spanning human rights expertise, experience bridging principles across various supply chains, and experience implementing responsible AI practices. Five key insights emerged from that conversation, which we believe will enable us to better understand and tackle challenges in developing responsible AI data supply chains:
Human rights frameworks offer a strong foundation for protecting data enrichment workers: Internationally recognized human rights principles provide trusted foundational guidelines for navigating complex supply chains. They prioritize the rights of the most vulnerable and can help identify appropriate actions for preventing, mitigating, and remedying harms along the AI data ecosystem. These internationally recognized frameworks have been tested in complex contexts with multiple levels of responsibility, much like the data enrichment ecosystem.
Existing infrastructure can be leveraged: Many companies already have teams and processes focused on supply chain human rights. These can be expanded to address data enrichment work, accelerating progress without reinventing the wheel.
Data enrichment brings unique risks: While data enrichment workers face some challenges similar to those in traditional supply chains, they also face distinct issues unique to the mentally taxing work of data enrichment (e.g. the mental and emotional cost of reviewing traumatic content). Due diligence tools developed for traditional supply chains will need to be carefully adapted to address the unique challenges of work that has psychological harms.
The fragmented and varied nature of data supply chains must be addressed: Data enrichment work happens across offices and homes, through various employment structures including full-time work, contractors, subcontractors, gigwork. This makes it difficult to map out who is doing what and where. Addressing this complexity will require more transparency and standardized accountability mechanisms.
Worker voices are essential: Finally, we cannot design ethical AI supply chains without directly engaging workers themselves. Cross-sector collaboration, including with worker organizations, must be at the center of future frameworks.
Why This Matters
Data is the essential building block of AI. If the people who curate, label, and enrich that data are treated unfairly, it undermines not only the integrity of those systems, but also their safety, quality, and trustworthiness. Ignoring the conditions of data enrichment workers isn’t just a human rights failure, it’s also a technical and business risk.
By leveraging existing human rights frameworks and intentionally embedding responsible data enrichment practices within them, we can build AI systems that are not only more ethical, but also safer and more reliable.
This International Workers’ Day, let’s commit to building AI that respects every contributor, especially those whose labor remains unseen but essential. The foundation for responsible AI already exists in human rights frameworks. Now, it’s up to us to build on it.