AI is powered by data enrichment workers, the people who prepare, clean, and label the large datasets machine learning (ML) models are trained on. Despite growing awareness of their often precarious working conditions, there has been limited transparency about how AI practitioners are sourcing enriched data from these workers and little guidance on how they should. This not only impacts the wellbeing of data enrichment workers, it also affects the quality of the data AI technology is built on.
To help fill this gap, Partnership on AI (PAI) is sharing the Responsible Data Enrichment Sourcing Library, a set of resources AI organizations can use to formalize their data enrichment practices and have a positive impact on the lives of data enrichment workers. These resources, based on PAI’s previous work and developed in partnership with DeepMind, have been put into practice by DeepMind. To demonstrate the process we took to transform PAI’s recommendations into actionable guidance for an applied setting, we are sharing an accompanying case study. This case study also details why DeepMind committed to responsible sourcing practices, how we collaborated, and the impact of adopting our recommendations.
At the heart of the Responsible Data Enrichment Sourcing Library are five worker-centric guidelines that AI practitioners should follow when designing a project involving enriched data. By providing these guidelines and resources, we aim to lower the barriers for AI organizations to improve their data enrichment practices and commit to better practices.
In accordance with our mission to advance positive outcomes for people and society, PAI undertook this collaboration in the public interest. We thank DeepMind for collaborating on this project, both increasing transparency about how enriched data is sourced and demonstrating the results that standardizing data enrichment practices can have.
PAI is committed to making a positive difference in the AI field. To ensure that we are building an actionable path for AI practitioners to improve conditions for workers and that any developed guidance truly results in better conditions for data enrichment workers, we will continue to build on these resources. We plan to incorporate feedback from AI practitioners on usability, workers on their resulting experience, and the broader AI community on how to further raise standards for workers.
If you are interested in acting on these guidelines and resources or have feedback on how they can be improved, please get in touch.