DeepMind is focusing on responsible data collection
DeepMind is dedicated to upholding the highest standards of safety and ethics in all of its pursuits. As a major part of this mission, DeepMind has worked in collaboration with the Partnership on AI (PAI) to create standardized best practices and processes for responsible human data collection.
Human Data Collection
Three years ago, DeepMind established the Human Behavioural Research Ethics Committee (HuBREC) to oversee research including experiments with human subjects, particularly in the field of AI decision-making. The AI community has also been increasingly involved in data enrichment, which includes tasks like data labeling and model validation. This area can raise ethical concerns related to worker welfare and pay, which led to the need for stronger guidance in these practices.
The Best Practices
DeepMind and PAI collaborated to develop best practices for data enrichment, including five specific steps that AI practitioners can follow. These steps focus on ensuring fair pay and working conditions for individuals involved in data enrichment tasks.
Looking Forward
DeepMind is committed to continuously improving its data collection practices and hopes that its work with PAI will spark broader conversations about how to develop responsible data collection norms and industry standards in the field of AI.
As part of its commitment to upholding the highest standards of safety and ethics, DeepMind has developed best practices and processes for responsible human data collection in collaboration with the Partnership on AI (PAI). This includes a focus on data enrichment and human data collection. The hope is that this work will lead to broader conversations and the development of industry standards for responsible data collection in the field of AI.