The paper describes a new large-scale dataset for action recognition, containing actions across many different countries, where faces are blurred for privacy reasons, and all videos have the Creative Commons license. R2 considers the dataset will be beneficial to the community, but recommends rejection, arguing that a more comprehensive experimental analysis related to bias would be needed for the paper to be accepted. While the concerns raised by R2 are legitimate, both the AC and SAC agree with R1 and R3 that the paper passes the acceptance bar of NeurIPS, as this work is a very important step towards datasets that are designed for everyone, for any culture, while attempting to protect privacy and copyright. The authors should add the clarifications in the rebuttal to the camera-ready version.