Yes. It is a training issue. They need to train the algorithm in a different way as it relates to children and not use certain practices, which they are currently using, with this data and metadata, and only be based on explicitly stated interests for children.
No comments