higher dimensional bias in BERT disinformation classifier

Posting here, so that we can track not only articles and issues of general
relevance to the broader aspects of AI KR, but also specifics to be
followed up in technical work.
How can AI KR help to define and address Higher dimensional bias (to be
defined @patrick)
Higher-dimensional bias in a
BERT-based disinformation classifier

see the original resource in Github
https://github.com/NGO-Algorithm-Audit/Bias_scan/blob/master/Technical_documentation_bias_scan.pdf