Putting ethical principles at the core of the research lifecycle
Hugging Face 4 years ago 43
Hugging Face's multimodal learning group published an ethical charter establishing principles to address potential harms from machine learning, including data privacy issues, algorithmic bias, and misuse. The charter, finalized in May 2022, outlines five core values: transparency about data sources and decisions, open and reproducible research accessible to non-specialists, fairness monitoring to mitigate unwanted biases, self-criticism about imperfections, and proper credit attribution. The team will document how these principles are operationalized throughout the project while acknowledging that some values may conflict and require case-by-case evaluation of risks and benefits.