Volunteer taggers at a folk-dance archive mark dialect speech as “rude” more often than the same content in the house dialect. The model then copies that habit. Where did the bias enter?
Select an answer to reveal the explanation.
Short Explanation
Taggers mark dialect as “rude” more often than the house dialect. That habit became the training signal. Bias entered during labeling, not only in who was filmed, and not in batch size or a missing kernel.
Full Explanation
Bias can be introduced while data is labeled, not only in who appears in the recordings. When taggers systematically mark dialect as rude, the model learns that habit. Serving-batch size and hardware kernels are unrelated. Associate items should locate the labeling process as a first-class bias path.