Bias in training data

If training data over-represents one group, AI becomes biased. A facial recognition system trained mostly on light-skinned faces works worse for darker skin.

• Unbalanced data creates biased AI.
• Diverse representation in data reduces bias.