AI bias doesn't appear magically. It enters systems through specific pathways that designers can identify and address:

• Data bias occurs when training materials don't represent all users equally, leading algorithms to perform better for the majority groups. For example, a healthcare AI trained primarily on data from male patients might miss important symptoms that present differently in women.
• Societal biases often occur because of inaccurate information taken from historical datasets
• Algorithmic bias emerges during model development when certain features receive disproportionate weight or when optimization targets inadvertently favor particular outcomes.
• Interaction bias happens at the user interface level when design choices create different experiences for different groups.^[1]

For example, a voice assistant might struggle with certain accents, or a photo app might apply "beauty" filters that reflect narrow cultural standards. Designers must audit for bias at each phase: during data collection by ensuring diverse representation, during model development by testing across demographic groups, and during interface design by making systems adaptable to different user needs. Addressing bias requires ongoing vigilance rather than one-time fixes.