Page 58 - AI & Machine Learning for Beginners: A Guided Workbook

Loan Application Discrimination AI-driven lending models,

trained on historical financial data, have unfairly denied loans to
certain demographic groups. If past lending practices were biased,
AI may inherit and perpetuate these discriminatory patterns,
affecting financial inclusion.

Hiring Algorithm Bias AI tools used in hiring may favor

candidates with backgrounds similar to those of successful past
employees. This creates a feedback loop, potentially excluding
qualified candidates from underrepresented groups and
reinforcing existing workplace inequalities.

Addressing AI Bias

Ensuring diverse, unbiased training data, implementing fairness
audits, and regularly refining AI models are crucial steps toward
building more equitable AI systems.

How Bias Creeps Into AI Systems

Despite AI’s potential to enhance fairness and objectivity, bias can
infiltrate these systems through several key sources. Understanding
these factors is critical to developing more equitable AI models.

Biased Training Data – If an AI model is trained on

historically biased data, it inherits and reinforces those biases.
For example, if a facial recognition system is trained primarily on
lighter-skinned faces, it may struggle to accurately recognize
individuals with darker skin tones.

Flawed Algorithm Design – Sometimes, bias is

unintentionally embedded in an AI’s design. If an algorithm
prioritizes certain features over others or lacks diversity in its
training methodology, it may amplify existing inequalities rather
than correct them.

Human Bias in Data Labeling – Even when raw data is diverse,

the process of labeling or categorizing it can introduce bias.
56