Page 59 - AI & Machine Learning for Beginners: A Guided Workbook

Human annotators may subconsciously favor certain patterns,
leading to skewed results when the AI learns from this labeled
information.

Mitigating AI Bias

To reduce bias, AI developers must:

Use diverse and representative training datasets

Audit algorithms for unintended biases

Improve transparency and accountability in AI decision-

making
Activity: Identifying Potential Bias in AI Hiring

Scenario: A company develops an AI tool to screen job
applications. The AI is trained on data from the company’s past
successful hires, who are predominantly male.

Potential Biases in this AI System

Gender Bias: The AI might favor male candidates based on

historical hiring patterns.

Lack of Diversity: If the system learns from past hires, it may

unfairly disadvantage women and underrepresented groups.

Skill & Experience Filtering: If past hires had similar

backgrounds, the AI may prioritize familiar profiles rather than
evaluating all applicants fairly.

57