Home/AI/Bias in AI Systems/Introduction

⚖️ Bias in AI Systems

Understand, detect, and mitigate bias to build fair and equitable AI models

Your Progress

0 / 5 completed

←

Previous Module

Knowledge Distillation

Understanding Bias in AI

🎯 What is AI Bias?

AI bias occurs when a model produces systematically unfair outcomes for certain groups of people. This can result from biased training data, flawed algorithms, or biased interpretations of results.

⚠️

Critical Impact

Biased AI systems can perpetuate discrimination in hiring, lending, healthcare, and criminal justice.

📊 Real-World Examples

💼

Hiring Algorithms

Amazon's recruiting tool penalized resumes containing "women's" (chess club, colleges)

Trained on historical data from male-dominated tech industry

👮

Criminal Risk Assessment

COMPAS system showed higher false positive rates for Black defendants

Predicted future crimes with racial disparities

🏥

Healthcare Algorithms

Risk prediction tools underestimated health needs of Black patients

Used healthcare spending as proxy for health needs

👤

Facial Recognition

Higher error rates for women and people with darker skin tones

Training data lacked diversity in skin tones and genders

🔍 Types of Bias

Historical Bias

Training data reflects past societal prejudices and inequalities

Example: Word embeddings associate "programmer" with male pronouns

Representation Bias

Certain groups are underrepresented or misrepresented in training data

Example: Image datasets with fewer examples of certain demographics

Measurement Bias

Proxy variables don't accurately measure the intended concept

Example: Using zip code as proxy for creditworthiness

Evaluation Bias

Benchmark datasets don't represent real-world diversity

Example: Testing on homogeneous test sets

⚖️ Fairness Definitions

Multiple mathematical definitions of fairness exist, often in conflict:

Demographic Parity

Equal positive prediction rates across groups

P(Ŷ=1 | A=0) = P(Ŷ=1 | A=1)

Equal Opportunity

Equal true positive rates across groups

P(Ŷ=1 | Y=1, A=0) = P(Ŷ=1 | Y=1, A=1)

Equalized Odds

Equal TPR and FPR across groups

TPR and FPR equal for all A

Calibration

Predictions equally accurate across groups

P(Y=1 | Ŷ=p, A=0) = P(Y=1 | Ŷ=p, A=1)