CS5720 - Week 13
Slide 242 of 260

Bias in Deep Learning Models

Types of AI Bias

AI Bias occurs when algorithms systematically produce results that are unfairly prejudiced for or against certain groups of people.
🚨 The Hidden Danger
Bias in AI isn't always intentional - it often emerges from seemingly neutral decisions about data and algorithms.
  • 🤖
    Algorithmic Bias
    Unfairness built into the learning algorithm itself
  • 📊
    Data Bias
    Prejudice inherited from biased training datasets
  • 🧠
    Cognitive Bias
    Human prejudices that influence AI development
  • 📈
    Evaluation Bias
    Flawed metrics that miss important fairness aspects

Where Bias Comes From

Understanding the sources of bias is crucial for preventing and mitigating unfair AI systems.
  • 📚
    Historical Inequities
    Past discrimination embedded in historical data
  • 👥
    Representation Gaps
    Missing or underrepresented groups in data
  • 📏
    Measurement Bias
    Systematic errors in how data is collected
  • 🔄
    Aggregation Bias
    Inappropriate generalization across groups

Real-World Bias Examples

💼
AI Hiring Tools
Resume screening algorithms showing bias against women and minorities
⚖️
Criminal Justice Systems
Risk assessment tools with racial bias in recidivism predictions
🏥
Healthcare AI
Diagnostic systems performing poorly for underrepresented groups
👤
Facial Recognition
Higher error rates for women and people with darker skin
← Previous Next →
Prepared by Dr. Gorkem Kar