0.2 AI Safety, Bias, & Fairness

Autonomous

AI systems ->

Affect Human Lives

AI decisions ->

Safe, Reliable, Fair

aside from being accurate, AI systems must satisfy the ff. requirements: _ _ _ (1)

Transparent, Accountable

aside from being accurate, AI systems must satisfy the ff. requirements: _ _ (2)

Bias

systematic errors that unfairly favor or disadvantage certain individuals or groups

Bias

does not necessarily arise because an AI system is intentionally discriminatory

Bias

often originates from the data, assumptions, objectives, or measurement processes used during system development

Patterns

AI learns _, not morality

Data Bias

occurs when the training data fails to represent the popularity accurately

Algorithm Bias

even when data appears balanced, algorithms may still amplify undesirable patterns

Algorithm Bias

optimization procedures attempt to maximize predictive accuracy

Measurement Bias

sometimes the wrong quantity is measured

Equal Treatment

everyone undergoes the same decision-making procedure

Equal Opportunity

individuals with similar qualifications should have similar chances of receiving favorable outcomes

Equal Outcomes

different demographic groups should receive comparable overall outcomes