AI-Bias-and-Fairness WGU D333

AI Bias

The presence of systematic and unfair discrimination in artificial intelligence algorithms or systems.

Fairness in AI

Ensuring that AI systems treat all individuals and groups equally, without discrimination.

Skewed Sample

A sample that does not fairly represent all groups or situations, leading to biased training results.

Limited Features

A situation where some data types are far more common than others in the training set.

Tainted Examples

Training data that contains biased or incorrect information, potentially influencing AI outputs negatively.

Proxy Bias

Bias that arises when neutral information is linked to sensitive topics like race or social class.

Fairness Metric

A measure to evaluate how fair an AI system is, assessing treatment of different groups.

Protected Class

A category of individuals who share a legally protected characteristic against discrimination.

Protected Feature

The specific attributes within a protected class that should not unfairly influence AI decisions.

Pre-processing

Altering training data before feeding it into an AI system to reduce bias.

In-processing

Adjusting the learning process of an AI during training to promote fairness.

Post-processing

Modifying an AI's outputs after decision-making to ensure fair results.

Disparity

An unequal distribution of data types within a training dataset that can lead to biased models.

Biased Information

Data that reflects stereotypes or falsehoods, affecting AI behavior and accuracy.

Cultural Representation

The inclusion and equitable representation of diverse cultures in AI training data.

Data Balance

Ensuring equal representation of all groups and types in AI training datasets to avoid bias.

Algorithmic Fairness

The principle that algorithms should be designed to treat all groups fairly.

Bias Detection

The process of identifying and measuring bias present in AI systems or data.

Equitable AI

AI systems that provide equal treatment and outcomes across different demographic groups.

Stereotype Reinforcement

The phenomenon where AI systems perpetuate harmful stereotypes present in the training data.

Diversity in Data

The representation of a wide range of demographics and attributes in input data for AI systems.