Chapter 3: Data and Fairness in AI

Vocabulary flashcards covering key concepts from Chapter 3: Data and Fairness in AI, including dataset bias, fairness principles, and real-world examples.

Bias (in AI)

A condition where data does not represent everyone equally, making the training data unbalanced or unfair.

Joy Buolamwini

A computer scientist who tested facial recognition technology from major tech companies and discovered it worked well for light-skinned men but made more mistakes for women and people with darker skin tones.

Bias Check

A method to ensure fairness by testing AI programs to see if they recognise or treat various categories differently.

Human Supervision

The practice of having humans review AI decisions instead of trusting them blindly.

Transparency (in AI)

Understanding how an AI system works and how it makes its decisions.

Data Imbalance

A flaw in the training data where certain groups or classes are overrepresented while others are underrepresented, leading AI to favor more frequent examples and mispredict rarer ones.

Gender Stereotype in Machine Translation

A form of AI bias where an algorithm assigns gender to a gender-neutral input (such as translating the Hindi phrase "वह डॉक्टर है" to "He is a doctor") because the training dataset contained more male examples.

Teachable Machine

A Google web platform (teachablemachine.withgoogle.com) used to train machine learning models to classify image datasets, such as distinguishing between cats and dogs.