Understanding AI Bias and Feedback Loops

What is algorithmic bias?

• A set of mathematical rules used to solve problems.

• A situation where real-world biases are mimicked or exaggerated by AI systems.

• A type of computer code that helps AI learn new things.

• A system that helps people make fair decisions.

What can happen if a facial recognition AI is trained using data that has many more examples of one group of people's faces than others?

• The AI will become more efficient at recognizing all faces.

• The AI might struggle to accurately recognize faces from underrepresented groups.

• The AI will develop a better understanding of human diversity.

• The AI will automatically correct any imbalances in its training data.

Why might an AI system designed to grade essays on standardized tests not accurately assess the quality of writing?

• It focuses on complex elements like creativity and structure.

• It prioritizes human-like understanding over measurable features.

• It often relies on easily measurable elements like sentence length and vocabulary.

• It is designed to be easily fooled by template essays.

In the context of AI, what does a "positive feedback loop" mean?

• The AI receives positive reviews from users, improving its popularity.

• The AI learns to correct its own biases over time.

• The AI amplifies past patterns or outcomes, whether good or bad.

• The AI provides helpful suggestions to improve data collection.

What caused Microsoft's Tay chatbot to start posting offensive tweets?

• It was programmed to be offensive from the start.

• It learned from biased and manipulated conversations with users.

• A technical glitch corrupted its core programming.

• It was hacked by a rival company.

In the "Hire Me!" AI system, why did John Greenbot receive a low likelihood of being hired, even if he was qualified?

• The system had a bug that always rejected applicants named John.

• The AI was trained on past data where other applicants named John were unsuccessful.

• John Greenbot's resume was incomplete.

• The hiring manager manually overrode the AI's initial positive recommendation.

What potential issue did the "Hire Me!" AI system show when it suggested hiring younger applicants?

• It incorrectly linked age with lower tech knowledge, leading to discrimination.

• It prioritized applicants with less experience, which is always a good strategy.

• It was designed to only hire people under 30.

• It failed to consider the applicant's education level.

What is the first important step in addressing algorithmic bias?

• Immediately rewrite all existing algorithms.

• Understand that algorithms will naturally have some bias.

• Only use AI for tasks that have no human impact.

• Collect less data to prevent bias from forming.