Module 11: AI Ethics and A Brief Introduction to Morality

Question 1

Identify and explain the three core components of AI Trust

Answer 1

1. Understanding – Define hazards and thresholds
2. Action – Guardrails to mitigate hazard likelihood & severity
3. Explanations – Communicate risk behaviors & events, i.e. compliance documentation & visualizations.

Question 2

List the five defined measurements of performance

Answer 2

1. Data Quality
2. Accuracy
3. Robustness
4. Stability
5. Speed

Question 3

List the five defined measurements of operations

Answer 3

1. Monitoring
2. Compliance
3. Security
4. Humility
5. Business Rules

Question 4

List the four defined measurements of ethics

Answer 4

1. Interpretability
2. Bias/Fairness
3. Governance
4. Social Impact Assessment

Question 5

Dataset is skewed towards certain group or may not reflect the real world

Answer 5

Skewed Sample

Question 6

Feature collection for certain groups may be informative or reliable.

Answer 6

Limited Features

Question 7

Unreliable label, historical bias

Answer 7

Tainted Examples

Question 8

Do we have enough data?

Answer 8

Sample Size Disparity

Question 9

Zip Code or school can be proxies for race School or sport activity can be proxies for gender.

Answer 9

Proxies

Question 10

What are four suggestions are made for tackling AI bias?

Answer 10

1. Identify- Identify protected features in your dataset
2. Select- an appropriate fairness metric for your use case and value system
3. Build insights to identify & understand your model’s potential bias
4. Mitigate bias uncovered in your data or model

Question 11

Transform your data such that the target would not be correlated to protected attributes

Answer 11

Pre-processing

Question 12

Modifying the loss function in your algorithm to have fairness constraints

Answer 12

In-processing

Question 13

Changing the model predictions to avoid discrimination.

Answer 13

Post-processing