Lecture 8 - Responsible AI

Question 1

Why responsible AI? (3)

Answer 1

1. Bias in data
2. Unknowable AI
3. Inappropriate AI

These factors show that we have a responsibility as creators of AI and AI systems to guide how they’re implemented and created based on our social values.

Question 2

AI is more than an algorithm: 3 important features:

Answer 2

1. Adaptability
2. Interaction
3. Autonomy

Question 3

Ethics: taking responsibility: (3)

Answer 3

1. In Design, in design process
2. By Design, putting ethical reasoning in programmed behaviour of AI
3. For Design, societal embedding of AI limiting the users

Question 4

Ethics IN design:

Answer 4

1. Doing the right thing (goals) and doing it right (process)
2. There is a need for design methods
3. Importance of use plan: “Danger is not AI taking over the world but misuse and failures”

Question 5

AI = ART

Answer 5

1. Accountability
2. Responsibility
3. Transparency

Question 6

Responsible data science: FACT framework:

Answer 6

1. Fairness (without prejudice)
2. Accuracy (without guesswork)
3. Confidentiality (without revealing secrets)
4. Transparency (provide transparency, answers become indisputable

Question 7

Explainable AI has (5):

Answer 7

1. Understandability (understand its function)
2. Comprehensibility (represents its learned knowledge)
3. Interpretability (explain or provide the meaning of terms)
4. Explainability (produces details or reasons)
5. Transparency (if it is understandable by itself)

Question 8

Why Explainability? (7)

Answer 8

1. Trustworthiness, confidence
2. Causality
3. Transferability to other domains
4. Informativeness
5. Fairness
6. Interactivity
7. Privacy awareness

Question 9

Two XAI directions:

Answer 9

1. Glassbox models: make the model transparent from the start
2. Blackbox models: make non transparent models and explain the output by extra means.