2.7 Transparency, Interpretability and Explainability

Question 2

AI-based systems are applied where users need to trust them

Answer 2

“For safety

Question 3

Most users interact with AI systems as

Answer 3

“Black boxes”

Question 4

Users’ awareness of interacting with AI

Answer 4

Occasionally, users are unaware

Question 5

Complexity of AI systems led to

Answer 5

Explainable AI (XAI)

Question 6

Aim of XAI

Answer 6

Users understand how AI systems get results, increasing trust

Question 7

Reasons for XAI (The Royal Society)

Answer 7

User confidence,

Safeguarding against bias,

Meeting regulatory standards/policy requirements,

Improving system design,

Assessing risk/robustness/vulnerability,

Understanding/verifying system outputs,

Autonomy/agency/meeting social values

Question 8

Desirable XAI characteristics (stakeholder perspective)

Answer 8

Transparency,

Interpretability,

Explainability

Question 9

Transparency (XAI)

Answer 9

Ease of determining what algorithm and training data was used to generate the model

Question 10

Interpretability (XAI)

Answer 10

Understandability of AI technology by various stakeholders, including users

Question 11

Explainability (XAI)

Answer 11

Ease with which users can determine how the AI system comes up with a particular result