Explainability & Human Trust in AI

Question 1

What is a common basic definition of trust?

Answer 1

Trust is a directional transaction between two parties where A believes that B will act in A’s best interest, accepting vulnerability to B’s actions.

(Jacovi et al., 2021)

Question 2

According to the APA, how is interpersonal trust defined?

Answer 2

Interpersonal trust is the reliance on or confidence in the dependability of someone or something, specifically the degree to which each party can depend on the other to do what they say they will do.

Question 3

What are trust-relevant situations characterized by?

Answer 3

Trust-relevant situations involve vulnerability and require some stake, such as social, financial, personal, or organizational risk.

(Parkhe & Miller, 2000)

Question 4

What is affect-based (or social) trust?

Answer 4

Affect-based trust is formed when a trustee is perceived as well-minded, warm-hearted, and adhering to social norms.

Question 5

What is cognition-based (or performance-based) trust?

Answer 5

Cognition-based trust is formed when a trustee is perceived as competent, understandable, and predictable in terms of the performance required.

Question 6

What is the potential issue with inappropriate over-trust in AI systems?

Answer 6

Inappropriate over-trust may lead a human operator to overestimate a system’s abilities and rely too much on them.

Question 7

What is the significance of cognitive components of trust in AI?

Answer 7

Cognitive components should be more relevant for robotics/AI than affective components, as users must understand the AI’s competence to perform tasks.

Question 8

What does Explainable AI (XAI) aim to achieve?

Answer 8

XAI aims to make AI decision-making processes understandable to humans, moving away from the ‘black box’ nature of traditional AI models.

Question 9

List the key points of Explainable AI.

Answer 9

• Trust
• Accountability
• Regulatory Compliance
• Challenges in Achieving Explainability
• Approaches to XAI

Question 10

What are the two mnemonics mentioned for remembering key concepts in AI trust and explainability?

Answer 10

• TRUST: Transparency, Reasons, Understanding, Simplified models, Tools
• ABCD: Avoid over-reliance, Beware of under-trust, Cultivate balanced trust, Develop transparency and education

Question 11

What does Automation Bias refer to?

Answer 11

Automation Bias refers to the tendency of individuals to over-rely on automated systems, ignoring contradictory information.

Question 12

What is Algorithm Aversion Bias?

Answer 12

Algorithm Aversion Bias involves a preference for human judgment over algorithmic recommendations, even when algorithms are more accurate.

Question 13

What is the ‘Goldilocks Zone’ of Trust in AI?

Answer 13

The ‘Goldilocks Zone’ of Trust refers to a balanced level of trust in AI that is neither excessive nor insufficient.

Question 14

What are barriers to trustworthiness in AI?

Answer 14

• Lack of Transparency
• Bias in Data
• Overconfidence

Question 15

What components are essential for developing trustworthy AI systems?

Answer 15

• Transparency
• Robustness
• Uncertainty Quantification
• Safeguarding against biases
• Thorough testing and validation

Question 16

True or False: Explainability is the only solution to address challenges posed by AI systems.

Answer 16

False

Question 17

What should be integrated with explainability to build trustworthy AI systems?

Answer 17

Explainability should be integrated with robust regulatory frameworks, ethical guidelines, and continuous monitoring.