| Inter-Pillar Relationships |
| Pillar: Operational Integrity |
|
Governance vs. Interpretability |
■ Reinforcing
|
Governance supports interpretability by enforcing standards to ensure AI systems are understandable and transparent
(Bullock et al., 2024 )
|
AI regulations mandate interpretability to validate algorithmic outputs, ensuring systems comply with governance frameworks
(Bullock et al., 2024 )
|
|
Interpretability vs. Robustness |
■ Tensioned
|
Interpretability can compromise robustness due to increased complexity in models
(Hamon et al., 2020 )
|
Interpretable models in safety-critical applications may reduce robustness, increasing vulnerability to adversarial attacks
(Hamon et al., 2020 )
|
|
Explainability vs. Interpretability |
■ Reinforcing
|
Explainability aids interpretability by clarifying complex model outputs for user understanding
(Hamon et al., 2020 )
|
In financial AI, explainable models improve decision insight, ensuring model actionability
(Hamon et al., 2020 )
|
|
Interpretability vs. Security |
■ Tensioned
|
Security demands limited openness; interpretability requires transparency, creating inherent conflict
(Bommasani et al., 2021 )
|
Interpretable models in healthcare might expose vulnerabilities if too transparent, affecting security
(Rudin, 2019 )
|
|
Interpretability vs. Safety |
■ Reinforcing
|
Interpretability aids safety by enhancing understandability and identifying system flaws
(Leslie, 2019 )
|
In medical AI, interpretable models allow doctors to verify predictions, improving safety
(Leslie, 2019 )
|
| Cross-Pillar Relationships |
| Pillar: Ethical Safeguards vs. Operational Integrity |
|
Fairness vs. Interpretability |
■ Reinforcing
|
Interpretability fosters fairness by making opaque AI systems comprehensible, allowing equitable scrutiny and accountability
(Binns, 2018 )
|
Interpretable algorithms in credit scoring identify biases, supporting fairness standards and promoting equitable lending
(Bateni et al., 2022 )
|
|
Inclusiveness vs. Interpretability |
■ Reinforcing
|
Interpretability enriches inclusiveness by ensuring AI systems are understandable, fostering wide accessibility and equitable application
(Shams et al., 2023 )
|
Interpretable AI frameworks enable diverse communities’ meaningful engagement by clarifying system decisions, supporting inclusive practices
(Cheong, 2024 )
|
|
Bias Mitigation vs. Interpretability |
■ Reinforcing
|
Interpretability aids bias detection, supporting equitable AI systems by elucidating model decisions
(Ferrara, 2024 )
|
Interpretable healthcare models reveal biases in diagnostic outputs, promoting fair treatment
(Ferrara, 2024 )
|
|
Accountability vs. Interpretability |
■ Reinforcing
|
Accountability and interpretability enhance transparency and trust, essential for effective AI system governance
(Dubber et al., 2020 )
|
In finance, regulators use interpretable AI to ensure banks’ accountability by tracking decisions
(Ananny & Crawford, 2018 )
|
|
Interpretability vs. Privacy |
■ Tensioned
|
Privacy constraints often limit model transparency, complicating interpretability
(Cheong, 2024 )
|
In healthcare, strict privacy laws can impede clear interpretability, affecting decisions on patient data
(Wachter & Mittelstadt, 2019 )
|
| Pillar: Operational Integrity vs. Societal Empowerment |
|
Interpretability vs. Sustainability |
■ Neutral
|
Interpretability and sustainability operate independently, focusing on different AI aspects
(van Wynsberghe, 2021 )
|
An AI model could be interpretable but unsustainable due to high computational demands
(van Wynsberghe, 2021 )
|
|
Human Oversight vs. Interpretability |
■ Reinforcing
|
Human oversight bolsters interpretability by guiding transparency in AI processes, ensuring systems remain clear to users
(Hamon et al., 2020 )
|
Interpretable algorithms in medical AI gain user trust through human-supervised transparency during their development
(Doshi-Velez & Kim, 2017 )
|
|
Interpretability vs. Transparency |
■ Reinforcing
|
Interpretability enhances transparency by providing insights into AI mechanisms, fortifying user understanding
(Lipton, 2016 )
|
Transparent models boost public trust, as stakeholders understand how AI decisions are made clearly
(Lipton, 2016 )
|
|
Interpretability vs. Trustworthiness |
■ Reinforcing
|
Interpretability boosts trustworthiness by enhancing users’ understanding, encouraging confidence in AI systems
(Rudin, 2019 )
|
Understanding AI predictions in healthcare improves trust in medical diagnostics
(Rudin, 2019 )
|
RAISEF
