Defining AI Trustworthiness
An assessment of an organization’s AI solution would evaluate certain desirable characteristics for the AI system to contain before it can be deployed. Together, these characteristics contribute to the “trustworthiness” of the AI system. These characteristics correspond to different stages within the AI lifecycle and involve different “AI actors” and stakeholders. AI Actors can be defined as parties who are actively involved in the lifecycle of the AI system, such as those responsible for its development and deployment. Measuring AI trustworthiness should occur across the lifecycle of an AI system. According to a report published by the OECD that analyzes different approaches to AI Risk Management, there are a few common characteristics for trustworthiness that are embedded within the Risk Assessments:
-
Benefiting people and the planet:
Those who play an active role in the AI system lifecycle should engage in creating AI systems that can contribute to inducing inclusive growth, sustainable development, and well-being.
-
Human-centered values and fairness: AI actors should respect the rule of law, human rights, and democratic values throughout the AI system lifecycle.
-
Transparency and explainability: AI actors should promote greater awareness of their systems and commit to responsible disclosures to allow AI actors to comprehend the outcome and challenge the decision when needed.
-
Robustness, security, privacy, and safety: AI systems need to ensure traceability and function appropriately, and AI actors need to apply systematic risk management approaches to address, among others, safety and security risks.
-
Accountability: AI actors should ensure the proper functioning of AI systems and be accountable for upholding the above principles based on their roles, the context, and consistent with the state of the art.
