Glossary

AI Stakeholder Identification Methodology (Terms and Definitions)

AEQUITAS has developed a preliminary methodology for identifying relevant stakeholders to be involved in the design process of AI systems and to set a common ground for the used terminology. By utilizing a combination of desk research, expert knowledge, and resources from previous projects, we have created a questionnaire that guides the selection of targeted user groups and stakeholders to be involved in the design process of AI systems. The questions are divided into three categories to identify: #. Stakeholders affected by the AI system (‘Affectees’) #. Stakeholders that have power over the development and deployment of the AI system (‘Decisionmakers’) #. Stakeholders that have information that would aid with the development of a fair AI system (‘Domain Experts and Users’).

Identification of stakeholders affected by the AI-system (Affectées)

AI unfairness can lead to both negative and positive effects for different stakeholders depending on the case at hand. These stakeholders can include individuals or groups of individuals, such as citizens, patients, workers, students, children, parents, consumers, women, men, racial minorities, person(s) with a disability, person(s) with a low socio-economic status, person(s) with a high socio-economic status, etc. Beyond (groups of) individuals, AI-Fairness is also relevant for other types of stakeholders such as society at large, the environment, democracy, the economy, etc.

To identify stakeholders affected by AI unfairness the following questions serve as guidance:

  • Who/what could directly/indirectly be harmed by the AI unfairness in the case at hand?

  • Who/what could directly/indirectly benefit from the AI unfairness in the case at hand?

Once identified, the stakeholders can be categorized in two groups:

  • Positively affected

  • Negatively affected

Per group, the stakeholders can be further categorized based on the level of impact:

  • Directly affected

  • Indirectly affected

Identification of Decisionmakers

We define the Decionmakers as the takeholders with power over the development, deployment and governance of the AI-system.

Achieving AI-Fairness requires the involvement of those who have the power over the development, deployment and governance of AI systems. These stakeholders can include a CEO or manager, a head of a Government Agency. They can include Legal Compliance officers and CSR officers. But they can also include a responsible minister or state secretary, members of parliament or a parliamentary committee, a supervisory authority, a notified body etc.

To identify stakeholders with power over the development, deployment and governance of the AI system the following questions serve as guidance:

  • Who has the final decision to use the AI system?

  • Who is managing (aspects of) the AI project?

  • Who takes care of the governance of the AI system?

  • Who is auditing the AI system?

  • Who is regulating the AI system?

  • Who is supervising the AI system?

Once identified the stakeholders can be categorized into three groups:

  • Deployer (e.g., CEO, Manager, Head of Government Agency)

  • Governance (e.g., Legal Compliance Officer, CSR Officer, Sustainable Development Officer)

  • Authority/Supervisor (e.g., Privacy Authority, Market Authority, Financial Authority)

  • Policy (e.g., Minister, Parliamentary Committee)

Per group, the stakeholders can be further categorized based on the level of involvement necessary:

  • Direct involvement

  • Indirect involvement

  • Continuous involvement

  • Ad-hoc involvement

Identification of Domain Experts and Users

Domain experts and users are stakeholders that can aid the development of the AI system. Achieving AI fairness also requires the involvement of those who have specific expertise regarding the domain in which the system will be used. These include technical developers such as computer scientists, machine learning experts, data scientists, statisticians. They also include domain experts with expertise on the use case at hand, such as physicians, judges, lawyers, caseworkers, teachers, financial experts, etc. But domain experts also include people who are expected to use or work with the AI system once it has been deployed, such as salespeople, call centre employees, nurses, etc. To Identify stakeholders who have information that aids the development of the AI system the following questions serve as guidance:

  • Who is involved in the development of the AI system?

  • Who has domain expertise regarding the actions of the AI system?

  • Who (else) will be using/working with the AI system?

  • Who has a stake in understanding the workings of the AI system?

Per group, the stakeholders can be further categorized based on the level of involvement necessary:

  • Direct involvement

  • Indirect involvement

  • Continuous involvement

  • Ad-hoc involvement