How do you assess the impact of AI on stakeholders?

1 Identify the stakeholders

The first step is to identify who are the stakeholders of the AI system, and what are their interests, values, and expectations. Stakeholders can be directly or indirectly involved in the design, development, deployment, or use of the AI system. For example, stakeholders of a facial recognition system may include the developers, the clients, the users, the subjects, the regulators, and the public. Each stakeholder may have different goals, preferences, and perspectives on the AI system.

2 Analyze the impacts

The next step is to analyze the impacts of the AI system on the stakeholders, both positive and negative, intended and unintended, short-term and long-term. Impacts can be measured by various criteria, such as efficiency, accuracy, fairness, privacy, security, accountability, transparency, trust, autonomy, well-being, and human dignity. For example, an impact analysis of a facial recognition system may consider how it affects the performance, convenience, safety, identity, consent, rights, and responsibilities of the stakeholders.

3 Apply ethical principles

The third step is to apply ethical principles to guide the evaluation of the impacts and the decision making process. Ethical principles are general norms or values that reflect the moral standards and expectations of a society or a community. There are different ethical frameworks and codes of conduct that can be used to assess AI systems, such as the Asilomar AI Principles, the IEEE Ethically Aligned Design, the EU Ethics Guidelines for Trustworthy AI, and the ACM Code of Ethics and Professional Conduct. For example, an ethical assessment of a facial recognition system may use the principles of beneficence, non-maleficence, justice, respect for human rights, and social responsibility.

4 Engage in dialogue

The fourth step is to engage in dialogue with the stakeholders and solicit their feedback, opinions, and concerns. Dialogue is a process of communication and collaboration that aims to foster mutual understanding, learning, and trust among the stakeholders. Dialogue can be conducted through various methods, such as surveys, interviews, focus groups, workshops, consultations, and participatory design. For example, a dialogue with the stakeholders of a facial recognition system may explore their views on the purpose, scope, limitations, risks, and opportunities of the system.

5 Implement solutions

The fifth step is to implement solutions that address the impacts and the ethical issues identified in the previous steps. Solutions can be technical, organizational, or regulatory, depending on the nature and the severity of the problem. Solutions can also be proactive or reactive, depending on the stage and the context of the AI system. For example, some possible solutions for a facial recognition system may include improving the data quality and diversity, applying privacy-preserving techniques, providing user control and consent options, ensuring accountability and oversight mechanisms, and complying with relevant laws and policies.

6 Monitor and evaluate

The sixth step is to monitor and evaluate the outcomes and the effects of the solutions implemented in the previous step. Monitoring and evaluation are essential for ensuring the quality, effectiveness, and sustainability of the solutions, as well as for identifying new or emerging impacts and ethical issues. Monitoring and evaluation can be done through various indicators, metrics, and methods, such as performance tests, audits, reviews, feedback loops, and impact assessments. For example, a monitoring and evaluation of a facial recognition system may examine how it meets the criteria, principles, and expectations of the stakeholders, and how it can be improved or modified if needed.

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?