Governing Ethical AI in Organizations: Stakeholder Perspectives on Barriers and Enablers

In today’s world, the need for ethical AI is at an all-time high due to the fact that AI systems are becoming more autonomous and less transparent, as well as the number of people impacted by their decision-making processes growing exponentially with those that are the most vulnerable. This qualitative study examined the views of stakeholders regarding the barriers and facilitators to implementing ethical artificial intelligence in technology organisations. Using thematic analysis to analyse 39 semi-structured interviews conducted with AI developers, organisational leaders, ethicists, and community members impacted by AI across 12 organisations, five main themes were identified: (1) The Principle-Practice Gap, where although organisations articulate their ethical principles, they are unable to implement these due to resource and operational demands; (2) Structural and Cultural Barriers to Ethical Implementation of AI, including the absence of governing bodies, the absence of organizational leadership support, and the presence of conflicting organizational priorities; (3) Algorithmic bias and fairness as persistent implementation challenges; (4) Privacy, transparency, and governance complexities; and (5) Pathways to ethical implementation, which included the utilisation of governance frameworks, the diversity of the team, and the involvement of stakeholders to increase the likelihood of ethical implementation of AI. Findings suggest that the successful implementation of ethical AI will depend upon the organisation-wide transformation of culture and values beyond just implementing technical solutions for ethical AI, which further will be dependent upon the authentic commitment of leadership and the inclusion of ethics throughout every aspect of the development life cycle. These findings contribute significantly to the body of knowledge in AI governance, organisational best practices, and policy development for promoting responsible AI.