AI ethics problem may lie with organizations, not engineers, study finds

The study from the University of Manchester has found that ethical issues in AI development often stem from organizational structures rather than individual engineers. While engineers are aware of problems like biased outputs, unfair decisions, and opaque systems, they lack the authority and incentives to address these issues within their organizations.

The research, based on interviews with engineers from various sectors, highlights a "structural capacity" problem: engineers can identify ethical concerns but are unable to implement safeguards due to organizational constraints. These constraints include compliance processes focused on paperwork rather than action, commercial pressures to meet deadlines, and reward structures that prioritize speed over rigor.

This creates what researchers call "compliance theater," where organizations appear ethically committed but do not follow through with ethical practices. The study suggests that current AI governance efforts focus on creating documentation and policies rather than changing the actual development processes. Lead researcher Alessia Vlasceanu emphasizes that training engineers alone won't solve the problem; instead, organizations must alter how they develop AI systems.

The findings challenge the notion that AI ethics is merely an issue of engineers' knowledge or care. Instead, it points to a deeper structural issue where the infrastructure for ethical AI is currently misaligned. The researchers argue for a shift in focus towards how AI projects are managed day-to-day, ensuring that ethical concerns are raised, addressed, and rewarded.

They also call for regulators to move beyond verifying compliance documents and to verify whether ethical safeguards are being followed in practice.