Why Ethical Questions Vanish When AI Works

AI is increasingly embedded across business and society, delivering economic, environmental, and social benefits – and often a strong competitive advantage for organizations that adopt it. Because AI outputs come from machines, we are easily led to assume they are objective, or at least more neutral than human judgment.

But AI systems do not operate in a vacuum.

They are designed, trained, deployed, and overseen by humans, within specific cultural, organizational, and societal contexts. This makes AI a socio‑technical system. AI designers, developers, and deployers do not only take “technical” decisions about datasets, algorithms, and system architectures; throughout the AI lifecycle, a series of human decisions are made –about data selection, design choices, objectives, success criteria, deployment, and use. Each of these decisions embeds ethical assumptions and priorities alongside technical considerations, often implicitly. When these assumptions are well aligned, AI appears to “just work” –and that is precisely when ethics becomes invisible. This may help explain why fewer than 30% of academic studies discuss the ethical and regulatory dimensions of AI.

Yet many things can go wrong when ethical principles are missing. Poor data quality or limited representativeness may lead to biased or inaccurate outputs, raising concerns related to fairness and non-discrimination. Limited explainability can undermine accountability by making it difficult to understand, challenge, or audit AI outputs and the resulting decisions. Weak anonymization or excessive data retention can compromise privacy and data protection. In each case, the absence of explicit ethical principles – such as fairness, transparency, accountability, and human oversight – translates into tangible operational and compliance risks.

In engineering contexts, these risks are not merely abstract; they can affect safety margins, regulatory compliance, and system reliability. For instance, in energy systems, AI-driven load forecasting models trained on historical consumption patterns may fail to account for the behavior of marginalized populations or atypical usage. This can potentially cause inequitable energy distribution or distorted pricing signals.

If these limitations in representativeness of data are not explicitly addressed, the resulting “optimized” designs may appear sustainable while embedding hidden trade-offs. Moreover, when decisions are taken based on such models, without sufficient transparency and accountability, those decisions become difficult to validate or contest.

This is where governance and regulation play a critical role. Some approaches focus on identifying and mitigating risks through self‑assessment, while others aim to embed ethical principles – such as accountability, responsibility, and human oversight – throughout the entire AI lifecycle. The EU AI Act reflects this latter approach, promoting human‑centric and trustworthy AI while supporting innovation.

Still, governance is often perceived as something that slows down innovation and creativity. Yet innovation built without ethical foundation often struggles to earn trust from users, clients, and society. In practice, the absence of governance creates uncertainty, not freedom. Clear ethical boundaries and accountability structures are what allow innovation to scale responsibly, without relying on trial‑and‑error at the expense of trust.

Is governance an obstacle to AI innovation – or the condition that makes it possible?

This article was written by Katherine Sáez Villanueva during her internship with the IPU team, where she was involved in Information Security Governance, Risk Management & Compliance (GRC).

If you have questions or want to know more about integrating in sustainable innovation, feel free to reach out to info@ipu.dk.