You emphasize translating complex frameworks like the AI Act and GDPR into practical guidance. In your experience, what are the most common misconceptions teams have about “ethical AI,” and how can they move from theory to actionable implementation?
One of the most persistent misconceptions I encounter is that “ethical AI” is either a purely theoretical discussion or a compliance exercise that happens at the end of a project. In many EU-funded projects, teams initially perceive ethical and legal frameworks as abstract, legalistic, and somewhat disconnected from the realities of development.
In practice, ethical AI is much more operational. It is about how decisions are made throughout the lifecycle of a system – how data is sourced and governed, how models are designed and evaluated, and how outcomes are communicated and monitored.
Another common misunderstanding is the assumption that all ethical principles must be applied equally and exhaustively in every case. This often leads to paralysis or unnecessary complexity. What proves far more effective is prioritisation. Different use cases carry different levels and types of risk, and ethical efforts should reflect that. When teams identify which principles are most critical in a given context and define how success will be measured, ethical AI becomes actionable rather than aspirational.
The real shift from theory to practice happens when teams stop treating ethics as a label and start treating it as a set of design and risk management decisions embedded in everyday work.
Having worked closely with EU-funded research consortia, how can projects effectively integrate ethics throughout the entire AI lifecycle—without slowing down innovation or creating unnecessary bureaucracy?
The fear that ethics automatically means bureaucracy is, in my view, usually a design failure rather than something inevitable. From my experience working with EU projects, the key lies in proportionality and timing. At the early stage, it is enough to carry out a lightweight risk screening to identify potentially high-risk use cases. During the design phase, ethical and legal requirements, such as explainability or human oversight, should be directly embedded into system specifications. In development, instead of relying on one large audit at the end, teams should apply continuous checks. Finally, at the deployment and monitoring stage, it is essential to introduce feedback loops, incident reporting mechanisms, and ongoing tracking of both model performance and its real-world impact.
What works best in practice is aligning ethical considerations with existing processes, such as quality assurance or risk management, rather than creating parallel structures. Using templates and repeatable workflows also helps avoid unnecessary complexity. Most importantly, teams should focus on key decision points rather than producing excessive documentation. Well-designed ethics governance does not slow projects down. On the contrary, it often accelerates them by reducing rework, avoiding regulatory surprises, and eliminating last-minute, reactive compliance efforts.
From your hands-on role as an Ethics Advisor and Mentor, what early warning signs indicate that an AI project may face ethical or compliance risks, and what simple steps can teams take to address them proactively?
There are several early signals that suggest an AI project may face ethical or compliance challenges. One of the most common is a lack of clarity around data. If teams cannot clearly explain where their data comes from, how it is processed, or whether it involves personal or sensitive information, this often indicates potential risks related to privacy, governance, and later regulatory compliance.
Another warning sign is when explainability and transparency are treated as afterthoughts. If these aspects are not considered during the design phase, they tend to be difficult and costly to address later, and the resulting system may struggle to build trust with users or meet regulatory expectations.
A further issue arises when responsibility is diffused. If no one within the team has clear ownership of ethical and compliance aspects, gaps tend to emerge, particularly in complex, multi-partner environments.
Addressing these risks does not require heavy processes. What makes the biggest difference is introducing early visibility and simple, iterative checks. Teams benefit from identifying key risks at the outset, clarifying roles and responsibilities, and revisiting these aspects regularly as the system evolves. When ethics is approached as an ongoing, lightweight process rather than a formal obligation, it becomes much easier to manage proactively.
Ultimately, the most effective approach is not to eliminate all risks upfront, but to ensure that they are recognised early, prioritised appropriately, and continuously managed throughout the project lifecycle.