Europe doesn't need to chase the biggest AI models to lead in artificial intelligence; instead, it should focus on building systems that are fair, explainable, and genuinely accountable to the people who use them. That's the argument from Francesco Ferrero, leader of the Flagship Initiative on Artificial Intelligence at LIST (Luxembourg Institute of Science and Technology), who contends that the current race for ever-larger models on ever-larger datasets is economically unsustainable, environmentally costly, and leaves most organizations dependent on tools they cannot control or understand.

What Does "Responsible AI" Actually Mean in Practice?

The term "responsible AI" gets thrown around frequently in tech circles, but Ferrero offers a precise definition with real teeth. According to him, responsible AI must meet four specific criteria:

• Factual Accuracy: The system produces correct information and does not hallucinate or generate false claims.
• Explainability: The AI can explain its decisions to the people affected by those decisions, creating genuine accountability rather than blind authority.
• Fairness: The system does not encode or reproduce bias and does not discriminate against individuals or groups in decisions affecting jobs, credit, or healthcare.
• Efficiency: The system is frugal with computational resources, making it accessible and controllable for organizations beyond the handful of tech giants with massive infrastructure.

Each element addresses a real problem. A system that cannot explain its decisions is not accountable in any meaningful sense; it simply imposes authority. A system that encodes bias will reproduce it reliably at scale, shaping access to jobs, credit, and healthcare in ways that harm vulnerable populations. And a system that requires enormous computational resources to operate forces most organizations to depend on external providers without understanding or controlling what they deploy.

Why Is the "Bigger Is Better" Approach Unsustainable?

The current AI paradigm emphasizes scale: building ever-larger models on ever-larger datasets, requiring ever-greater computing power. The results can be impressive, but the costs are mounting. Only a handful of organizations on Earth have the infrastructure to train frontier models, leaving hospitals, public agencies, and small businesses either unable to access AI or forced to use systems they do not understand or control.

The environmental footprint is also a genuine constraint. Training large-scale AI models consumes enormous amounts of electricity and water, making the approach increasingly difficult to justify as resource scarcity grows. From economic, societal, and environmental perspectives, the bigger-is-better approach is not sustainable.

"The bigger-is-better approach is not sustainable from either the economic, societal and environmental perspectives," explained Francesco Ferrero.

Francesco Ferrero, Leader of the Flagship Initiative on Artificial Intelligence and Head of the Human-Centred AI, Data and Software Research Unit at LIST

How Can Organizations Test Whether Their AI Systems Are Safe and Fair?

LIST has developed practical tools to address the accountability gap. The LIST AI Sandbox allows organizations to stress-test AI systems against real-world conditions before deploying them in consequential decisions affecting healthcare, justice, or financial services. For most organizations, the honest answer to the question "Is our AI system safe, fair, and doing what we think it is?" is that they do not know. The Sandbox changes that by providing independent scrutiny that should be routine but largely is not.

At a larger scale, the Luxembourg AI Factory is designed to reshape how AI gets adopted across the economy, with a particular focus on small and medium-sized enterprises (SMEs) and startups that are most likely to be left behind by the current model. The Factory guides organizations from initial ideas through working deployment, connecting research, tools, and sectoral expertise to turn responsible AI from an abstract principle into something an SME can actually build with.

"The LIST AI Sandbox allows everyone to test the AI systems that they use," stated Francesco Ferrero.

Francesco Ferrero, Leader of the Flagship Initiative on Artificial Intelligence and Head of the Human-Centred AI, Data and Software Research Unit at LIST

Why Does Trust Matter More Than Raw Performance?

AI systems built without explainability, fairness, or accountability mechanisms generate resistance. Not immediately, perhaps, but steadily, from citizens who experience their effects, from regulators asked to oversee them, and from societies that find the terms of the technology were set elsewhere by others for other purposes. The harder and more important work is building AI that earns trust.

Ferrero argues that frugal, human-centered AI is the right approach for Europe because in a world increasingly affected by conflicts and resource scarcity, people need a technology they can embrace rather than one they will resist in the long term. The question is not whether AI can transform healthcare, work, and geopolitics; it is whether it can do so while reinforcing democracy instead of eroding it, making organizations more competitive without making people more vulnerable, and allowing innovation and ethics to travel together.

"Frugal and human-centred AI is the right approach for Europe because in a world that is increasingly affected by conflicts and scarcity of resources, we need a technology that the people can embrace instead of one they will fight in the long term," noted Francesco Ferrero.

Francesco Ferrero, Leader of the Flagship Initiative on Artificial Intelligence and Head of the Human-Centred AI, Data and Software Research Unit at LIST

The narrative that Europe is lagging behind the United States and China in the bigger-is-better AI competition misses the point entirely. Europe's advantage lies not in chasing raw scale but in building AI that is trustworthy by design, efficient with resources, and genuinely oriented around the people who use it. In a world of growing instability and resource pressure, that is a more durable bet than raw computational power.