Today we know enough about artificial intelligence to recognize its enormous potential. But perhaps the most important question for decision-makers is not what AI can do today, but what we do not yet…
Today we know enough about artificial intelligence to recognize its enormous potential. But perhaps the most important question for decision-makers is not what AI can do today, but what we do not yet know about where it is taking us.
As its capabilities accelerate, our responsibility is no longer limited to governing the technology we understand. It extends to managing the uncertainty surrounding its development and future impact.
These are not theoretical concerns. Stanford University’s 2025 AI Index showed that AI systems’ performance on three advanced benchmarks improved significantly within a single year. What these numbers reveal matters more than the numbers themselves: the benchmarks we use today to measure AI capabilities can shift very quickly.
The International AI Safety Report 2026 introduces a telling concept, the “evidence dilemma.” Systems are becoming more capable quickly, while evidence about their risks emerges more slowly and is harder to assess.
If decision-makers act too early, they may lock in ineffective interventions. If they wait for conclusive evidence, they may leave society exposed to potentially serious harm. This is why the unknown begins at the frontier of capability itself.
We do not know how far these systems can go, how fast they will get there, or where their technical and cognitive limits will appear.
The picture is further complicated because capability gains are uneven. Advanced systems may excel at highly complex tasks while still failing at others that seem simpler.
Then comes the question of explanation and accountability: will we still be able to understand the decisions made by increasingly complex and autonomous systems? As AI’s role expands in decisions that affect people and society, transparency and explainability are no longer just technical issues. They become matters of trust and responsibility.
We also do not know whether today’s safeguards will be enough for tomorrow’s systems. The International AI Safety Report points to persistent significant gaps and notes that the real-world effectiveness of many safeguards remains uncertain.
Uncertainty extends to the economy and labor. An IMF study drawing on more than 700 million job postings across 110 economies found a decline of nearly 10% in demand within the jobs most exposed to AI, while also recording rapid growth in demand for AI-related skills.
This shows why the future cannot be reduced to the claim that AI “will eliminate jobs” or “will create jobs.” It may do both. The bigger question is how gains and losses will be distributed, and who will capture the largest share of the new economic value.
The social transformation may run deeper than the labor market. AI is gradually entering education, access to information, decision-making and human communication. In media and journalism in particular, a fundamental question arises: how will AI reshape our relationship with information, truth and trust?
There is another paradox in the environmental domain. AI can help improve energy efficiency and support many environmental applications, yet it also requires ever-growing physical infrastructure.
The International Energy Agency expects global electricity demand from data centers to more than double by 2030, reaching around 945 terawatt-hours, with AI as the most important driver of this increase. The agency also notes significant uncertainties in these estimates.
This raises a question that should be central to the debate: not only how AI can help us achieve sustainability, but how we can make AI itself sustainable.
Then there is the geopolitical dimension. Will this technology spread knowledge and capabilities more widely, or concentrate power among the few who own the compute, data, models and infrastructure?
We have no final answer, but the answer will deeply shape the future of economic power, technological sovereignty and international relations. This brings us to the biggest institutional question: can governance move as fast as technology?
The problem is not that decision-makers lack all the evidence. They must make important decisions while the evidence itself is still evolving. That is why we need governance that can learn, evaluate and adapt continuously, rather than assuming that legislation suited to today’s technology will necessarily suffice for tomorrow’s.
Here I return to an idea I have proposed before: the world urgently needs a specialized United Nations institution dedicated to artificial intelligence. AI crosses borders, and its effects on security, the economy, information, labor, the environment and international relations cannot be treated only as separate national issues.
The mission of such an institution should not be to slow innovation or create a new layer of international bureaucracy. It should be to build a shared global foundation for knowledge, assessment and governance, helping countries track the evolution of capabilities and risks, exchange scientific knowledge, align principles and standards, and strengthen cooperation on challenges that do not respect national borders.
The need for such an institution becomes more pressing because of what we do not know. If future capabilities and risks cannot be fully predicted, the world needs an international mechanism to help it prepare for them. What is needed is not an institution that claims to know the future of AI, but one that helps the international community manage uncertainty about that future.
Beyond all this remain “the unknown unknowns.” Capabilities, risks and opportunities may emerge that we have not even thought to ask about, because AI may create new patterns of relationships and interdependence that humanity has never experienced. This should not lead us to fear technology or halt innovation.
Rather, it should push us toward institutional humility and proactive preparedness. Hence the need for a broader vision captured in one principle: “AI for Security, Sustainability and Peace.”
Security, to protect people and society from evolving risks. Sustainability, to ensure technological progress does not come at the expense of its environmental and social foundations. Peace, so that this new power becomes a means of cooperation and shared prosperity, not a new source of conflict and division.
International governance is necessary not only because of the risks we know today, but also because of uncertainty about what AI may become tomorrow.
If AI is a global force that crosses borders, its governance can hardly remain a purely national matter. We have spent a long time asking: what will AI be able to do?
Perhaps it is time to add a more urgent question:
Do we have the national and international institutions capable of dealing with what we know about it today, and preparing for what we do not yet know?



