The real challenge of artificial intelligence lies beyond the singularity

The technological singularity describes the hypothetical point at which AI does not only reach or surpass human intelligence, but rather accelerates its capabilities so rapidly that it essentially starts programming itself – without humans being able to keep track. Technological advancement would then pick up pace rapidly and would no longer remain manageable. This scenario has been discussed for decades. At the moment, I perceive it as a long-term possibility, not an imminent reality.

As CEO of one of the leading AI companies, Altman is not only a researcher but also a businessman. His statements therefore do not merely describe a technological vision, but also shape expectations, attract investment and underscore the strategic significance of his own technology – especially ahead of a planned stock market listing. What we are experiencing is more akin to an economic upheaval: Even today, AI is changing knowledge work radically by writing software, supporting research, analyzing data and developing drugs. 

In many tasks, it is faster and more productive than humans. But humans remain the innovation engine. The pattern stays unchanged: Humans develop better AI systems, which then support humans in creating the next generation. We are still far from an AI that independently builds increasingly powerful AI.

The real challenges do not lie in computational power or larger models either, but rather in missing capabilities: continuous learning, robust world understanding, causal thinking and the lasting integration of experiences. Even more important, however, is the embedding of AI into human and societal contexts.

Nowadays, AI systems are able to talk about ethics, name risks or summarise moral arguments. What they do not experience, however, are the consequences of their actions. They bear no responsibility, feel no empathy and develop no independent understanding of decisions and their effects on humans. Their assessments rely on statistical patterns and prescribed goals.

The more autonomous AI becomes, the more important its ability to weigh its own actions against their consequences for humans, society and the environment becomes. An agent built for efficiency can make excellent local decisions and yet promote societally problematic developments. Economic efficiency, legal compliance, fairness and sustainability are not identical goals – between them, conflicts inevitably arise.

This is why it is not enough to build increasingly powerful models. The actual task consists in embedding AI into a system that takes physical consequences, economic effects, legal frameworks, societal values and long-term impacts into account simultaneously. To achieve this, AI will not be able to do without a combination of world models, knowledge graphs, physical knowledge and explicit representations of societal norms. Yet technological solutions are not sufficient. They require a regulatory framework that reliably secures transparency, responsibility and human controle. This is where Europe comes in with its EU AI Act, which has been creating clear rules for the development and the use of AI since August 2026.

Even more fundamental is the question of whether it makes sense to let AI decide in an unbiased, objective and rational manner. Human societies function remarkably well precisely because individuals do not think identically. Companies see opportunities, researchers uncertainties, legal experts risks, artists new perspectives, politicians different interests. This diversity leads to conflicts and slower decision-making – but at the same time protects against collective maldevelopments. 

The choice of goals is not a technical, but rather a societal and political decision. Emotions are not a disruption of rational action, but rather condensed experiences that expand our judgement. Fear draws attention to risks, empathy protects the vulnerable, trust enables cooperation, curiosity drives innovation. An intelligence optimised solely for efficiency could lose precicely those attributes that make human societies stable and adaptable in the long term.

Perhaps AI should not even try to deliver the one optimal answer. More valuable would be an AI that makes different perspectives visible, presents goal conflicts transparently, quantifies uncertainties and simulates potential consequences. The actual decision would then consciously remain with humans or democratically legitimised institutions.

Accordingly, the question for the coming decades is whether AI will become more intelligent than humans. Depending on the perspective, it already is in many respects – or soon will be. The real challenge, however, lies in embedding this intelligence into our society in a way that strengthens human values, preserves diversity and supports responsible actions.

This is, quite possibly, where the difference between intelligence and wisdom lies. Intelligence finds the best solution for a given goal. Wisdom first questions whether this goal is indeed the right one, what unintended consequences could arise and how different interests can be brought into a viable balance. The future of AI will depend less on how intelligent machines become, but rather on whether we succeed in giving them sufficient orientation, feedback and societal embedding. For it is not intelligence alone that decides the progress of a civilisation, but rather the way in which it is used responsibly.