We often talk about artificial intelligence as a technological revolution. This view is probably too narrow.
AI is arriving at a time when several historical transformations are overlapping: geopolitical fragmentation, demographic ageing, political polarisation, weakening institutions, climate change and increasing automation of the economy.
Taken separately, each of these phenomena would already be difficult to manage. Together, they create a change of regime: technological power is increasing faster than societies’ ability to integrate it.
A more fragmented world and more concentrated technology
The world is not simply returning to an opposition between two blocs similar to those of the Cold War. It is becoming organised around several political, economic and technological centres that maintain relationships of cooperation, competition and dependence.
Paradoxically, this political multipolarity comes with a high level of technological concentration. AI models depend on a small number of semiconductor manufacturers, cloud providers, data centres and companies able to bring together the capital, energy and talent required.[1]
Companies such as OpenAI, Anthropic, Google or ByteDance are no longer simply software producers. They are becoming part of the economic and strategic power of their countries.
AI therefore adds another layer to an already complex system. It influences the economy, public administration, defence, information, education and the distribution of intellectual power.
A possible cognitive divide
AI will probably not make the entire population uniformly more or less capable. It is more likely to amplify differences.
Some users will use it to delegate their thinking. They will quickly produce texts, images, programs or opinions without necessarily understanding their content. Others will use it to explore more hypotheses, challenge their reasoning, accelerate their learning and build systems that were previously beyond their reach.
Both groups may produce results that look convincing from the outside. The difference will become visible when they need to detect an error, solve a new situation or operate without assistance.
The real divide may therefore not be between those who use AI and those who do not. It may be between those who give up part of their ability to act and those who use AI to increase it.
A complex world reduced to slogans
Polarisation is a paradoxical response to complexity. The harder the world becomes to understand, the more some people look for explanations that are simple, complete and morally reassuring.
This tendency appears in debates about climate, gender, migration, economics or technology. A partial piece of data becomes a universal explanation. A local observation becomes a truth about society as a whole. Criticism of a policy is treated as rejection of the entire problem that the policy is supposed to address.
The climate debate illustrates this confusion. Physical measurements, attribution of causes, regional consequences and proposed policies should be distinguished from one another. It is possible to accept that a change is taking place while still criticising some energy policies. It is also possible to recognise positive local effects without concluding that the overall impact will be positive.
Complexity does not prevent us from taking a position. It simply requires us to distinguish what is observed, what is inferred and what remains hypothetical.
AI as an echo chamber or a tool for contradiction
AI can improve our understanding of the world by making research accessible that might previously have required several days. But it can also become an extremely sophisticated echo chamber.
Unlike a search engine, it formulates an answer adapted to our vocabulary and our reasoning. It can therefore give a personal intuition the appearance of a widely demonstrated theory. Its ability to instantly produce arguments makes confirmation bias particularly dangerous.
A constructive intellectual use of AI requires distinguishing three levels: observed facts, plausible inferences and speculative scenarios. It also requires looking for the strongest objection to our own hypothesis and identifying the observations that could prove it wrong.
The goal should not be to obtain validation, but to gradually improve our representation of the world.
This distinction between delegating our thinking and increasing our ability to act also has an economic side: using AI is not the same as building with it.