L'IA peut-elle prédire que toutes les civilisations humaines futures s'effondreront dans les 50 prochaines années ?

Si l'IA peut prévoir les trajectoires sociétales avec une grande confiance, elle pourrait redéfinir les politiques mondiales, la préparation aux urgences et même les cadres éthiques autour de la survie humaine. Ce pouvoir donnerait à l'IA l'autorité sur la planification à long terme, rendant potentiellement le destin de l'humanité dépendant de la prévoyance algorithmique.

Background

State-of-the-art forecasting systems model long-term societal trajectories by integrating climate, demographic and technological trends, yet they remain bounded by deep uncertainty in human agency and rare discontinuities such as war or pandemics. Empirical evaluations show that these models can reproduce past collapses under specific parameter regimes, but they fail to converge on a consensus when asked to project far-future singular events across every plausible civilization pathway. Named systems like the International Futures model and World3 variants illustrate how structural assumptions—rather than empirical calibration—dominate long-range outcomes, and even their authors caution against treating collapse as an inevitable forecast. A notable counterexample is the post-2020 rebound in global fertility and energy intensity gains, which recent projections did not anticipate, underscoring the sensitivity to unforeseen innovations and policy shifts. Thus, while the literature documents conditions under which collapse becomes probable within stylized frameworks, it offers no validated model that excludes the possibility of sustained civilization trajectories beyond 50 years.

SOURCE: no public reference