L'IA peut-elle désormais prédire le futur ?

L'IA peut-elle désormais prédire le futur ?

Can AI now predict the future?

Title history
  • L'IA peut-elle désormais prédire le futur ? — observed on 20261025-1537 UTC (YouTube)

by Les Revues du Monde · View the channel on PSI

Humanities, Social Sciences & Thought AI — News & Applications UYComputer scienceUYQArtificial intelligence (AI)
🎙 Les Revues du Monde 👥 1.2M 📅 October 25, 2026 ⏱ 31 min 👁 2K 📄 documentary 🧭 2026-10-25
Available in: English (current) Français

Keywords

AIpredictioncliodynamicsconflictmodeling

Summary

The video explores the possibility of predicting historical events using mathematical models and AI. Host Charlie Danger collaborates with data scientist Mohamed to test whether machine learning can forecast political conflicts. They explain weather prediction models like Météo-France’s Arpège as a reference, then adapt the approach to social systems. Using the XGBoost model trained on historical conflict data from the VIEWS project and UCDP, they attempt to predict outbreaks in Ukraine and Lebanon. Results show that while the model correctly predicts ongoing conflicts, it fails to anticipate new ones, such as the Lebanon war, with a two-month delay. They compare their results with the prestigious VIEWS system and find that even specialized models struggle. The video then discusses Peter Turchin’s cliodynamics, which uses massive historical datasets to identify long-term cycles and structural-demographic theory, leading to his 2010 prediction of US instability around 2020, which seemingly came true. However, the black swan event of COVID-19 illustrates the unpredictability of rare events. The episode emphasizes the challenges of data quality, model assumptions, and the risk of overfitting, concluding that while history can inform trends, exact prediction remains elusive.

186 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial value by offering a concrete, hands-on test of predictive models on real historical data, rather than just theoretical discussion. The argumentation is solid: the creators transparently discuss their methodology, data sources, and the arbitrary choices involved in feature selection. They also critically compare their results with a state-of-the-art system (VIEWS), demonstrating humility and scientific rigor. The inclusion of historical context, such as Turchin’s predictions and the Seshat databank, adds depth, while the discussion of black swan events underscores the inherent limits of forecasting. The narrative is coherent and evidence-based, with each claim grounded in either their own experiments or cited literature.

Scientific Rigor, Source Quality, Title Accuracy

The video demonstrates strong scientific rigor by using peer-reviewed concepts (cliodynamics, VIEWS, UCDP) and providing a detailed Google Doc with sources. The creators acknowledge the problems of historical data quality, such as propaganda in ancient sources, and the risk of overfitting. The title accurately reflects the content, which investigates whether AI can predict the future through historical modeling. Despite the challenges, the video maintains a balanced perspective, neither overpromising nor dismissing the potential of such models. The integration of expert interviews and original experiments enhances credibility.

206 words

Title / Content Match

The title's question is directly addressed through experiments and expert interviews, making it a good match.

Quality & Reliability

8/10

The video presents a rigorous, self-aware investigation, including original experiments with a data scientist, and references to established academic work, while clearly acknowledging limitations and data uncertainties.

Chapters

Cited Sources

External References

Contribution & Novelties

The video contributes an original experiment where the creators train their own XGBoost model on conflict data, testing it on Ukraine and Lebanon, and openly compare its performance with the VIEWS system. This hands-on approach offers a tangible illustration of the challenges in predictive modeling for social phenomena. It also synthesizes existing research, such as Turchin’s cliodynamics, making complex ideas accessible to a broad audience. The honest discussion of model limitations and data biases adds educational value.

Pour aller plus loin :

  • Cliodynamics — The interdisciplinary field that applies mathematical models to historical dynamics, central to the video’s discussion.
  • Black swan theory — Nassim Taleb’s concept of unpredictable rare events with massive impact, directly relevant to the COVID-19 example.
  • Uppsala Conflict Data Program — The UCDP database used in the video for conflict data.
  • Seshat: Global History Databank — A large-scale historical database referenced through Turchin’s work, illustrating the data collection effort.

152 words

Radar Profile

The radar profile would show high scores in information quantity and quality, a moderate technical level, and slightly lower reliability due to acknowledged limitations. This reflects a well-researched but cautious exploration of a complex topic.

Reliability 7/10