On February 6, 2025, Institut Polytechnique de Paris hosted the IA Action Summit at École polytechnique, a high-level event bringing together leading researchers, industry experts, and policymakers to discuss the transformative role of artificial intelligence. Among the distinguished participants, Anna Simoni and Vianney Perchet, researchers at CREST, presented their latest work on AI applications in macroeconomic forecasting and algorithmic optimization, respectively.
Macroeconomic Nowcasting with AI and Alternative Data – Anna Simoni
Anna Simoni (CNRS, ENSAE Paris, École polytechnique, Hi!Paris Fellow), presented her work on integrating artificial intelligence and alternative data to improve macroeconomic nowcasting. She addressed the challenge of forecasting key economic indicators such as GDP growth, inflation, and financial cycles in real time—an essential task for policymakers, financial institutions, and businesses.
Her research highlights the potential of AI to handle large, complex, and mixed-frequency datasets from both traditional (e.g., national statistics, financial markets) and non-traditional sources (e.g., Google search data, satellite imagery, mobile phone traffic, credit card transactions). By applying machine learning techniques, including dynamic factor models, neural networks, and Bayesian methods, her work demonstrates how AI-driven models can enhance nowcasting accuracy, detect economic shifts early, and provide better insights into economic trends.
Through real-world applications, including an analysis of Google search data for GDP nowcasting, her findings reveal that AI methods can significantly improve short-term economic predictions, especially in periods of high uncertainty or economic recessions. However, she also emphasized that AI models must be carefully designed to ensure interpretability and reliability, as pure data-driven approaches do not always outperform traditional methods.
AI-Powered AI: Enhancing Algorithmic Decision-Making – Vianney Perchet
Vianney Perchet, (GENES, Hi!Paris Fellow), introduced the concept of AI-powered AI, where artificial intelligence assists and improves algorithmic decision-making in various fields, from optimization to reinforcement learning and recommender systems.
He explored how AI-oracles—AI systems that provide tentative solutions to computational tasks—can be leveraged to accelerate decision-making processes. Through examples like binary search with AI hints, he demonstrated the trade-offs between consistency and robustness: algorithms must be able to trust AI-generated predictions when they are accurate but remain resilient when faced with imperfect or misleading information.
Perchet highlighted several fundamental challenges in AI-assisted algorithms, including:
- Strategic AI usage: AI-oracle calls can be costly, requiring efficient selection mechanisms.
- Combination of multiple AI sources: Different AI models may offer varying predictions, requiring aggregation methods to improve reliability.
- Incentivization: AI models may have their own objectives, necessitating mechanisms to align them with user needs.
- Learning from structure: AI-assisted algorithms must adapt and refine predictions over time for better long-term performance.
- Beyond worst-case scenarios: Moving beyond rigid worst-case analyses toward realistic, instance-dependent performance guarantees.
His research opens exciting new perspectives for integrating AI into optimization problems, online learning, and algorithmic decision-making, with applications ranging from robotics and computer vision to financial forecasting and recommender systems.
CREST’s Commitment to AI Research
CREST’s involvement in the IP Paris IA Action Summit, through the contributions of Anna Simoni and Vianney Perchet,demonstrates the lab’s strong engagement in artificial intelligence research applied to economics, forecasting, and decision-making. Their work aligns with ongoing research at CREST on machine learning, reinforcement learning, and statistical AI methods, led by scholars such as Nicolas Chopin, Paola Tubaro, Bruno Crépon, and Olivier Gossner.
As AI continues to transform economic modeling and optimization, CREST continues its research of exploring how AI can enhance forecasting, decision processes, and strategic learning in uncertain environments.