- Bring your own headphones for some conferences.
- Bring your water bottle, as water points will be available at the event.
- Printing your access badge is not mandatory to access the event. You will be asked to present the QR code at the entrance to the event and the different conference rooms.
- Access to the event is strictly reserved for accredited individuals. Each participant must present their individual badge. In order to verify identity, proof of identity (card or digital copy) may be requested at the entrance.
AI and Truth: from the fantasy of 100% reliability to the era of calibrated trust
Sep 23, 2025 — 11:45 am CEST - 12:15 PM CEST
SALON INFINI
You need to register and log in to access this functionality
Already registered? Log in now to personalize your experience!
Description
Can we really expect artificial intelligence to deliver absolute truth? Behind the promise of objectivity lie biases in training data, technical limitations of models, and the specter of hallucinations. Current approaches, such as CoT and RAG with RLHF, are making progress but remain insufficient to guarantee total reliability. The challenge is therefore no longer to aim for infallible AI, but to learn how to calibrate the trust we place in it. Between human supervision and user awareness, this discussion will explore how to build a pragmatic relationship of trust with AI.
Resume
L'utilisation de l'intelligence artificielle soulève des questions sur la confiance en ses résultats. La qualité des données d'apprentissage est cruciale pour réduire les risques de biais et d'erreurs. Différentes approches, telles que le RAG et le Chain of Thought, visent à améliorer la fiabilité de l'IA. Il est essentiel de calibrer la confiance en fonction de la vérifiabilité des résultats et des conséquences des erreurs. Les entreprises doivent être conscientes des risques éthiques et de dépendance liés à l'IA. Une supervision humaine et une gouvernance adaptée sont recommandées pour garantir une utilisation responsable de l'IA.
Ce résumé a été généré par une IA.



