Complying with the EU AI Act: Innovations in Explainable and User-Centric Hand Gesture Recognition

Transparency matters for real-world AI: The paper demonstrates that explainable AI methods can make hand gesture recognition systems more transparent by identifying when the system encounters unfamiliar or anomalous gestures—a key requirement under the EU AI Act for high-risk AI applications.

Real-world data reveals hidden problems: By expanding their dataset to include 24,000 out-of-distribution gestures (gestures the system hadn't seen before) collected across multiple users and locations, researchers found that systems perform differently in real conditions than in controlled settings, highlighting why diverse, real-world testing is essential for responsible AI deployment.

Personalization improves fairness and usability: The system achieved at least a 15.17% improvement in user adaptability through transfer learning techniques, suggesting that AI systems designed to adapt to individual users may be more equitable and user-friendly than one-size-fits-all approaches.

Detecting system uncertainty is crucial: The integration of anomaly detection improved the system's ability to identify when it was uncertain about a gesture by 11.50%, with 97.5% accuracy in characterizing these uncertain moments—demonstrating that responsible AI systems should acknowledge their limitations rather than making confident wrong predictions.

Regulatory compliance can drive better technology: This work shows that meeting EU AI Act requirements for transparency and robustness doesn't just satisfy legal obligations; it can lead to technically superior systems that are more trustworthy and reliable in real-world use.