From Black Box to Glass Box: Making AI Transparent and Understandable
At the FAIR General Conference 2025, presented the project “From Black Box to Glass Box” (BigBox), developed by Kode within the FAIR cascade funding scheme of the PNRR.
Rome, 12 December — Yesterday at the FAIR General Conference 2025, Andrea Spinelli and Chiara Baraglia from Kode presented the project “From Black Box to Glass Box” (BigBox), developed by Kode in collaboration with the KDD Lab – University of Pisa, within the FAIR cascade funding scheme of the PNRR – SPOKE 1 Human-Centered AI.
The project’s main objective is the development of FairXAI, a modular and scalable framework for the explainability of AI models.
In a context where complex artificial intelligence models are increasingly used in industrial, healthcare, and financial sectors, transparency and explainability have become essential requirements.
The project aims to provide a modular and scalable framework capable of explaining the behavior of complex AI models—often perceived as “black boxes”—through counterfactual explanations based on logical rules.
With FairXAI, the framework developed within the project, it is possible to obtain clear and verifiable explanations of AI decision-making processes, thanks to the optimization of existing XAI models (particularly LORE and its derivatives), accessible even to non-expert users.
The innovation of FairXAI lies not only in its ability to explain model behavior, but also in its capability to integrate multiple interpretability methods into a single, flexible system, ready for industrial and research applications. The platform stands out for:
- its modular and extensible architecture,
- its multi-domain compatibility, working with tabular data, images, and time series,
- its guided user interface, designed to simplify interactions with complex algorithms and make the system usable even by non-expert operators.
Among the main expected benefits, the project fosters trust and adoption of AI systems in sensitive domains such as healthcare, finance, and industrial maintenance, promoting transparency and the responsible use of artificial intelligence. Moreover, FairXAI contributes to spreading a culture of accessible and interpretable AI, involving operators and decision-makers in the process of understanding model behavior.
The project demonstrates how research and innovation can transform AI’s so-called “black box” into a glass box—accessible, understandable, and ready for safe and reliable use in real decision-making processes.
