A Hybrid-DLT Based Trustworthy AI Framework
Andrea Pelosi, Claudio Felicioli, Andrea Canciani, Fábio Eduardo Severino · 2023
While Artificial Intelligence (AI) is making significant strides in a variety of sectors, an exclusive focus on accuracy can overlook the critical aspect of trustworthiness, especially in contexts where it should be a primary concern. In this paper, we propose a novel framework for the development of trustworthy AI systems, leveraging Hybrid Distributed Ledger Technology (Hybrid DLT). We explore the concept of shifting from an accuracy-based paradigm to an approach where trustworthiness is an integral part of the design. Our framework facilitates collaboration between different entities across the data preparation, model training, and the classification phase of a supervised learning ML solution. It uses a shared ledger which offers a tamper-resistant audit log of every operation, ensuring non-repudiation and replicability. We discuss how employing our proposed framework leads to significantly enhanced trustworthiness in AI systems.