Building Trustworthy AI Systems: AI Inference Verification with Blockchain and Zero-Knowledge Proofs

Abstract

In recent years, the use of deep learning models in sensitive applications increased exponentially. There is a strong need of having a mechanisms for a transparent and secure inference verification. To this end, we propose a system leveraging Zero-Knowledge Proofs (ZKPs) and Blockchain technologies to ensure the validity of model inferences without revealing neither input data nor model details.In this paper, we propose a system that is capable of making non-interactive proofs that are verified on a Blockchain thus creating a trustless environment between the prover and the verifier. The solution is based on the Easy Zero-Knowledge Inference (EZKL) [1] library and leverages ZK-SNARK [2] proofs. We provide a detailed descriptions of the system’s architecture, the implementation as well as the benefits of this approach in enhancing transparency and security in Artificial Intelligence (AI) applications.