Trust the Process: Zero-Knowledge Machine Learning to Enhance Trust in Generative AI Interactions

TL;DR

This paper proposes Zero-Knowledge Machine Learning (ZKML) using cryptographic proofs to verify AI outputs, enhancing transparency, fairness, and privacy in generative AI applications like transformers.

Contribution

It introduces snarkGPT, a practical ZKML implementation for transformers, and provides empirical analysis of its scalability and performance in ensuring fairness and trust.

Findings

snarkGPT effectively verifies transformer outputs without revealing model details

ZKML improves transparency and fairness in generative AI models

Empirical results demonstrate the feasibility of ZKML in real-world scenarios

Abstract

Generative AI, exemplified by models like transformers, has opened up new possibilities in various domains but also raised concerns about fairness, transparency and reliability, especially in fields like medicine and law. This paper emphasizes the urgency of ensuring fairness and quality in these domains through generative AI. It explores using cryptographic techniques, particularly Zero-Knowledge Proofs (ZKPs), to address concerns regarding performance fairness and accuracy while protecting model privacy. Applying ZKPs to Machine Learning models, known as ZKML (Zero-Knowledge Machine Learning), enables independent validation of AI-generated content without revealing sensitive model information, promoting transparency and trust. ZKML enhances AI fairness by providing cryptographic audit trails for model predictions and ensuring uniform performance across users. We introduce snarkGPT, a practical ZKML implementation for transformers, to empower users to verify output accuracy and quality while preserving model privacy. We present a series of empirical results studying snarkGPT's scalability and performance to assess the feasibility and challenges of adopting a ZKML-powered approach to capture quality and performance fairness problems in generative AI models.