A Fair and Trustworthy Remuneration Framework for AI Model Training Using DLT

Abstract

Currently, artificial intelligence (AI) models – particularly those of, but not limited to, Large Language Models – are trained over large amounts of data. Training often happens with little consideration, and even less remuneration, of input data that is content protected by Intellectual Property (IP) rights (e.g., copyrights). The recent rise in sophistication and popularity of generative AI models has further highlighted this issue, as traditional IP licensing models remain largely inadequate. In this paper, we present a proof of concept for an automated, fair, and trustworthy remuneration system for AI model training data contributors leveraging Distributed Ledger Technology. We propose the use of attribution methods for rewarding the most relevant sources for any given request, and smart contracts for the enforcement of the mutually beneficial revenue-sharing agreements between the model creator and training data copyright holders.