Theory & Proposal

A group of researchers posed a really good question. As AI models become more prevalent in our lives, from language processing to content generation etc, a critical question comes about: Who really owns the data used to train these models?

Well known data privacy regulations like General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA) give user control over their personal data but how many individuals are aware to what extent these act affect them let alone are even aware they’re in place.

Rather than idling by and hoping government bodies will agree to protect them some have proposed a blockchain-based framework called IBIS to tackle this issue. They argue that existing solutions treat datasets and copyrights as mere non-fungible tokens, failing to capture the dynamic nature of iterative model training and fine-tuning processes. IBIS aims to create a robust system for tracking data provenance, lineage, and copyright compliance throughout the AI model lifecycle. I think this idea makes a lot more sense

Findings

Let's break down what IBIS is all about. First off, it's a blockchain-based framework that integrates on-chain registries for datasets, licenses, and models. It also has off-chain signing services to facilitate collaboration between AI model owners (AOs) and copyright owners (COs). It’s a blockchain-based framework that integrates on-chain registries for datasets, licenses, and models, alongside off-chain signing services to facilitate collaboration among AI model owners (AOs) and copyright owners (COs). So here’s the lowdown…