Addressing Data Sovereignty and Empowering Users by Promoting Trustworthy Low Resource AI Systems

Abstract

In the last few years, advancements in artificial intelligence (AI) have dramatically transformed thedigital world, with AI tools being integrated across a multitude of industries. The widespread adoption of Large Language Models (LLMs) has led to numerous benefits, such as improved data analysis, customer support, and plain language explanations. However, the proliferation of LLMs in digital services has also raised concerns related to cost, environmental impact, privacy, and algorithmic fairness. This research explores if a locally trained and run low-rank adaptations (LoRAs) can enable community-based organizations to create AI tools that can fine tune LLMs and address their specific needs while mitigating concerns around privacy, algorithmic fairness, cost, and environmental impact. Furthermore, this research provides guidelines for low-resource organizations to adopt this AI tool on local hardware.