Data Foundation Through PRESS

This chapter critically examines the pivotal role of data quality in AI system performance, highlighting how suboptimal data foundations precipitate failures across traditional machine learning, generative AI, and specialized models. Through case studies, such as Unity Technologies’ $110 million loss and T-Mobile's $500 million breach, it demonstrates that data-related issues are often misattributed to algorithmic failures. The PRESS framework (People, Responsible, Explainable, Safe, and Sustainable) is introduced to diagnose and address data challenges, with low scores exacerbating issues. The chapter advocates for a data-first approach, emphasizing traceability, freshness, and ownership to transform data from a liability to an asset. It explores data trustworthiness, transparency, and ethical considerations, stressing the importance of reliable data for AI training and the paradox of data overload. Trust, transparency, and fairness are identified as cornerstones for ethical AI practices. The chapter presents the explainable principle, which tailors explanations to different audiences, and the safe principle, which addresses core data quality issues like missing or biased data. Vendor data management is discussed, with a focus on automated validation and legal implications.