Exploring the Freedoms in Data Mining: Why the Trustworthiness and Integrity of the Findings are the Casualties, and How to Resolve These?

Abstract

This work aims to improve the accuracy of data-mining. A problem with today’s system for Artificial Intelligence (AI) is the reliance on frameworks that lacks oversight: AI is trained on data that is not guaranteed to be representative and is constructed through software-APIs maintained by a large number of developers, each with their interests. In contrast, the society expects the systems to work, as the result is otherwise a zombie-like behavior of systems for industrial control of how new drugs are discovered, oversight of public governance, heating systems, etc. This paper aims at tackling this issue. The idea is to relate the freedoms in software and algorithms, thereby identifying the blind spots in AI. The results reveal how knowledge discovery is directly linked to hidden attributes (in software and algorithms). Thus, this work provides users with a recipe for improving the trust in their AI-predictions.