Algorithm Accountability, Algorithm Literacy and the hidden assumptions from algorithms

Abstract

Our societies are facing problems that are more and more complex so that decision making is often helped or even delegated to algorithms. Algorithmic decision making (ADM) processes are complex socio-technical systems which interact with society on a large scale. Credit scoring, automatic job candidate selection, predictive policing, or recidivism risk assessment are examples, among others, of already used ADM systems. In this talk, I will start with an overview of what is so far understood as Algorithm Accountability and Algorithm Literacy. I will then focus on algorithms that carry with them modeling assumptions (e.g., machine learning, data-mining algorithms...) and show what effects this has on the interpretation of the algorithms’ results and how we could, from a software engineering point of view, bring more explainability.