Causality in Machine Learning

The goal of machine learning is to find structure and correlations in data. However, correlations are not considered as causal relationships. Found relationships can for instance also be based on noise, too selective data or confounders. Based on the notion of causality defined by Jonas Peters, Dominik Janzing and Bernhard Schölkopf, we asked the question in the Machine Intelligence Seminar 2020 which influence and significance causality has in machine learning. The result is this collection of trend reports, which explores the meaning and importance of causality in several data science applications. Each report aims to gain an understanding of the current research on the influence of causality in a specific research area. The investigated research areas range from ethical and medical applications, human decision processes and human psychology, over law and politics, up to artificial general intelligence. For each area, trends influenced or driven by causality and their key drivers are identified. The trends are then projected into the future, revealing opportunities and threats as well as their impact on technology development and on our society.     «

The goal of machine learning is to find structure and correlations in data. However, correlations are not considered as causal relationships. Found relationships can for instance also be based on noise, too selective data or confounders. Based on the notion of causality defined by Jonas Peters, Dominik Janzing and Bernhard Schölkopf, we asked the question in the Machine Intelligence Seminar 2020 which influence and significance causality has in machine learning. The result is this collection of...     »