Dr. Holger Steinmetz (ZPID): "Prediction vs. Explanation: The Best Of Both Worlds"

Big data and machine learning are experiencing a huge rise in the economy and many scientific fields due to its enormous predictive capabilities. While most data scientists will currently reject early bold claims that big data would be end of theory (and, thus, traditional science), the precise characteristics of the well-known "prediction vs. explanation" dichotomy are not well understood. While modern concepts of "explainable AI" try to overcome past limitations and criticism of "black box-approaches", behavioral scientists have still a different understanding what "explanation" means. The present talk will note three essential topics that helps to understand data and its occurance, namely "causation", "sampling", and "measurement". The talk presents suggestions about potential blind spots with regard to these topics, briefly introduce these topics and present some initial approaches to combine the best of both worlds.