Learning Causal Structures from Data

Authors:

  • Johan Schubert
  • Ronnie Johansson

Publish date: 2020-02-07

Report number: FOI-R--4882--SE

Pages: 18

Written in: English

Keywords:

  • artificial intelligence
  • machine learning
  • causal learning
  • causal discovery
  • decision support
  • common operational picture
  • military

Abstract

In this report we perform a horizon scanning over the research field of causal learning and discovery within artificial intelligence (AI). The motivation for investigating this field of research is to find out how important causality is in military decision-making. The area of machine learning within AI has seen great progress in recent years. However, there is an inherent limitation in the machine learning methods that focus on finding correlations in data. If correlation between two events is detected, we do not know if either one has caused the other or whether a third event is causing both. By also giving algorithms an idea of causality, it becomes possible to better understand and reason about the outside world. In order to allow the use of machine learning in military systems that normally only identify correlations between events and phenomena, one must know all possible causal relationships in advance. This requires a very high understanding of all possible events and phenomena. If this knowledge is not available, methods for learning causal relationships are necessary.