A ground level causal learning algorithm

Abstract

Open domain causal learning involves learning and establishing causal connections between events directly from sensory experiences. It has been established in psychology that this often requires background knowledge. However, background knowledge has to be built from first experiences, which we term ground level causal learning, which basically involves observing temporal correlations. Subsequent knowledge level causal learning can then be based on this ground level causal knowledge. The causal connections between events, such as between lightning and thunder, are often hard to discern based on simple temporal correlations because there might be noise - e.g., wind, headlights, sounds of vehicles, etc. - that intervene between lightning and thunder. In this paper, we adopt the position that causal learning is inductive and pragmatic, and causal connections exist on a scale of graded strength. We describe a method that is able to filter away noise in the environment to obtain likely causal connections between events.