Causality, Probability, and Time

Cause and effect come up in nearly every area of life, whether someone is tracing a link between diet and illness or trying to account for a sudden drop in the stock market. Philosophers have debated causation for centuries and computer scientists have studied it for decades, yet getting software to infer and explain causes on its own is still unsolved.

Most earlier efforts have paid little attention to when a cause takes effect or how tangled its relationships can be, even though both shape how well we can predict outcomes, explain events, and decide what to change. The problem is becoming more urgent as large collections of observational data, such as patient records and social network activity, grow more common.

The author proposes a method that accounts for timing and complexity, both for drawing relationships out of a dataset and for explaining why a particular event happened. Theoretical and experimental case studies show how the method works in practice and test whether it succeeds.