A Measure-Theoretic Axiomatisation of Causality

Overview

• The paper proposes a new framework for understanding causality, viewing it as an extension of probability theory and a study of what happens when intervening on a system.
• The authors argue for building on Kolmogorov's measure-theoretic axiomatization of probability as a starting point for an axiomatization of causality.
• They introduce the concept of a "causal space", which consists of a probability space and a collection of "causal kernels" that encode the causal information of the space.

Plain English Explanation

The paper tackles the fundamental question of how to formally define and reason about causality. Causality is a crucial concept in many fields, from science to philosophy, but there is still no universally agreed-upon way to define it mathematically.

The authors propose building on the well-established mathematical theory of probability, proposed by Andrey Kolmogorov. They suggest that causality can be seen as an extension of probability, where we not only look at how events are correlated, but also what happens when we actively intervene and change the system.

To capture this idea, the authors introduce the concept of a "causal space". This consists of a standard probability space, along with a set of "causal kernels" that describe how the probabilities of different events change when we intervene on the system. This framework allows the authors to rigorously analyze various aspects of causality, including cycles, latent variables, and stochastic processes, which have been challenging for other approaches.

Technical Explanation

The key idea in the paper is to view causality as an extension of probability theory, rather than as a separate concept. The authors start by arguing that Kolmogorov's measure-theoretic axiomatization of probability provides a solid foundation for building a formal theory of causality.

They then introduce the concept of a "causal space", which consists of a probability space (a set of events, a sigma-algebra, and a probability measure) along with a collection of "causal kernels". These causal kernels are transition probability functions that describe how the probabilities of different events change when we intervene on the system.

For example, if we have a variable X that causally influences a variable Y, the causal kernel would describe how the probabilities of different values of Y change when we set X to a particular value. This allows the authors to formally capture the notion of "what happens when we intervene on a system", which is central to many causal reasoning tasks.

The authors show how this framework can shed light on various long-standing challenges in causal reasoning, such as handling cycles, latent variables, and stochastic processes. By grounding their approach in measure theory, they are able to provide a rigorous mathematical foundation for causal reasoning.

Critical Analysis

The authors' proposal for a measure-theoretic axiomatization of causality is a valuable contribution to the ongoing efforts to formally define and reason about causality. By building on the well-established foundations of probability theory, the authors have created a framework that is both mathematically rigorous and conceptually appealing.

One potential limitation of the approach is that it may require significant technical expertise to fully understand and apply. The concepts of causal spaces and causal kernels, while well-defined, may be challenging for some readers to grasp. Additionally, the authors do not provide extensive examples or applications of their framework, which could make it harder for researchers to understand how to put it into practice.

Furthermore, the authors acknowledge that their framework does not yet address all the complexities of causal reasoning, such as the role of time and the potential for feedback loops. While they argue that their approach can shed light on these issues, further research may be needed to fully incorporate these aspects into the framework.

Despite these potential limitations, the authors' work represents an important step forward in the effort to develop a comprehensive and rigorous theory of causality. By bridging the gap between probability theory and causal reasoning, they have laid the groundwork for fruitful future research in this critical area.

Conclusion

The paper proposes a novel framework for understanding causality, grounded in the measure-theoretic foundations of probability theory. By introducing the concept of a "causal space", the authors have created a rigorous mathematical foundation for reasoning about causal relationships and the effects of interventions on a system.

While the technical complexity of the framework may present some challenges, the authors' work represents an important contribution to the ongoing quest to formally define and reason about causality. Their approach has the potential to shed light on long-standing issues in causal reasoning, and may pave the way for further advancements in this crucial area of research.