Position: Probabilistic Modelling is Sufficient for Causal Inference

Causal inference is a key research area in machine learning, yet confusion reigns over the tools needed to tackle it. There are prevalent claims in the machine learning literature that you need a bespoke causal framework or notation to answer causal questions. In this paper, we make it clear that you can answer any causal inference question within the realm of probabilistic modelling and inference, without causal-specific tools or notation. Through concrete examples, we demonstrate how causal questions can be tackled by writing down the probability of everything. We argue for the advantages of the generality of the probabilistic modelling lens, when compared to bespoke causal frameworks. Lastly, we reinterpret causal tools as emerging from standard probabilistic modelling and inference, elucidating their necessity and utility.

Science and machine learning often deal with questions of cause and effect. For example, does a drug actually cure a disease, or do healthier people just tend to take it? A common belief is that to answer these “causal” questions, you need a specialised mathematical language, separate from standard statistics.This paper illustrates that you don’t. We show that the familiar tools of probability are all you need. The key is to use a single statistical model for both the world we observe but also the hypothetical worlds we want to ask questions about. By explicitly specifying how they relate to one another, we can make inferences about events in the hypothetical worlds using data from the observed world only.Imagine that you have data showing that people who take more aspirin often get worse headaches. To find out if aspirin is actually helpful, or if the correlation between positive effects and higher aspirin doses spurious, we can add a hypothetical "intervened world" to our model—one where we decide the aspirin dose for everyone. By connecting these worlds through shared assumptions (like how the human body works), we can use real-world data to see what would happen in our hypothetical one.This approach simplifies answering causal questions, framing them as standard problems in probability. It shows that the prevalent specialised causal tools are just convenient shortcuts within this broader framework, making causal reasoning accessible to broader range of machine learning researchers.