Relational Structural Causal Models

An artificial intelligence must have a model of its environment that is causal, supporting reasoning about interventions and counterfactuals, and also combinatorial, supporting generalization to unseen combinations of objects. In this work, we formally study when and how such a model can be learned. We develop relational structural causal models, extending structural causal models (Pearl 2009) to settings where objects and their relations vary. First, we show how answers to not only causal but also observational queries about unseen combinations of objects can not be identified without further assumptions. To enable such identification---including in the presence of unobserved confounding---we define relational causal graphs and derive symbolic identification criteria. Finally, we propose relational neural causal models, a provably correct approach that outperforms non-relational baselines on simulated traffic scenes with varying cars, signals, and pedestrians.

An important problem for artificial intelligence is how to develop models that can do well in settings that were not seen in their training data. For example, a self-driving AI may encounter a configuration of signals, pedestrians, and cars that it has never seen before. How can we guarantee it will act and decide correctly in this situation? In this work, we develop an approach to solving this problem by introducing relational structural causal models. We first give mathematical guarantees for when an AI can generalize in this way when given a set of guiding principles (a graphical model). Then, we develop an approach that uses neural networks to perform such reasoning, predicting not only correlational but also causal and counterfactual outcomes in settings with unseen objects and relations between them.