Learning about causation : dynamic epistemic modal logic with causal awareness

In this paper, we enrich the existing dynamic epistemic modal logic of causality with dynamic awareness of causal variables. In our setting the agent can become aware of (previously unavailable) causal variables. This approach allows studying important aspects of the development of agents’ knowledge about causal structures. First, we propose the language and discuss some notable validities. We consider scenarios that feature the expert (teacher) who has full knowledge of the causal dependencies, and the layperson (learner) whose knowledge is limited to only the variables she is aware of. Our models and our logic are dynamic: They allow observations by both the expert and the layperson, and they accommodate gaining awareness of causal variables by the layperson. The logic also allows interventions—counterfactual manipulations of values of variables. In the epistemic causal awareness models, the different epistemic levels of the expert and the layperson correspond to implicit and explicit knowledge, respectively. Those different perspectives can also be seen with respect to the causal graphs, in a 2-layer representation (so-called hybrid causal graph) that makes use of ‘shadow’ causal variables. Overall, our framework connects the existing lines of research on dynamic epistemic logic of causality and dynamic awareness with learning and pro-activity in AI.