Contingency detection plus domain-specific inferences

This is a comment posted on Monday in the virtual debate around Pascal Boyer and Clark Barret's paper on Causal Inference.

The authors argue that rather than looking for a module of causal detection, detection of contingencies should be investigated in order to understand its role in constraining downstream inferences. Roughly, the system detects a contingency relation in a specific environment and matches it with a set of inferences that lead to relevant conclusions in that environment. For example, the detection of contingency relations between two moving geometric shapes, like two triangles, may trigger an intentional reading of the movement, that is, an activation of a “look-for-animate-agents” system. So, the idea is that we detect a contingency relation in a specific context and this triggers specific inferences. But don’t we need those specific inferential patterns to detect the contingency relations? In the case of the moving shapes, what is the difference between detecting “intentional-looking” contingencies, as Happé has defined them, and detecting contingencies in the context of intentional expectations? What would be the advantage of a two-step explanation of a contingency-detection that triggers a domain-specific inference? And also, what is precisely ‘causal’ in the detection of the co-presence of two or more stimuli in the environment, if their only effect is that they trigger an set of domain-specific inferences? How does this explain our expectations about causality, that is, our expectations about the connection of events and not simply their correlation?