Learning Composable Signals for a Cognitive Substrate

According to the cognitive substrate hypothesis, human intelligence arises from the integration of specialist parts that are mostly shared with other mammals. In order for such specialists to cooperate in novel situations, however, they must agree on a system of signals that can describe aspects of the new situation using relations between familiar elements. This poses a problem because many signals can only be learned from experience, yet different specialists may experience the world in very different ways. This paper demonstrates that distributed learning of such signals is possible, and also that the apparent problem is actually a benefit, using a vision specialist and a hearing specialist that together observe a simulated four-way intersection. Using a heuristic method based on Allen’s time relations, these two specialists agree on a set of composable signals, and some dynamics of the simulation are captured in the differences in how the two specialists interpret signals.