Figure 1.

This figure represents the relationships between the various states due to the Markov blanket partition: active states are directly influenced by internal states and sensory states are directly influenced by external states. Only through these influences are the internal and external states are coupled to one another. The figure illustrates the process of an emotion being communicated between two human agents. Another person (external state) expresses an emotion audibly, which furnishes sensation for the agent. The agent performs inference, by listening and interpreting the sound based on prior knowledge, experience, and contextual expectation, performing motor action if needed (such as listening attentively). This leads to the agent forming an understanding and updating their internal states accordingly. Here, our classification of emotional states is quite simple: emotions are either part of an agents' “internal states” or “external states” (which are just the internal states of another agent) meaning they are a feature of the self or the other. In this sense, emotional states exist as the result of a sensation (i.e., as an inference about the cause of that sensation). In terms of communication, this then means that the awareness of other people's emotional states results in a kind of advantage for a human, in the form of a more accurate generative model for the reasons behind other humans' actions. Clearly this is a very coarse-grained view of emotional inference. For greater detail on the intricacies of emotion research, please see (LeDoux, 2000; Seth, 2013; Smith and Lane, 2016; Panksepp et al., 2017).