Bodily models, not internal models

I want to briefly advocate for thinking about active inference type ideas in terms of bodily models rather than internal models. A bodily model is a system of relationships constructed by a collection of parts that share interoceptive signals with each other. The interoceptively connected collection is the body; the system of relationships is the model.

There are a few advantages to thinking about bodily models instead of internal models. The main one is the emphasis it places on interoception sharing between agentic parts. The bodily model is created by the interactions between the agents and serves as an means for them to coordinate with each other to achieve mutually compatible plans. Whenever this kind of phenomenon is important to the object of study, then the bodily model concept will be more useful than the internal model concept.

Bodily models are also more general than internal models because there’s no concept of location. 2D and 1D objects require us to rethink what “internal” usually means, but not what “bodily” means. There’s less human judgment involved in drawing the boundary between body and non-body because you just look for the interoception sharing. The economy, for example, doesn’t have an interior in the usual sense, but clearly has a body.

The emphasis on generalized bodies also allows us to draw on ideas about embodied intelligence and apply them to areas where we usually wouldn’t, like the economy, AI, and collective intelligence. Thinking about bodily models makes it easy to draw on related ideas.

Bodies also offer many interesting examples of phenomena that seem peculiar but may generalize. Bodies get Parkinson’s disease, which may be macroeconomic in nature. Using the idea of bodily models helps us access more examples and more data, allowing for more general and more powerful discoveries.