How does the brain model the physical world?

Josh Tenenbaum has argued that the brain must harbor something like an “intuitive physics engine,” which it uses to simulate reality. This physics engine is likely hierarchical in nature: the top level contains a high-level description of the world (including all the objects, the environment in which they are located, and their predicted path), which then drives lower layers.

Using fMRI in monkeys, we have identified a region in the macaque brain, CIPS, that seems to be specialized for coding 3D space. We hypothesize that this region harbors the physics engine. We seek to understand (1) the neural code for 3D object shape and location, (2) the neural code for environment geometry, and (3) the neural mechanism for simulating the future trajectory of objects.




Architecture of the brain, viewed as an intuitive physics engine.At a very high-level stage in the brain (represented by the box “cat”), there is a piece of machinery capable of encoding, using very high-level code, the objects in the environment together with the environment itself. Furthermore, this machinery encodes intuitive physical laws governing these objects, allowing it to predict what will happen next.  The deviation between what is predicted (“predicted frame”) and what actually occurs (“actual frame i+1”) drives learning, to update the model(adapted from Lin & Tegmark, 2016)