Animals aren't "prediction machines" which "build models". This computational statistical framing is as much a misdirection as the "cogs & springs" model of the

Animals aren't "prediction machines" which "build models". This computational statistical framing is as much a misdirection as the "cogs & springs" model of the era when that was the new technology. Animals are biological organisms which grow. They grow their muscles, tissues, organs, nervous system, and so on. They learn to swivel their eyes, and coordinate it with their hands. (Growth isn't a computational process.) This can be characterised as "prediction" in the same sense that a gas flame under a pot "predicts" the boiling of the water. And "modelling" as much as a burn on the pot "models" the flame. Play is a sensory-motor physical adaption of the bodies of animals so they can grow sensory-motor techniques to manipulate their bodies and environments. It is in a very small number of animals do these techniques become abstracted into a discrete form of propositional intelligence. It is absolutely critical however, that this propositional intelligence is just grounded in sensory-motor techniques. You cannot "build" a model of an environment, because environments are infinitely informationally dense (ie., they are continuous and any discrete computational process renders that density as an infinity). So "model acquisition" is a continuous process, in this case, a type of growth not 'construction'. When composing propositions in 'propositional intelligence' humans are able to use discrete processes to form novel concepts, but that is rare in the animal kingdom -- and grounded in the growth patterns which furnish cognition with primitive sensory-motor concepts.