Should we be basing models off of preexisting theories of the mind?

For example suppose that I believe that in order for me to manipulate an object in the mind I need to first train an archetype/schema of that object, as a basis of a mental model, before any manipulations can take place.

Should I make an AI that specifically compresses/trains to a schema model and then synthesizes interactions from the schema or do I assume that results of sucessful manipulation from more vague transformations means that the model was embodied?

artificial example mind schema train