Action Representation and Purpose: Re-evaluating the Foundations of Computational Vision