Standard, neo-classical economic theory taught in American universities is predicated largely on a model of the so-called homo economicus, or rational man. This vision involves an agent with rational preferences and capacity to advance ones own self-interests as defined by those preferences.
Now, is this accurate? You might say it is a mere model of reality, an attempt to simplify behavior in a way that succumbs to analysis and that in doing so, it is useful. But, questions of epistemology aside, the utility of the model is often times in the eye of the beholder.
Modelers who use statistical regression in econometrics often are pleased with models that can explain about ten percent of the variations data. Think about that. The other ninety percent of the information goes unexplained, unexploited.
This is not to say that the revolution started by luminaries such as Marshall, Jevon and Walras have built a house on sand. But, with models that can often times hardly explain a minority of the data, let alone the majority of the data, you start to see the problem.