It's not that economists ignore human psychology / behaviour, rather economists are interested in finding simplified models that provide some kind of generalizable insight. The models will include varying levels of complexity depending on what question the economist is trying to answer and what data the economist could reasonably access.

In many cases what motivates a particular individual provides little insight into what happens in the economy as a whole. For example, a model of start-up success that included founder education level and sci-fi movies watched by founders would provide little, if any, greater insight than a model the model that looked at just education level. It's similar to over-fitting in machine learning – the extra parameter is probably harming the usefulness of the model by making it less general.

On the other hand, CEO cocaine addiction might be a very good predictor of company profits, but you would have a hard time getting accurate data for that regardless of how good your AI is. Furthermore, I could reasonably hypothesize that some CEO's would respond to cocaine by out-performing non-cocaine users while some would under-perform depending on physiological factors. Again, without massive invasions of privacy, economists will not be able to factor this into their models.