> humans are not rational,
Okay, stop with this. Just stop for a moment and define rationality: chosing the best course of action based on the agent's available information (let's call information data evaulated according to the agent's preferences, maybe even mention utility function). Now, if people value shinyness over solving poverty, you get a lot of Apple customers.
Lo and behold, that's what's happening. People's utility functions are not simple money, time, iPhone, Nexus 6, what should I buy!? It also includes the whole spectra of every walk of life, spending 1000 USD on a shiny iWhatever is not just a device for iWhatevering, but also has the value you gain by having a nice device that your peers will want to look at. (Or maybe having a non-iWhatever will have a negative value because you'll feel left out, peer pressure, embarrasement! People are rational social agents.)
Yes, models usually don't deal with these stuff, but that's changing, mostly because now we have ways to quantify it (from twitter mentions to FB likes and easy sentiment analysis and DIY machine learning/NLP/etc.).
Just think about the discussion regarding healthcare. From an economics standpoint the continous situation for the past decade is a clusterfuck, a government mandate in such a well-developed state such as the USA would solve the problem, in a few years. Done. Cheap, effective, healthy, pick three. Oh wait, you can't because politics, because social status, because discrepancies in how people view the world. And even though you can model politics itself, there's not much point in it, because the actors/agents/dickwads are either too simplistic (so some kind of personal agenda driven) or too influenced by invested parties (healthcare providers) to care about your models, or your models about them not caring about your models.
And that's (one) full circle, but of course we could go on.