I think this is the least wrong post here. If we assume that our pattern recognition ability, which has obvious evolutionary advantages, is the source of empathy, which makes sense to me in terms of individual selection, then we can see that looking at a far situation will trigger certain pattern recognitions, specifically looking at our past experience. Based on my past experience more gain comes from being the underdog and winning than being the assumed winner and winning. Because I can see that, emotionally I will identify with the underdog more often because the outcome will be greater for individuals in that group and people tend to identify with individuals in far populations rather than groups. I'd add that personally being a part of the underdog group and winning would have much more of an impact than being a part of the assumed winning side and winning, much like a gambler remembers the wins more than the losses, and thus I would be pulling for the underdogs.
This can explain why my reasonableness will lead me to support the overdog in close situations. If there is a split in my group and I have to choose which side I'm on pattern recognition helps me realize that I am more likely to come out ahead if I ally with the overdog.(uberhund?) Thus, in such a situation I would be more likely to support the uberhund than the underdog because it directly affects my situation.
If I have to stick with a very unpredictive hypothesis, I have a decreased ability to predict the world, I will therefore do worse.
Not true. If you have a prediction model that is non-random and wrong you will get better results from simple random predictions.