From the outside looking in, it's hard to say how much of a human element there is. Certainly it must operate algorithmically (hands-off) most of the time, because Uber operates in way too many markets for someone to babysit each one (including significant sub-regions of each, such as Manhattan and SOMA) around the clock.

My theory is that they may have goosed the algorithm based on prior years, or perhaps their algorithm considers historical data for all (or just important) dates. Creating an algorithm that correctly models demand / pricing for all regions is pretty tough, and there are always going to be situations where what seemed reasonable gives you incorrect results.