There are good reasons to have polls and predictions but they are far from necessary.
The original claim was that election prediction is useless -- are you willing to concede that election prediction isn't useless?
1. First, when many pundits were basically saying there was no chance that Trump would win, Silver was particularly highlighting not just that it was in the realm of possibility, but would not even be very surprising if Trump won (e.g. I think Nate does a good job with his "same probability of a team down by 3 points at the beginning of the 4th quarter coming back to win" to clarify to a lay person that this isn't unusual).
2. When making political predictions based on polling that turn out to be "wrong" (i.e. end up going the way of the < 50% option), it's helpful to think of why this might happen. In Trump's case, it was pretty obvious he was a very different type of politician than what came before, so it's worth thinking whether that difference might result in widespread error in some polls. Also, whenever a race is very close, it's obvious that any small movements one way or the other can affect the outcome. In 2016 Trump lost the popular vote by a substantial margin but won the electoral collage by a significant but not huge amount. In fact, the "all or nothing" way the US electoral college system works has the effect of amplifying smaller differences into larger vote outcomes.