First, I think we need to at the very least separate out an opinion from a reward adjustment. If a minnow doesn't have enough voting power to upvote content, they should at least be able to signal they "like" the content (or in some way view a post in some positive way) without having to spam the comments with lots of one word replies. And also, we need to separate out the notion of flagging (which has very negative connotations) from simply lowering an excessively rewarded post.

So that first reason alone may justify some of the UI changes discussed in this proposal (as well as the notion of a generic upvote reaction at least), but it alone does not justify why more than one positive reaction would be necessary (or why any non-flag negative reactions would be necessary). And obviously I do think more reactions are beneficial (although I admit that we should keep the number very small to avoid UI bloat and user cognitive burden).

So why do I think multiple reactions can be beneficial? There are very common reactions that many people will have to content but most may not have anything more meaningful to say. Without reactions, these users are forced to either leave a comment with just a handful of words (taking up more space in the UI, and even adding more computational burden on the Steem blockchain than the custom_json_operation that a reaction would entail) or avoid transacting with the blockchain at all (which I would say hurts social interaction of the platform).

Furthermore, structuring these common responses allows us to aggregate them and display them in more compact and useful ways. It also allows client UIs to even use this meta-data to implement new ranking algorithms. For example, imagine a ranking algorithm that used the Worthwhile reaction given by a set of accounts you are following. You could discover the posts your friends thought were intriguing and maybe controversial (stuff that could expand your horizon and mind) while deprioritizing the stuff they all like and approve of (which likely is repetitive stuff you have seen many times before and that you may not be in the mood for at that time). Or maybe you have a ranking algorithm that considers the Funny reaction given by users weighted according to how often you have given them a Funny reaction; this could allow you to just focus on funny content (when you are in the mood for a laugh) ranked based on people determined to have similar senses of humor as you. I'm sure you can imagine more possibilities. All of this becomes much harder when people are forced to use unstructured ways of providing reactions such as comments; although machine learning could still do wonders there.