For example, I can post a comment decrying Blub with a snide remark (e.g., "You wrote a 1,000 line Blub program? Was it 500 getters and 500 setters?" in a thread discussing software projects) that is both information free and mean (perhaps Blub wasn't the author's preferred choice, but chosen for him or required in order to build an application for the iBlubber). People on this site generally dislike Blub, so the comment will get upvotes without adding any value to the discussion (an example of adding value would be saying you were able to do this in 100 lines of Flub using its cool new hygienic macros with a link to a paper on hygienic macros in Flub).
That's not to say all comment score data should be gone. Comment scores can still be kept and comments could be displayed on stories in the other in which they're displayed now (a mix of comment score and how recently it was posted). Generally, what I've found is that comments showing up _first_ tend to be of higher quality i.e., overall algorithm works more often than not.
[NB: I work at LinkedIn and we do this for connection counts-- we want users to network with each other, but we don't want to make it a "who has the most connections" game, that's why when you have over 500 connections (which is perfectly legitimate and allowed), only "500+" is displayed as the count on your profile]
Users should live or die by their votes on that comment. If you vote up the blub comment, you should personally get the downvotes for it too. Upvotes should expose you to the karmic downside of superficial comments.
Especially because the really good comments, the ones most deserving of upvotes, don't seem to get a lot of downvotes; watch the scores on a 'patio11 comment closely sometime to see an example.
It's actually based on a real comment I saw: the discussion was about migrating a 10 MLOC (iirc) enterprise Java system to git. One comment said that this system must have been "5MM getters, 5MM setters". That struck me as particularly mean and below the belt strike against the programmers who worked on this system: it's very likely there is a good reason why it had to be in Java in the first place (and other JVM languages may not have been available when it was created) and even so, it didn't mean the programmers working on it would have chosen Java as the language themselves (but they were not there when the architectural decision was made). Further more, it added nothing to discussion.
This is not unlike poking fun of somebody for wearing the wrong kind of clothes on the school yard: cheap way to score social points with the plurality of others present, mean and ignorant (may be they can't afford the right kind of clothes, may be they are going hiking right after class).
An insightful comment would have been something like "That's great that you were able to get this into Git, changing a VCS is a painful task. Have you considered using Scala in some of the modules? Functional objects, case classes and implicits could help you model your business domain better, write thread safe code, and get rid of much of the boiler plate."