I love this idea, but it should really be initially implemented as a kind of A/B test.
Simply choose half of the UIDs at random and always show them the comment score, and don't show it to the other half. Track which comments are upvoted most when their score is visible versus when not (there are many other metrics to use as well). It should be pretty easy to determine how much groupthink plays into voting.
I suspect that most people only read (and therefore only vote on) already popular comments, so I suspect groupthink is fairly high here.
On the one hand, I completely agree with the scientific merit of your suggestion.
On the other hand, showing different HNs to different people doesn't exactly strike me as community-building behavior. I think it would make the site feel even more like Temple Grandin's barnyard, with us in the role of the cows. Will my species behave better if we design the floor like this? Will that tweak to the algorithm prevent us from starting fights among ourselves? Oh, no, a gathering of more than twelve people -- we'd better build a robot to break those up.
Simply choose half of the UIDs at random and always show them the comment score, and don't show it to the other half. Track which comments are upvoted most when their score is visible versus when not (there are many other metrics to use as well). It should be pretty easy to determine how much groupthink plays into voting.
I suspect that most people only read (and therefore only vote on) already popular comments, so I suspect groupthink is fairly high here.