It would be an interesting feature to add to allow other users to see what you upvoted. This way if you upvote something silly they can attack you like wild animals or more hopefully you can see what they like and like it too.
I think this should also work with comments as you can easily click the wrong direction by mistake. Anyway, I like the idea of negative voting with a cost. How about:
You need 20 karma to down vote and each vote costs one karma point.
Actually I like the idea - it would prevent (or at least raise the cost of) karma-bombing. I think the threshold should be higher than 20 though - maybe 50 or 100. It would probably be best to introduce it at a high minimum and then lower it after observing activity, like pg said he's doing with the polls.
I probably will never use the poll feature again, or if so very very very rarely. This was actually a question I asked myself a couple times these past couple weeks whenever I saw a post that made me think "why am I seeing this on HN". Combined with the new feature that I wanted to try out to see how it worked it made sense to make a poll. Now I've done it and the novelty is over.
There would have to be a cost to clicking on it, or people would click just to see who the voters are. I wonder if it would be enough to have it hide the article.
A downarrow doesn't have to work in the exact opposite way as the uparrow. In the trivial case the downarrow would do nothing, and then site wouldn't change at all, except cosmetically. In the more generalized case you would:
-weight it versus upvotes
-weight it based on user history
-weight it based on user's propensity to downvote
-weight it based on the number of distinct "user clusters" it offends
(If one group always votes for butter-side up stories, and another always votes for butter-side down stories, and a story annoys both of them, chances improve that the story is worth voting down.)
Another interesting metric might be contentiousness. If a topic generates an outlying number of downvotes from trustworthy users, and an insignificant number of upvotes (or many upvotes from unpopular users), its participants can be labeled as positive or negative forces. Someone who posts articles that generate a lot of downvoting in the comments might not be picking the best topics. There's probably a nice linear algebra way to generalize this in the same vein as PageRank.
A closed-source, nearly-null downarrow would at least give pessimistic users the feeling that their voice was being heard. (Heuristics is one area where it makes sense to keep things closed-source. Changes should be explained, but not necessarily in exacting detail.)
Added. This makes some sense. I often find that on reddit some stories that are actually interesting get downvoted to oblivion for no apparent reason and yet at a later time are up voted to the front page with a similar headline. I'm sure there are other cases where it would bring up other problems.
It's also an ability to manipulate. If I downvote every 2 point story around my 2 point story, people are more likely to see mine over a possibly better story.
I left Reddit because that's what happened with every story. You submit something and it gets burried within 10 seconds. If it was 10 minutes, sure some people red it but 10 seconds is barely time to read a title.