Maybe a simple weighted vote (relative to how new is the member/karma) would solve the problem.
Or at least the mean from random sample from older members and another sample from newer ones, and use variance to check if there is significant difference, and maybe register another sample from members who are voting through the like button (but haven't already gone to the article from HN). A simple comparison of IP timestamps would help doing that.
(these are just suggestions I have no idea how the current algorithm works)
As long as there are points, and comments are sorted by popularity, not showing the points, I don't see how it can mitigate the problem of people trying to game the system.
If it is somehow measurable who is trying to do so, in obvious way, in some automated way and if this activity is at a level that affects the rest, then the user could be somehow penalized off that karma gained. Nevertheless this is democracy, so I would only expect this behaviour being penalized only if it really affects other in some measurable way, so that the community could approve penalizing users who are trying to advantage of lack of automated moderation.
The contrary could be done i.e. instead of penalizing users who are suspectable in such behaviour, preferably award recursively to each comment in the same thread more points, therefore the commenters who are 'legit' are awarded. Therefore, add a category of comment points named 'HN ponts' which are awarded automatically by the system.
