-----Original Message----- From: Rowan Collins [mailto:rowan.collins@gmail.com] Sent: Friday, July 15, 2005 10:56 PM To: wikipedia-l@Wikimedia.org Subject: Re: [Wikipedia-l] try having this language discussion on Brainjammer

[Warning: this post will drift further off topic the more of it you read. :p]

On 15/07/05, Bennett Haselton bennett@peacefire.org wrote:

...

This language thread -- which I probably wouldn't be able to follow even if I read through the archives to see how it started before I joined -- seems like it has all the characteristics of the type of out-of-control Internet debate that Brainjammer was designed to avoid:

Perhaps before you start advocating the use of this system, you should put somewhere a description of what it *is*. Looking at some of the existing threads, I couldn't immediately see anything other than a bog-standard forum.

I presume your "killer feature", as it were, is the ability to mark posts with a "status" in terms of their relationship to the parent.

The earlier message that I posted, described how the algorithm worked in more detail, and why I believed that it would produce more useful results than algorithms that have been tried in the past: http://mail.wikipedia.org/pipermail/wikipedia-l/2005-July/040966.html

The "killer feature" is not just marking posts according to their status with respect to the parent. The killer feature is more like a rule (which is to say, a rule that is enforced by having users vote down posts that are challenged as not conforming with this rule) which says that rebuttals must only focus on one point at a time. (More than once I've replied to someone's post saying "You made several good points here but to go along with the rules you should break up this post into multiple replies.") My theory -- as yet untested, but which seems strongly intuitive -- is that people will argue (and vote, if it comes to that) more honestly when considering only one point at a time. If you're asked to vote between an essay making a series of points and another essay making a series of opposing points, there's too much wiggle room for the brain to rationalize voting in accordance with your preconceived beliefs. But with one point at a time, there's less wiggle room.

And since even one rebuttal against a point -- no matter how trivial -- is enough to disqualify an entire post (voters are specifically instructed to vote on the specific point and *not* to take into account how significant it is as a part of "the post as a whole"), the only posts that will survive are the posts that meet that high standard of truth all the way through.

This seems like an intriguing idea, but imperfectly implemented. I know it's just the germ of an idea right now, but given that you're publicly advocating its use, here are some criticisms:

• The most obvious imperfection is that the status labels are not

particularly clear - why is "awaiting response" styled the same as "incorporated into parent"?

Yes, the front end needs some work, but the focus should really be on the top-level posts that get generated as a result of all the feedback propagated up from the replies.

One difference between this and a traditional discussion forum is that the end result of each thread should be as compact as possible -- a single post, ideally -- rather than just an archive of everything that has been said.

• What's more, if the premise that posts can be made redundant by

altering their parent holds, why are "incorporated" and "withdrawn" posts shown at all? Deleting them outright would be foolish, but surely showing them by default defeats the object of

That's a valid point, and a UI decision -- eventually those will probably just be displayed at the bottom of each thread, below a barrier that clearly separates them from the replies that are still "active".

• Who decides when a post has been successfully incorporated? If Bob

replies to Alice's comment, and Alice amends the original comment, it seems to me that first Alice must tell Bob that she has done so, and then Bob must accept the amendment and withdraw the reply. If your existing model follows this process, it's not obvious from the presentation.

Current model: Bob replies to Alice's comment. Alice incorporate's Bob's comment and also inserts a note explaining how she incorporated Bob's comment into her own post. Bob receives a notification that his comment has been incorporated. If he doesn't think his whole point was incorporated properly, he can post another reply.

This does leave a weakness where Alice could cheat -- if Bob posts a reply clearly disagreeing with Alice and Alice "incorporates" it without making any changes, that can prevent Bob from ever forcing a jury vote. In the long run I'll have to decide how to tweak the architecture of the system to prevent this. In the short term, I think it will be really obvious if someone is doing this just to block a vote from coming up, and I could tell them to cut it out without generating too many complaints about "dictatorship". (Although it's true that one design principle is to eventually enable these discussions to take place without *any* centralized moderation.)

• Your presentation also lacks any kind of "audit trail" - it would be

nice to see how Alice has changed her post, and how Bob responded to her claim to have incorporated his response. Your forumish interface doesn't seem suited to tracking such things.

That's a good suggestion. Currently when you incorporate someone's comment, you add a note explaining how you incorporated their comment, and that note is not currently displayed in the thread view, but it probably should be. We could also add a field for Bob to add a comment about whether he feels his suggestion was incorporated correctly.

Eventually I might want to include version history for each post so you could see *exactly* how it was modified to incorporate a given comment, complete with diff functionality to show exactly what was changed, but I didn't have time to implement that this time around.

And finally, you realise this is more-or-less exactly how discussion works on a "traditional" wiki (e.g. c2.com, meatball, etc)? i.e. the aim is to refactor a discussion ("ThreadMode") into a coherent "DocumentMode", and to encourage opinions to be expressed by amending and expanding the existing text rather than simply creating a "rival" document. The main difference seeming to be that you are aiming to create a structured and non-redundant discussion thread, rather than a stand-alone document.

Actually the idea *is* for each Brainjammer thread to result in, ideally, a stand-alone document: the top-level post. But I think it's one of those situations where this new design kinda looks like what has come before (a Wiki), but with a few subtle design changes that I'm hoping will produce radically different results. Specifically, the enforcement of the rule that disputes of any points have to be as specific as possible and handled one at a time, on the theory that even when discussing highly controversial subjects like George W. Bush or Windows vs. Linux, people will argue and vote more honestly when focused on specific points.

-Bennett