Slashdot Mirror

As demonstrations and looting took place in Ferguson, some friends of mine and many public commentators expressed disgust with some of the most prejudiced comments tweeted with the #ferguson hashtag. A few high-profile cases led to incidents such as security concerns at one high school and a teacher being fired from another, but most of my friends paying attention said it was more about the steady drumbeat of subtly racist, ignorant, or epically point-missing tweets limping past, often larded with passive-aggressive sarcasm. (Typical example that I just pulled from #ferguson, courtesy of "Wayne Dupree Show": "Liberal Logic 101: Blacks don't commit crimes, Police are just racist. It's sad but that's the narrative being pushed #ferguson #ericgarner". But on the other side, hashtag names like "#BlackLivesMatter" are pretty passive-aggressive too.)

It reminded me of the corruption of the original #GamerGate tag, which today is infamously known for crude sexist trolling, but in its original incarnation (as coined by actor Alec Baldwin), the hashtag apparently referred to some somewhat reasonable questions being raised about ethics in gaming journalism and the statements of one (female) indie game developer. Regardless of what you think of the original arguments or the people making them -- even if you accept, for the sake of argument, that everything they were saying was wrong -- they didn't deserve for the hashtag to be associated with sexist piggishness that became synonymous with #GamerGate, to the exclusion of any discussion of the original points.

Whether a hashtag is corrupted by opponents (#ferguson) or by Neanderthals who nominally claim to be supporting you (#GamerGate), in either case it's possible for a sufficiently large mob to effectively ruin the discussion for many of the participants. In the case of #GamerGate, the point of the original discussion was drowned out completely; in the case of #ferguson, a high proportion of tweets are still aligned with the original point, but a reader is still going to quit reading if each victim-blaming tweet depresses them so much that the next 10 decent tweets won't make up for it.

So, what can you do? You could follow only the people you trust to say something thoughtful (or, at least, not proudly ignorant), and filter their posts for the #ferguson hashtag, but then you'd miss the overwhelming majority of other people's tweets on the subject, even the good ones. You can follow all posts with the hashtag and block the most egregious repeat "offenders", but that won't help much when the problematic messages come from so many different accounts.

What Twitter could do, on the other hand, would be to set up a system for browsing tweets under a given hashtag that would reward the tweets that are given the highest rating by other users following the same hashtag. That would not replace the current Twitter default of strict reverse chronological order for tweets, which hardcore Twitter fans consider sacrosanct. But it could be an alternative model for browsing the tweets grouped under a given hashtag.

Similar to the system I suggested for Twitter to adjudicate abuse reports, a tweet under a given hashtag could initially be shown to a random subset of, say, 100 users who are following that hashtag, and rated as to whether the tweet is funny, informative, interesting, etc. (sound familiar)? Then if the average rating is high enough, the tweet would be shown to users who are browsing the "highest rated" tweets on a given topic.

(The simpler and more obvious solution would be to display tweets as "highest rated" if they had been favorited or retweeted by lots of people. However, this is problematic because it allows a person to game the system by having all of their friends -- or sockpuppet accounts -- "like" a tweet in order to drive it to the top of the pile. By having the ratings come from a random subset of users, this resists attempts to game the system, because there's no way for a user to ensure that their friends will be among the random subset that is selected to rate the tweet.)

This is, essentially, the same algorithm that I've recommended for many other similar problems, even including, say, ways to identify the best new songs in a given genre (so that trance fans can rate the best new trance songs, country fans can rate the best new country songs, and in both cases, the new songs with the highest average rating get the widest promotion to all self-declared fans of that genre). However, there's a signficant twist in the case of rating tweets under a political hashtag. Fans of trance music can be reasonably sure that country music fans are not going to sign up to rate trance songs and given upvotes to the stupidest trance music. But on the other hand, if you create the #ferguson hashtag to discuss reforms to the justice system, there's a good chance that plenty of trolls will sign up to follow the #ferguson hashtag if it gives them the opportunity to upvote racist and victim-blaming tweets that defeat the purpose of the original discussion. Even if you assume that the racists and victim-blamers constitute a minority of users following the hashtag, it might also be the case that they will have a higher response rate whenever they happen to part of a random sample which is asked to "rate" a given tweet to determine whether that tweet is promoted to a wider audience. The trolls might end up constituting a majority of votes cast, which would defeat the purpose.

So perhaps a modified version of the algorithm could work better. As before, new tweets under a given hashtag would be rated by a random subset of users following that hashtag. However, for some random subset of those tweets, the tweets would also be rated by a random subset of all Twitter users. (How to solicit ratings from the general population of Twitter users is a good question. If you simply displayed those tweets to random Twitter users in a sidebar and asked, "Please rate this tweet, even though it's for a hashtag that you're not following," the response rate would likely be very low. But whatever the low rate was, if you display the tweet and the rating request to enough users, eventually you will get a sample of ratings that is statistically significant.) If the system determines that, in many cases, the rating of the tweet's quality from average Twitter users is significantly far apart from the rating from users following that hashtag, then that hashtag can be considered "compromised" (i.e., the majority of people following tweets on that hashtag are probably trolls, or at the least, voting far differently from how average Twitter users vote). And then, perhaps, the highest-rated tweets under that hashtag could be displayed with a disclaimer saying that the ratings have probably been manipulated and are not reliable (but here are the highest-rated tweets anyway, in case you want to read them).

This does raise a philosophical question: What if some subset of Twitter users -- whether skinheads, or communists, or Beliebers -- want to engage in a discussion where posts are rated according to their appeal to members of that in-group, without regard for those posts' appeal to the rest of the user base? Isn't that a perfectly valid form of discussion? My sympathies lie against that point of view. Apart from the fact that the group obviously has the legal right to engage in whatever in-group discussion they want to have, I don't think it's healthy to engage only with like-minded people whose mindset is radically different from almost everyone else's. (In any case, the system could still display the highest-rated tweets, just with the ever-present reminder that those ratings are wildly different from the average ratings given by users who are not following the hashtag. Unfortunately that might just embolden members of the in-group who take pride in the fact that their philosophy sets them apart from most of the rest of the world.)

Unfortunately a "deference to the majority" also means that the protocol wouldn't do much good in cases where the majority really is wrong. If Twitter had existed 60 years ago and had implemented something like what I'm describing, then Twitter discussions of homosexuality or interracial marriage might never have gotten off the ground, because the majority probably would have downvoted anything advocating or even tolerating those lifestyle options. (What year would you guess was the first year in which surveys showed that a majority of Americans supported interracial marriage? 1997.) Peer review, even in the random-sample, non-gameable fashion that I'm talking about, doesn't do much good to advance the discussion when we are the trolls, oblivious to the things we're bigoted and ignorant about that we'll look back and shake our heads at in another fifty years.