"We suck at dealing with abuse and trolls on the platform and we've sucked at it for years," wrote Twitter's CEO, Dick Costolo, in an internal memo in February. And he was right. From death threats to members of the Parliament of the United Kingdom to an unending campaign of harassment against Anita Sarkeesian to GamerGate to high-profile celebrities and their families receiving floods of abuse and threats, Twitter had became an increasingly toxic environment.
It seemed as though Twitter has been standing idly by while trolls spread their hate, and Costolo's memo made it clear that not enough was being done. Since late 2013, Twitter has been tweaking how it responds, and quietly testing new methods of flagging harassing messages and acting against user accounts who posted them. The changes announced April 21 codify some of what people had observed was happening in tests. The proof is in what happens next, but the conditions that Twitter has set up could lead to a substantially improved experience for many on Twitter after a long period of little visible progress.
The asymmetric power relationship
The problem with Twitter's previous approach, as I outlined in an article here at Boing Boing last August on third-party collaborative blocking tools, is that it's far easier for an individual or a group of people—whether formally associated or a mob organized around a hashtag call to action—to harass targets than it is for targets to mitigate the harassment. It remains easy to register mass numbers of Twitter accounts or buy such accounts from others, and assemble fake followers. GamerGate demonstrated how "brigading" or "dog piling" could be used to have thousands of real and fake accounts relentlessly flood the mentions of a user, rendering that account almost unusable.
This can be especially awful when someone who is relatively well known retweets or mentions a person within a tweet who isn't, regardless of whether the reference is supportive or intended to encourage a negative response. For someone who has never encountered a large amount of excessive targeted negativity, it can be a surprise kick in the teeth. (This happens to regular people all the time, such as "Alex from Target," who was singled out for his looks, then received enormous backlash as well.)
The collaborative blocking tools, like Block Together, allow some automation. Randi Harper's Good Game Auto Blocker, which analyzes a few key GamerGate figureheads' followers, and creates a regularly updated filter list used by Block Together of accounts that intersect, allowed me and others to suppress the bulk of bad actors.
Twitter has gone through a few waves of improving how accounts are reported for abuse, making it a matter a few months ago of a few clicks, rather than requiring copying tweet URLs, pasting, and so forth.
Yet Twitter's fundamental problems remained:
It's too easy to create new accounts without any validation except an email address. While few people would suggest Twitter switch to Facebook's terribly implemented "real name" policy, the company interest in acquiring users outstrips its ability to throttle fake or disposable account creation. (Block Together options include blocking any account that is under seven days old that mentions your account, as well as any account that has fewer than 15 followers.)
Users have no tools available within the Twitter infrastructure that act at scale against an incoming barrage. Block and mute (as well as third-party client mute) are one-at-a-time actions.
Third-party clients weren't prepared for large-scale blocklists. With Twitter's Streaming API, in which one's timeline is fed in real time to Twitter's and third-party clients, blocking occurs in the client. Most clients weren't set up for this. I use Tweetbot, which like other apps retrieves just the most recent 5,000 blocked accounts, while I have over 12,000. A near-term Tweetbot update will raise the limit to at least 10fold, says the developer.
Twitter's enforcement seemed underwhelming. Accounts that released personal information or made specific and credible threats of harm would sometimes be suspended for hours or days and then return. Returning users often bragged that they had to do nothing to get their account back; many are recidivists, engaging in the same behavior repeatedly, and being suspended and resumed repeatedly.
The new policies could partially or largely solve aspects of all of these problems.
Timeouts and an invisibility ray
We've all assumed Twitter mines its data as well as third parties can, which would mean it should be very well placed to characterize accurately patterns of use that should let it identify abusers and act upon that information before reports are filed. And so they are. While this feature remains in testing, it could prove one of the most beneficial as it will act silently. Some users who receive regular abuse say they're already seeing a difference.
Twitter will automatically suppress tweets that mention a user from appearing in that user's timeline if they match patterns and content of abuse previously seen. Those who are continuously harassed often see abusive messages from people with accounts that are obvious giveaways: the biography includes abusive hashtags or imagery; the account is either new or has a low tweet count, and was put into use for this purpose; the abusive party isn't following the account it's talking at. This will help tweeters who are suddenly the subject of scorn as well as frequent targets.
People you follow won't have their tweets filtered, but others you don't who meet this criteria just won't have their messages show up in your stream. They can still post their messages, but the intended victim won't get them. This is effectively what collaborative blocking offers by creating a joint list of accounts that one person doesn't have to manually or continuously update. (This is also how per-user mute works in Twitter: the muted party can still follow, favorite, and retweet the muting account, and is unaware their behavior is invisible to them.)
On the enforcement side, Twitter is stepping up some behavior that it's definitely tried before, and codifying and extending it. Accepting reports and evaluating whether someone's message or series of actions constitute a violation of its rules is one thing. (Even there, Twitter has broadened one measure of abuse to read, "Users may not make threats of violence or promote violence, including threatening or promoting terrorism." It previously stated threats had to be "direct" and "specific.")
The new enforcement methods can include up to three elements: a specific amount of time a user is put in the corner before the account can be reactivated; the requirement to enter a phone and verify receiving a message at it; and the necessity of deleting one or more identified tweets. In the reactivation process, a user also sees a screen that reinforces the policy guidelines with links to more information.
The specific duration of a suspension is, I believe, new. In the past accounts were locked out for arbitrary periods of time, which weren't specified, or which required someone to contact Twitter to request reinstatement. (A much smaller number of accounts are banned, and those Twitter names are typically also permanently retired.)
Having to validate an account by providing a phone number reduces a user's anonymity. There's a lot to be said about the positive side of Twitter's current account registration policy for allowing freedom to speak (not freedom of speech) anonymously.
However, when Twitter evaluates reports or finds analytically recognized bad behavior, requiring an SMS code sent to a number could result in one of two outcomes: the person is chastened and realizes they can be identified should Twitter be subpoenaed; the user walks away from the account, which if it's one they've built up for a purpose, is a significant penalty. While it's possible to get disposable phone numbers for SMS and engage in other workarounds, the phone system is more limited than email, Twitter can identify re-used numbers and ranges used for the purpose, and only a subset of people will try to circumvent the system.
The final step is rather clever. If Twitter identifies tweets it wants deleted because they violate policy, it doesn't remove them itself. Rather, it requires the user to click Delete before getting an account reinstated, again setting up a choice to abandon an account or not. A suspended account's tweets are entirely unavailable, thus if they won't delete some, the user is effectively deleting all.
Its bat and ball
Forcing a user to delete tweets is, to be sure, rubbing users' noses in their behavior who are deemed to have crossed a line. But it's Twitter's world; we're just living in it. By tightening their definition of acceptable speech in their commercial realm, where free-speech rights are always constrained, this may make some uncomfortable.
Free-speech absolutists want all speech, no matter how offensive, to be legitimate. But this isn't true in the real world, even in a country like America that allegedly has among the broadest, if not the broadest, protections for speech. One should consider that police and executive powers, often with courts' cooperation, have restricted our rights to free speech in public places and free assembly for dissent.
Yet Twitter is a commercial space, not a government-controlled forum. And Twitter's rules typically relate to interpersonal interaction, not the kind of speech. One may say all sorts of things that others find offensive, but as long as they aren't targeted to an individual or specifically defined group and include statements designed to cow or coerce—or, worse, a preview of an actual plan of violence—Twitter won't remove those.
There are many things I may not want to hear, and I absolutely support the right of people to say them. Directed threats and other abuse are not in that set of ideas. In the physical world, such words could result in arrest, restraining orders, and convictions.
Now, it's time to watch and see how these policies play out. The best part about them is how invisible they will be to the vast majority of users who clearly use the medium sparingly, and a smaller group that simply wants to talk without the threat of violence, whether stating popular opinions or niche views that should be heard.
Image: From the 1923 edition of Grimm's Fairy Tales illustrated by Gustaf Tenggren