Twitter abuse and magic wands
Some top-of-the-head thoughts on the issue of dealing with abuse on Twitter. There seems to be a widespread impression that this is a clear-cut issue with an obvious solution; while I absolutely agree that there is a major, systemic problem with online abuse, particularly that directed towards women, I think many of the calls for action are glossing over the practical difficulties of dealing with it. It feels a bit odd having to say this, but: just because there is an obvious problem doesn’t mean that there is a simple solution. And there are many reasons to be wary when people start saying “something must be done”, without having a clear idea of what that something is. All too often, you start by saying that something must be done, and end by saying “oh god, we didn’t mean that”.
These are some of the more obvious questions that the debate has raised. To be clear, this isn’t an argument that no action should be taken, or that campaigns should cease - merely that these are some of the practical questions Twitter already have to address, and that those campaigning for Twitter policy changes should therefore consider as well:
- Who can report abuse? Is it only the victim - a person @messaged into a tweet - or can third parties flag abuse as well? If the latter, will the number of abuse reports a tweet gets influence the priority with which it is dealt, or influence the outcome of any decision? The former is limited (in the same way the PCC’s “only those directly affected can complain” policy was) and leaves wide avenues of abuse open, while the latter is hugely problematic. Firstly because third parties will, like moderators, often miss the context of a message or conversation and take it the wrong way, significantly increasing the number of abuse reports to deal with. And secondly, because it will entrench a level of power in the follower count that will damage many of Twitter’s best aspects. (This is already the case to an extent, of course, although I think claims about the power differential between those with large followings and those with less is often wildly overstated - Twitter is still far flatter in its power structure than most other mediums or institutions.)
- Is the demand simply for an easier way to report abuse, or is it for Twitter to actively police tweets more heavily - to ban more users, to pro-actively delete tweets, to react quicker, etc? How should it work in a real-time environment? Should users be instantly banned, or should there be a series of warnings before action is taken? Is there a right of appeal? Should there be consequences for frivolous or malicious abuse reports? How do you prevent the system being gamed to shut people down (as the “report spam” button already is)?
- What counts as abuse or harassment? Is it direct threats of violence only, or more general cruelty and unpleasantness? How will this be determined? What’s the boundary for legitimate argument and criticism? How should Twitter’s policies work in relation to legal frameworks; and which legal frameworks? Will moderation policies be different in different jurisdictions? Which jurisdiction has precedence when users live in different countries?
- Who will be doing the moderation? How does it scale to the size Twitter operates at? How many staff will Twitter need to employ to do this? How will this extra cost affect the nature and business model of Twitter? Will the moderation service be offered in every language that Twitter users speak, or is it English-only? How will moderators deal with dialects, sub-cultures, cultural references they may not immediately understand? Is there a danger of subtle, unintended discrimination emerging from having a pool of moderators with a broadly homogenous cultural background? How will the precedent of more active moderation of Twitter, and the creation of the infrastructure required to do it, be treated by governments who would like to legally suppress certain kinds of speech? Are there other ramifications of taking an interventionist stance on this issue that could affect Twitter in the future, on other issues?
- Is the demand that Twitter stop requiring users to fill in a form to report abuse likely to lead to better or worse moderation decisions? Does there need to be a mechanism for moderators to be given further context and background for the complaint, or must they go into it blind? Will this actually reduce the ability of moderators to correctly identify abuse?
- Is the idea of a paid-for Twitter workable? At what level should the fee be set for a global service like Twitter? A £5 entry fee might work to discourage drive-by abuse - but then again, it’s almost a week’s wages in Bangladesh. Does this vary by country?
- How will you react the first time somebody you like is banned because they lost their temper one night? Is there an unspoken assumption that people of a certain standing will be treated more leniently than others?
As a side note, one suggestion for automating the process that’s got a lot of attention is this one. It’s worth discussing, because I think a lot of its faulty assumptions are a common subtext to the debate:
A better way would be to automate the troll catching. This would be relatively simple.
The average troll:
* has created the account within the last month or so.
* has fewer than 50 followers.
* sends a disproportionate number of messages to people with blue ticks.
* uses a few trigger words repeatedly in their messages. You know which ones.
Using these criteria, it’d be simple to write an algorithm that screened and flagged suspicious behaviour without any need for human censorship.
Users could then be offered a ” safe Twitter” option, which automatically muted any user who’d been flagged as a potential troll. An opt-in filter, basically.
This is appealing at first sight, but flawed. The assumptions about the characteristics of “the average troll” are evidence-free assertions, and I suspect are likely wrong - many of these people aren’t necessarily new, or friendless, or spend all their time sending abuse. Quite the contrary: a significant aspect of the problem is that they are often part of like-minded networks.
Moreover, using interactions with “blue ticks” as a red flag embeds the idea that abuse is something only ever directed at verified Twitter accounts, or that this is the only type of abuse that matters; this is deeply wrong, and frankly rather unpleasant.
In any case, this mischaracterising of trolls ultimately matters little, because the plan falls down regardless. The first three points on the list of “ways to spot a troll” are also the characteristics of most new Twitter accounts; Twitter specifically encourages new users to interact with celebrities as soon as they join. These are useless as troll-detecting signals, because they apply to hundreds of thousands, maybe millions of accounts at any one time. Which leaves you with only the last point, a keyword filter, which is by a country mile the bluntest, least effective and most easily-evaded tool in the moderation shed.
There’s a reason why phrases like “it’d be simple to write an algorithm” generally lead to hollow, despairing laughter from developers.
Furthermore, the idea of an opt-in “safe Twitter”, while a nice idea on the surface, actually implies building a class hierarchy into the Twitter ecosystem. You’d have the verified users, who are protected to the extent that interacting with them in non-approved ways can get users flagged as abusive; the established users, whose interactions get no special protection, but are exempt from suspicion of abusive actions; and the underclass of new users, who can effectively be shut out from all conversation. It’s a weird bit of ladder-pulling, and I’m not sure it’s a structure that benefits anybody.
I think a lot of this discussion comes down to a difference in how some users see Twitter, and how Twitter sees itself. Is it a closed community, or an open network? Is it more like a very large MetaFilter - which has been for many years one of the best managed communities on the web, thanks to a small paid barrier to membership, a core team of thoughtful and engaged moderators, a dedicated space for community discussions of policy, and a relatively homogenous user base with shared cultural norms? Or is it just a communication platform, more akin to the phone network? It’s clear that for many users, their expectation is something closer to the former - a community in which mutually agreed social standards should be maintained - while Twitter generally prefers to see itself far more as a neutral, non-interventionist communications layer. These two versions of Twitter are largely incompatible.
To be clear again, none of this is a call for inaction. I have no doubt that there are improvements Twitter can make to the process, and like most social media companies they could take user relations a lot more seriously - but I fear that the gains from these will be marginal. The problem is complex, and you don’t get to only deal with the clear-cut cases. It’s (sadly) easy to come up with examples of obvious abuse and harassment, but these will not be the majority of cases, or anywhere close to the majority. It is really really hard for a human, never mind an algorithm, to determine the context and intent of a message, and to make a judgement on it that is consistent with thousands of other cases. To do it well at the scale of Twitter would take an army, and even then would likely be the source of more controversy then the lack of it creates.
I wish this was something that could be solved easily. It’s abhorrent the things that women have to put up with for daring to have an opinion. But ultimately, putting our faith in technical solutions to cultural problems generally leads only to disappointment and unintended consequences.
(For an alternative take on this - which also notes the practical difficulties of the issue, but comes to a somewhat more optimistic conclusion about how feasible this is than I do - Mary Hamilton’s blog post is excellent and insightful.)
(UPDATE: See also this post from @flayman which, while not a total solution, does offer an interesting take on how the particular kind of “swarm” attack that prompted the current debate might be handled.)
12 Notes/ Hide
- iamamro likes this
- bryceelder reblogged this from flashboy and added:
- sharonodea reblogged this from flashboy
- waldramings reblogged this from flashboy and added:
- pootling likes this
- bubbalou likes this
- adland reblogged this from flashboy
- adland likes this
- alshaw reblogged this from flashboy
- timtumble likes this
- tracylynnmurray reblogged this from flashboy
- tracylynnmurray likes this
- flashboy posted this