Content Moderation At Scale Is Impossible; Naughty Kids In Wuhan Edition
from the masnick's-impossibility-theorem dept
I keep trying to point out that content moderation at scale is impossible to do well for a whole variety of reasons, including the fact that sooner or later some people -- or some large groups of people -- may try to game the system in totally unexpected ways. Witness this amusing example from the London Review of Books, reporting on the situation in Wuhan, China, which was ground zero for the Covid-19 coronavirus outbreak. With everything shut down in and around Wuhan, schools have moved to online learning -- and some naughty kids seem to have worked out a way to try to get out of having to do schoolwork: getting the app the schools rely on pulled from the app store via fake negative ratings.
Schools are suspended until further notice. With many workplaces also shut, notoriously absent Chinese fathers have been forced to stay home and entertain their children. Video clips of life under quarantine are trending on TikTok. Children were presumably glad to be off school – until, that is, an app called DingTalk was introduced. Students are meant to sign in and join their class for online lessons; teachers use the app to set homework. Somehow the little brats worked out that if enough users gave the app a one-star review it would get booted off the App Store. Tens of thousands of reviews flooded in, and DingTalk’s rating plummeted overnight from 4.9 to 1.4. The app has had to beg for mercy on social media: ‘I’m only five years old myself, please don’t kill me.’
Must tip my cap to the cleverness here, but on the content moderation side it shows, yet again, just how difficult it is to handle content moderation. No one running an app store or other platform prepares for a situation like this. In this case, at least, it seems likely that with so many negative reviews -- and now press attention -- the platform might take notice and discount the most recent thousands of reviews, but imagine having to keep track of every case where this is happening, often on a much smaller, less obvious, scale?
What seems easy about content moderation almost never is. Everyone seems to think it's easy until they're actually running a platform.
Thank you for reading this Techdirt post. With so many things competing for everyone’s attention these days, we really appreciate you giving us your time. We work hard every day to put quality content out there for our community.
Techdirt is one of the few remaining truly independent media outlets. We do not have a giant corporation behind us, and we rely heavily on our community to support us, in an age when advertisers are increasingly uninterested in sponsoring small, independent sites — especially a site like ours that is unwilling to pull punches in its reporting and analysis.
While other websites have resorted to paywalls, registration requirements, and increasingly annoying/intrusive advertising, we have always kept Techdirt open and available to anyone. But in order to continue doing so, we need your support. We offer a variety of ways for our readers to support us, from direct donations to special subscriptions and cool merchandise — and every little bit helps. Thank you.
–The Techdirt Team
Filed Under: content moderation, content moderation at scale, coronavirus, covid-19, dingtalk, ios, remote learning, students, wuhan
Companies: apple
Reader Comments
Subscribe: RSS
View by: Time | Thread
This is why I still think that a up or down choice works better than a rating. Providing people with a way to amplify their own like (or dislike) of something leads to a skewing of results.
[ link to this | view in thread ]
Those ratings are super useful when they're honest. But it's far too easy to weaponize them to either artificially boost an app or product's rating or to tear it down. A safer metric might be number of downloads versus number of uninstalls.
[ link to this | view in thread ]
The problem isn't the moderation
I would argue that moderation isn't really the problem here, the problem is schools relying on third-party services that they have no control over and which were designed for a very different use case. It's an easy enough problem to solve, just post an APK on your own website and nobody can take that down but you.
[ link to this | view in thread ]
Steam
During the beginning of the whole Epic Store exclusive fuss, Steam saw several games severely tank on review (most notably the newest Metro game)... they managed to put a solution into place as an attempt to mitigate review bombing... it's changed how the review bombing impacts the overall reviews, but still doesn't really solve the problem.
So there will be "something" done, but overall it's just going to muddy reviews all together (there are legitimate cases for a sudden spike in poor reviews, like the removal of a feature for no reason... or to put it in a 'higher tier')
Maybe someone should make a game about reviews :0
[ link to this | view in thread ]
Actually
Guess you haven't heard of Wikipedia.
[ link to this | view in thread ]
Re: The problem isn't the moderation
And then tell people to side load apps made by whoever that haven't been checked with Google's scanner (assuming they use Android)?
That solution may be easy, but it isn't good.
[ link to this | view in thread ]
Re:
It's especially ridiculous that you have to give someone a star to indicate they're terrible. How does that make any sense? Besides that, it's well documented that these ratings skew high. "Got what was promised" should be the average experience with a 50% rating, whereas people are encouraged to give maximum ratings for that.
[ link to this | view in thread ]
Re: Re:
I actually prefer a four-tier choice. Good or bad, great or terrible. No middle tier "3 star" choice that provides no help. No pointless 10 or 100 point scale that only allows trolls to bring down scores because, as you say, the larger the scale, the higher it tends to skew because people don't normalize their scores. 100 point scales are the worse because people think in terms of school grades where anything under a 60% is a usually considered a failure instead of just average.
[ link to this | view in thread ]
Re: Re: The problem isn't the moderation
If you don't trust the manufacturer of the app, why are you installing it? In most cases I trust them more than I trust Google.
Google catches the obvious malware, but they're also the delivery system for the less obvious malware. Moderating for viruses is no easier than moderating for content (it's probably harder, as bad content often isn't trying to hide that fact.) and malware has gotten through in the past and will in the future. Better to download from a reputable source in the first place rather than downloading any random garbage that pops up in a search result and assuming it's safe.
[ link to this | view in thread ]
Re: Re: The problem isn't the moderation
Given that it's China, they're very probably using Android but not Google Play.
[ link to this | view in thread ]
Re:
But even that's not a good metric: I've uninstalled lots of great games on Steam, GOG, and itch.io for the very simple reason that I finished them.
[ link to this | view in thread ]
Click-through "reviews" are rendered pretty much useless by self-selection bias. Basically, the only people who bother to leave them are sycophants and haters giving maximum or minimum rating, respectively, without thinking. The majority who didn't have any strong opinion, usually won't bother to review at all.
IMO, everyone who uses an app without reviewing should be given a default neutral rating, to represent that silent majority.
[ link to this | view in thread ]
At least dogs can breathe a sigh of relief with this news; they’re no longer being framed for eating homework the kids didn’t do!
[ link to this | view in thread ]
Re: Re:
Meh, that's still providing useful information about replayability, so it would be a good metric to have in addition to a star rating. But even better I think would be to weight star ratings based on the play/use time. A one star review from someone who has played a hundred hours counts a hundred times more than a one star review from someone who played for less than an hour. And you'd need some expiry method too, so that a five star review from someone who hasn't logged in for a year isn't overruling newer ratings. So maybe [star rating] * [play time - time since last login]...and then normalize that by dividing by the value as if every user had given five stars.
[ link to this | view in thread ]
moderation.
Isnt moderated.
There are allot of things that you can Cut, delete, augment but moderation tends to be Hard.
Trying to get things What??
you can get them to moderate
Speech,
how things are said and expressed
How things are done
But you cant balance things when the group DONT LIKE YOU..
[ link to this | view in thread ]
Re: Actually
https://lmgtfy.com/?q=wikipedia+edit+wars
[ link to this | view in thread ]
Re:
But how do you define "users"? Is it everyone who ever downloaded the app? Everyone who ever logged in? Everyone who logged in in the past day? Everyone who installed it once upon a time and forgot about it and left it running as a background service? I have a lot of installed apps that could break and I wouldn't notice. I even have some apps that I use daily that could have major bugs in major features and I would never notice because I'm only using one minor piece of the app. I don't want to be giving others a false impression that these apps work well when I really don't know or care.
As I posted elsewhere on this article, I think a better method would be to weight reviews based on the play/usage time of the user writing the review. So you still only get reviews from people who are actually invested in the app in some way, but one review from a loyal, long-term user will overrule hundreds from people who are just review bombing.
[ link to this | view in thread ]
Re: Re: Re:
As long as it's unambiguous what people should pick for something that goes well, but is nothing special. That's the usual interaction ("taxi got me where I asked to go, at the listed price"—that's not "5 stars"). I don't even know that "bad" vs. "terrible" is a useful choice; in practice that seems more based on how vindictive someone feels, rather than a real judgment of the product/service.
Really, I'd say anyone reporting a bad experience should have to identify the type of problem they had. At least whether the problem is in the product, the advertising/listing, or the seller. (What does "1 star" mean in the context of this Wuhan thing? "I didn't find it useful", "wasn't as advertised", "product is malware"—Apple could have very different responses to those.)
[ link to this | view in thread ]
Communication works well in a community--one where people genuinely care about each other and about the subject. It doesn't work--at all--where people are driven by malice. If neighbors hate each other, if their idea of a good time is to get together to burn cars or churches or Asian grocery stores, then telling them to TALK NICE is going to be rather a waste.
People who don't have enough local fellow-misanthropes used to go to the big city. Now they just go online. And you can't tell them not to do something because people don't like it. They are doing that thing precisely because people don't like it.
[ link to this | view in thread ]
Re: moderation.
Words salad goes down better with some punctuation or formatting dressing.
[ link to this | view in thread ]
“Impossible” Or Not, It Has To Be Done
When these companies have the power to make or break livelihoods on a whim, it becomes clear that their behaviour cannot continue unchecked.
This is why we have Governments and legal systems: to step in and impose rules once the “Wild West” no longer becomes a tenable way to live.
[ link to this | view in thread ]
Re: Actually
xkcd #1167
Perhaps you give Wikipedia too much credit...
[ link to this | view in thread ]
Still nope
When these companies have the power to make or break livelihoods on a whim, it becomes clear that their behaviour cannot continue unchecked.
That example of yours does not show what you think it does. While it's certainly aggravating to have trolls and/or puritanical pinheads flagging photos Facebook is not the one deciding to 'break' anyone's livelihood there, as that is simply another example of the difficulty of moderation at large scale, why you don't put all your eggs in one basket and yet another instance of a story as old as human civilization, one or more losers deciding to screw with someone using the tools available.
This is why we have Governments and legal systems: to step in and impose rules once the “Wild West” no longer becomes a tenable way to live.
Still waiting for you to list exactly what you think the platforms should be forced into doing that will magically make those problems go away. I'll even narrow it down, what 'rules' do you think should have been in place that would have prevent the example you linked above that wouldn't cause even more damage?
[ link to this | view in thread ]
Re: “Impossible” Or Not, It Has To Be Done
"When these companies have the power to make or break livelihoods on a whim, it becomes clear that their behaviour cannot continue unchecked."
Oh, hey, Baghdad Bob, welcome back. I thought I recognized your usual brand of inflammatory anti-google rhetoric.
And no, the fact that humans are being human STILL isn't a reasonable excuse to abolish actual freedom of speech. Rumors being harmful is something we've lived with for some time now and we still haven't put a cop in every pub to closely monitor what the patrons are saying.
[ link to this | view in thread ]
Re: Re: Actually
That isn't a counterargument to Wikipedia being effective moderation at scale. It is, despite the occasional skirmish.
[ link to this | view in thread ]
Re: Re: Re:
I feel like that would then be biased towards the hardcore fans of the game that put hundreds or thousands of hours into a game. Instead of just the gamers that played the game enough to be well informed.
Clearly there'd need to be a cap on how far playtime weighs into it in that hypothetical rating system.
But really, I'm not sure if such a system would actually work well.
[ link to this | view in thread ]