Thursday, July 22, 2021

Twitter Corruption

Principle: that which can be abused will be abused. 

Twitter's algorithms are all sorts of whacked out, but apparently they're not wacky enough, because they're opening new loopholes. 

Allegedly dislikes are purely for giving information to Twitter. This is probably how it was sold to any vestigial-conscience employee of Twitter. In reality it will be used to drive all sorts of unsavoury behaviour. Dislikes will determine algorithmic promotion and demotion, affect moderation decisions, identify 'low quality' tweets which do not appear by default, and generally give Twitter an excuse to censor anything it happens to enjoy censoring. 

On the plus side it means I get to make concrete predictions with a suitably-fast confirm/deny cycle.

First, because dislikes aren't public, there's three games you can play: having the algorithm dislike tweets instead of real tweeters, having some dislikes be more equal than others, and preventing some accounts from affecting the dislike total.
I suspect twitter analytics will show dislikes to tweet authors, but if you'll notice, you can see specifically who liked a tweet. There's no feasible way to audit the dislikes. You would have to survey 100% of twitter interactions, have 100% of them answer you, and then you can do the accounting.* This means some twisted mental-case Twitter employee can conjure up thousands of dislikes on your tweets and nobody can catch them. Similarly, you can be dislike shadowbanned, both giving and receiving. Banned by fake dislikes or by bluchecka dislikes specifically, or be banned from giving dislikes (from disliking a bluchecka tweet one too many times). They will still display a dislike to you but it won't register to the algorithm.

Even with the private dislikes, because it's not auditable, the author's analytics panel may well display a fake number of dislikes. Impressions are already rather fake; if you have an old enough account you can see that shadowbanned accounts still register impressions from followers who don't see the tweet, which can be measured by engagement ratios. "I guess they didn't like the new tweets." No you're just banned. 
If your tweets are being shadowbanned by dislike ratios, they may hide this by not showing you the dislikes. Even without an audit if you get 1000 impression and 10,000 dislikes, obviously something's up. This sort of behaviour provides Twitter with a layer of plausible deniability, helping them preserve their vestigial moral compasses from realizing what they're doing.

Likewise if you see a tweet has been buried and only has 3 dislikes, something is clearly hecked up. Bluchecka and other favoured accounts will have their dislikes count for much more - or perhaps simply add 1000 to the database or something of that nature. Thus the analytics panel will display fake numbers to you so the bad actors can't be caught red-handed.

The algorithm will likely have a few dislike thresholds, based on whatever made sense to the boozed-up or sleep-deprived/caffeinated Twitter employee who couldn't dodge the responsibility in time. If you have access the database you'll see that bluchecka don't get de-prioritized even if they pass the thresholds, and others do get hidden even if they don't.

Third, Twitter doesn't want to hire enough moderators, because moderators cost money, and filtering them correctly for character is, like, work. The algorithm will determine which Tweets are heresy uh I mean 'hateful' by checking for dislikes, especially more-equal dislikes, and secondly prioritize moderator attention by dislikes. 

Perhaps Twitter will play games with 'leaking' dislike totals...having already fudged them to hell and back, they will show the real database entries.

*Refresher: accounting is a superpower. Likely limited to to the top 1% the population or less. In theory up to 12% could learn to, but in practice they don't.

No comments: