Don’t Say “Gay”, and Other Secrets of the Algorithm

The upsides may not be many after Elon Musk’s takeover of Twitter. For one thing, the company appears set on charging users for the exposure of their content via the $8 per month Twitter Blue option. In this, it is following the apparently inevitable path taken by everyone from Google and Amazon to Tinder.

But as some compensation, Musk has been opening the company’s books to scrutiny, at least to ideologically aligned observers. This has provided a window into the secret workings of the company’s algorithm before the Musk take-over.

One such observer is Dave Rubin, who last month posted a thread about two days he spent behind the scenes at Twitter’s offices in San Francisco. According to Rubin, users’ accounts were (are?) being hit with invisible strikes that downgrade their exposure on other users’ feeds. 

Now, this included perfectly understandable downgrades. An account would be downgraded in the aftermath of a suspension for publishing something blatantly discriminatory. 

But Rubin revealed that Twitter also had a softer category of downgrade, NSFA. Meaning “not safe for ads”, it is modelled on NSFW, “not safe for work” (i.e. porn). 

Unfortunately, the NSFA keywords were many and the database is context blind. Accounts are also not informed if they have been marked NSFA.

Meta (i.e. Facebook and Instagram) may be the endpoint of this sort of content moderating. Zuckerberg frequently describes the company’s policy as creating “positive communities”.

In practice, this has given both platforms a saccharine sheen. Automatic downgrades come into place for too much “negativity”, as with posts that include keywords like “kill”, “hate” or “ugly”. 

This makes political mobilisation challenging. Even discussing the news is hampered: Meta has been known to take down pictures of the Taliban and Hamas, regardless of context.

Meta, however, notified high-profile accounts of blemishes on their records, and even gave them the opportunity to remove or edit the targeted post. This may be extended to its incoming $12 per month paid “verification” service.

Like the two-tiered Blue system at Twitter, it is a means to keep the most influential people as happy users, while leaving the rest of us in the dark about the whims of the code.

Follow Christian on Twitter for more news updates.

Feature image courtesy of @krakenimages via Unsplash.

Sign Up To Our Free Newsletter