Vice News reports that some prominent Republicans on Twitter are being shadow-banned by the site. Specifically, Vice found that the names of several prominent Republicans do not appear in an autofill list when searched. Meanwhile, every member of the progressive caucus does appear in the autofill choices when searched. Twitter acknowledged the difference but denied it was the result of content discrimination:

The Republican Party chair Ronna McDaniel, several conservative Republican congressmen, and Donald Trump Jr.’s spokesman no longer appear in the auto-populated drop-down search box on Twitter, VICE News has learned. It’s a shift that diminishes their reach on the platform — and it’s the same one being deployed against prominent racists to limit their visibility. The profiles continue to appear when conducting a full search, but not in the more convenient and visible drop-down bar. (The accounts appear to also populate if you already follow the person.)

Democrats are not being “shadow banned” in the same way, according to a VICE News review. McDaniel’s counterpart, Democratic Party chair Tom Perez, and liberal members of Congress — including Reps. Maxine Waters, Joe Kennedy III, Keith Ellison, and Mark Pocan — all continue to appear in drop-down search results. Not a single member of the 78-person Progressive Caucus faces the same situation in Twitter’s search…

Presented with screenshots of the searches, a Twitter spokesperson told VICE News: “We are aware that some accounts are not automatically populating in our search box and shipping a change to address this.” Asked why only conservative Republicans appear to be affected and not liberal Democrats, the spokesperson wrote: “I’d emphasize that our technology is based on account *behavior* not the content of Tweets.”

GOP Chair McDaniel posted a link to the story with this statement:

The notion that social media companies would suppress certain political points of view should concern every American. Twitter owes the public answers to what’s really going on. https://t.co/i9UtbxJrt0

— Ronna McDaniel (@GOPChairwoman) July 25, 2018

As mentioned above, Twitter denies content discrimination is taking place. When asked why prominent Republicans like McDaniel are not showing up in the search autofill, Twitter directed Vice to this May 15th blog post describing the company’s new approach to creating “healthy conversations.” Here’s a portion of that:

Some troll-like behavior is fun, good, and humorous. What we’re talking about today are troll-like behaviors that distort and detract from the public conversation on Twitter, particularly in communal areas like conversations and search…

There are many new signals we’re taking in, most of which are not visible externally. Just a few examples include if an account has not confirmed their email address, if the same person signs up for multiple accounts simultaneously, accounts that repeatedly Tweet and mention accounts that don’t follow them, or behavior that might indicate a coordinated attack.

So what conclusions can we draw from all of this? First, Twitter agrees that Vice’s findings are basically correct, at least with regard to what is happening on the site. Second, when asked for an explanation, Twitter pointed to a blog post about shadow-banning trolls and then refused to answer subsequent questions.

At first blush, it sounds as if Twitter just called GOP chair Ronna McDaniel a troll. But the third point to consider here is that Twitter is rolling out some sort of correction to address the problem. That suggests they don’t believe McDaniel is a troll after all.

I’m speculating, but maybe what Twitter is trying to say here is that McDaniel (and other Republicans) were false positive in Twitter’s new troll net. Yes, they are being shadow-banned under the site’s new policy, but no they should not have been caught up in it.

Even if all of that is true, which I think is pretty generous, it’s still odd that only Republicans appear to have triggered false positives. Why didn’t any progressives get caught up in the machine-learning dragnet? Doesn’t the outcome itself suggest Twitter’s new approach targets and disadvantages the right, even if the signals in question aren’t based on content?

After all, the only reason anyone noticed this is happening is that Ronna McDaniel is a prominent GOP party chair. The more important question is how many ordinary conservatives have been caught up in the same false positive and are currently being shadow-banned by the site without a good reason to do so? If Twitter cared about winning back some confidence in its even-handedness it would release some information on what went wrong here and how many people were impacted. More importantly, it would be helpful to know if the fix Twitter is rolling out going to merely restore visibility to McDaniels and a few prominent Republicans or create a broader fix for other people currently being impacted?

Update 9PM: Kayvon Beykpour is the Twitter product lead who wrote about why the shadow-banning was taking place this morning. In response to one of his tweets, I asked how many people were impacted and whether all were Republicans. He responded a short time ago stating the number is “larger than the handful that have been mentioned” but still unknown.

“How many” is larger than the handful that have been mentioned (we are looking into quantifying but this isn’t an easy number to discern given how ranking works), and definitely not limited to a specific political affiliation.

— Kayvon Beykpour (@kayvz) July 26, 2018

Here’s my response.

We're only aware of the handful because they are prominent figures. How far does this go? Also, the fact that the impact isn't limited by political affiliation doesn't mean the impact isn't disproportionate by same. So far, it appears to be. Can you say that it's not?

— John Sexton (@verumserum) July 26, 2018

I’m taking him at his word that some people on the left got caught up in this too (more false positives) but that doesn’t mean the impact is proportional by political affiliation. Based on what we’ve seen so far, it seems the impact is disproportionately falling on Republicans for some reason. Granted, the number involved is unknown, but obviously, if you expand the disparate impact to 10,000 users or a million, that’s a significant partisan advantage.