Reflections on Birdwatch: Week 1
one week in. Is Birdwatch ready for prime time?
In an earlier post, we noted that one of Twitter’s not-so-secret projects is named Birdwatch.
Recently, one of the Newslettr’s editors was added to the Birdwatch open-beta program.
The project is not as complete as we had hoped.
Regardless, they have reached the “there are a lot of questions” stage of the project. Perhaps we will have some answers.
Writing the News
How does a Birdwatch comment start? A member of the program sees a Tweet, and writes1 a commentary on it. There is no queue2 of “high-volume Tweets that might need a Birdwatch comment” — at least not publicly.
After a week, I have one comment: it is impossible to write Birdwatch comments quickly while getting the tone right. After a few comments, there is a tendency to find you are writing “neutral commentary” that takes the form of “Mr. So-and-so is a stupid person who enjoys lying in order to mislead the public”.
On the other hand, the “I am a partisan pretending to be non-partisan in order to write partisan-leaning comments” tone is quite easy to get right.
In conclusion: you can’t rate Birdwatch commenters on an “Uber-like” 5-star system, where dropping to 4 stars means you’re fired3. That only rewards the consistently mediocre. And, well, Twitter is often mediocre without Birdwatch, there’s no need to rush more mediocrity.
Birdwatch Delayed is Birdwatch Denied
The saying goes that a lie can travel halfway around the world before the truth can put its boots on.
Twitter actually could fix this. Twitter could just say that Tweets only travel at 3000mph. To send a message from New York to Sydney on Twitter would now take three hours. I think our Aussie friends can survive with a delay in Elon Musk’s commentary.
But, that drastic a change might be a bridge too far. A far more elegant solution would be to simply have Birdwatch responses be ready faster.
Right now it takes 24-48 hours right now to get Birdwatch responses up. For a production-quality service, this needs to be done in a maximum of 6 hours. If you have enough workers, and the right priority-system, this shouldn’t be hard.
I assume it is slow because there isn’t a large enough army of anonymous volunteers who are willing to make complicated politically-charged decisions for no financial compensation.
I assume there is a second reason it is slow: unpleasant reviews tend to clog the system. An employee might be required to spend 10 minutes reviewing a Birdwatch comment on an unpleasant topic as part of their job duties. A volunteer will simply … not do that. And if there are enough unpleasant reviews, the volunteers will simply … not volunteer.
The Wheel of Content
The tweet above is from a content-mill account: it posts a historical image approximately once per hour. In the case above, it got the attribution wrong: the photograph is not of Salvador Dali, but of Alberto Cristini.
To a certain extent, Twitter should be paid by @historydefined if it is in the business of verifying facts. This is something of a “fire-and-forget” account: with a high volume of posts, and with no editors having any accountability, it is certain that errors will creep in.
If Twitter steps in and says “this caption is definitely wrong”, does that create a liability for Twitter to assess every other caption from this account as well?
The Israel-Palestine Situation
Dateline: Wednesday, May 11, 2022. According to France24, “Veteran Al Jazeera journalist Shireen Abu Akleh, a Palestinian-American who was among the network's most prominent figures, was shot dead Wednesday as she covered an Israeli army raid in the occupied West Bank.”
This set off a variety of partisans promoting their own versions of the truth.
There is only a minimal distinction between the partisan takes in Tweets, and the “neutral rebuttals” found in the Birdwatch queue. The Birdwatch replies to a Tweet by Rep. Andre Carson are an excellent example of “not very helpful”.
When there is nothing useful to be said, only those who have useless things to say will speak up. In other words: don’t expect Birdwatch to help matters here.
The Exaggerating Partisan
Brian Tyler Cohen (@NoLieWithBTC) is … well, a serial exaggerator. He has become a “Twitter personality” by spewing invectives and political opinions at break-neck speeds.
A lot of his tweets are problematic, particularly on abortion. (This pattern isn’t unique to Mr. Cohen. Most of the high-profile takes on abortion in the past week have been bad.)
Of course, it isn’t a tiny hateful minority of the country trying to “take away rights”. It is, by every account, a substantial majority4 of the populations of multiple states.
As another example, if Mr. Cohen says “let’s be clear”, it is a tell-tale sign he isn’t being clear, he is simply using half-truths to stir up shit.
And a final example, this one from a Matt Schlapp.
Presumably somebody informed him of something. However, it is a drop-dead certainty that Home Depot has at least one dishwasher available at a local store.
The only answer for zealots is tone-coloring5. Having the same Birdwatch comment: “Mr. Cohen is prone to hyperbole and partisan zealotry. You should not assume he or anyone else believes the literal truths of his statement.” repeated every day simply won’t work. It needs to be a symbolic indicator.
The Rumor Mill
Presumably, Mr. @Chicago1Ray is technically correct. Presumably somebody emailed such a statement. Possibly Chicago Ray himself. Doesn’t make the statement true.
There’s really nothing that “Birdwatch” can do to combat the likes of Chicago Ray. The only approach is escalating bans for violations of bright-line rules.
And “rumor-mongering” like this from a high-profile account needs a bright-line rule that leads to a block6. There is nothing Twitter can say here other than “you have to stop”.
Also, there is the detail that many of the people involved in “using Twitter to lie to the public” are quite litigious7. Perhaps Twitter does need to wait until it is de-listed to ensure that it isn’t trapped in legal hell.
There is one suggestion so obvious I assume Twitter is working on it: you can have GPT-3 do the comments. My best estimate is that type of software is 3-5 years away from working well enough to make a difference.
Well, if you follow enough people, you will see most of the super-viral tweets.
the Newslettr hated it when Uber ran its business that way, but at least it was fit-to-purpose.
Also, it is dishonest to the point of willful fraud to blame “the filibuster” for the fact that a motion which was opposed by a majority of senators failed. Then again, the Newslettr has retired from coverage of internecine Democratic messaging debates.
Tone coloring is “all the liberal propaganda is colored blue and all the conservative propaganda is colored red”, with a few dozen safety precautions on the side.
There are a few different ways of doing tone-coloring. Some of the simple approaches include: 1. Colored text. 2. Colored text background. 3. A pattern on the left side of the tweet. 4. the Checkmark system.
Perhaps with a “well, if we have a KYC relationship with someone who is paying us money, maybe we can unblock you” caveat …
Some of those people are above litigation, they simply point to an army. For some reason I doubt Mr. Putin would be happy with Twitter labeling a bunch of accounts as “zealous Muscovite partisans”.