Comment by BryantD | Hacker News Reader

Comment by BryantD | Hacker News Reader

I put around 15 minutes a day into Birdwatch for a couple of months before the purchase was finalized. This is purely anecdotal, so take it with a grain of salt.

The requirement for agreement seemed to work well at preventing weird factchecks from anywhere on the political spectrum. I saw a fair number of people trying to use Birdwatch to argue with each other, bad faith factchecks, and so on. None of them made it to general visibility.

I wrote 45 notes. I tried very hard to keep them unbiased, but I'm human and I have strong political opinions. 5 of them wound up approved. I suspect the requirement for agreement tended to keep anything that's more than a little divisive from getting approved; no evidence for this, though. I'd be curious to know what the average percentage was for active users.

There are way more Birdwatch notes getting written right now. A lot of them are terrible quality. However, they mostly aren't getting approved, so I think the system is working as designed.

It's a really smart idea. Look for agreement between people who often disagree as a signal of truth. Like you point out, it's probably a stronger signal of less divisive, less polarizing info than truth. But still decentralized and not biased right or left (biased center, effectively.)

I think it will only work if they can keep the bots out. Otherwise it will be gamed like everything else.

https://en.wikipedia.org/wiki/Argument_to_moderation seems like the envitable outcome of this system.

I don’t see that.

What parent is suggesting is that only the subset of things which are agreed by two parties who, demonstrably, disagree on most things, are approved and noted.

You’re not going to get “a fetus becomes human life at 4.5 months” out of such a system

Interesting example. Because the hardcore on both sides have zero percent agreement on that. To some people a fetus isn’t a person until intended birth, so how would a birdwatch context work on something nonfalsifiable like this?

Presumably it would not approve anything, and thus have nothing to say, which is probably the best outcome.

How then do you decide on the sets of things to allow it to moderate or not? That sounds like a bootstrapping problem since it is a question people will disagree on.

This is my concern as well. The idea here seems to be that the thing that offends the least must be fact.

This behaves very differently for falsifiable vs. non-falsifiable information, but the end result is not something in which I'd place much stock either way.

Edit: to be clear, from Twitter's perspective, this is excellent, i.e. this is precisely what a revenue-motivated entity would want. It's the rest of the world that I'm concerned about.

I would be worried about users falling off seeing their notes have ~10% approval rate, but it doesn't seem to bother you at all. I wish I had smarter question to ask about it, but... how come?

That's a great question, actually. Factors which I think play a role:

1. I'm not that heavily driven by ego in this kind of context. I know that approval is largely a function of how many people bother to vote on my notes.

2. The satisfaction I got from getting a note approved on one of Elon Musk's tweets was sizable. It is literally one of the most-read pieces of text I've ever written, which is kind of wild.

3. I define success here as "good notes are approved," rather than "my notes are approved." I'm used to casting a wide net and being happy at a few successes. I'm a hiring manager, so I have to know how to be happy with a low success rate.

4. I also know that other Birdwatch contributors were reading my notes, and I tried to write them in such a way that people can easily research and draw conclusions. (Not always easy, since I'm opinionated.) So even a note that isn't approved might do some good.

Tweets are near real time so how do they keep up with them when it's a consensus system that could take a lot of back and forth? The period between when a tweet is sent and when its reviewed is posted could be enough time to indoctrinate a lot of people, no? I think the average tweet impression over time graph looks like a reverse exponential.

I have seen people complain about bad check today (notably Radley Balko had one), so it seems like some bad faith checks are starting to creep in.

It is vulnerable to numbers, I think, for the same reason people believe misinformation. If a check looks plausible, some number of people will say it's correct without double checking.

I didn't see anything right-wing about Radley's fact check other than his assertion that it must be so.

For those who haven't seen it, Radley tweeted[1] that of 760 million Subway riders, there were only 8 murders, so it is intrinsically safer than riding in a car.

The birdwatch message retorted that he was conflating 1 year of Subway death statistics against the lifetime vehicular fatality rate, and implied that his stats weren't an apples to apples comparison. Birdwatch recalibrated his statistic using (the generally more meaningful) "miles traveled" and determined that riding a mile on the New York Subway was ~1.8 times riskier than that same mile in a car, which is also probably wrong, but right if you use _the data he provided_ against "miles traveled" vs overall ridership.

[1] https://mobile.twitter.com/radleybalko/status/15888941041333...

The Birdwatch note also assumed he was using lifetime risk of dying in a car crash, which he wasn't (I double checked).

I am not 100% sure that miles traveled is the most meaningful statistic, FWIW. Wouldn't it make more sense to use hours spent on each form of transportation?

Appears that that Birdwatch was removed. I assume this was from people rating it after it appeared and the annual versus lifetime error was found.

Deaths per X miles traveled is the standard metric. Deaths per hour would be a very strange way to compare forms of transit given that the purpose is travel not wasting time.

Only if everyone can always optimize their travel choices by time, which obviously we can’t or all travel in Manhattan would be via helicopter.

A good habit for "fact checking" groups would be to NOT give the "right answer" in a case like this. Pointing out the error (yearly vs. lifetime) is good enough. No need to try to be an ad-hoc social scientist giving quick answers to questions that are probably not easy - or at least, if you want to do it, do it on a regular account.

Reply on news.ycombinator.com