Dec 122022

Elon Musk has tweeted extensively about Twitter and journalism, and what that can mean for the future. I know a lot of people like to complain about his approach to running Twitter, but I think there’s something to his ideas. I once thought that WikiTribune would be the bridge that leads to a new type journalism due to its wiki-style approach. I was clearly wrong, as WikiTribune lasted about 2 years and is now a social media site. Twitter, however, may be able to succeed where WikiTribune failed, assuming it can figure out a business model that keeps the servers on.

What Twitter excels at now

The value of tweets are naturally short-lived – sure there are people who will dig through years of old tweets to find something to complain about, but generally speaking nobody really cares about old posts on Twitter. So while Twitter isn’t a great historical record, it does excel at contemporaneous commenting, hence how live-tweeting took off almost immediately. This is particularly useful for developing stories, and covering major events in real-time. Thanks to Twitter, it’s possible to get live updates on what’s going on, and get them more quickly than sitting around waiting for the report to trickle back to news channels.

The problems with this are: 1) anyone can tweet anything, and all tweets look equivalent, 2) we don’t want social applications being the people in charge of determining who are the authoritative sources that should be listened to and who should be ignored, and 3) convincing people to pay for this. But there’s good news, Twitter already has tools that can address at least the first 2 (which hopefully makes it valuable enough to solve #3).  

But what about all the misinformation I keep hearing about? 

Like I mentioned in problem #2, we don’t want sites like Twitter in the business of determining correctness. Twitter is a place where people post comments and status updates, it’s not equipped to determine whether posts are accurate or misinformation, nor is it a specialty that we can reasonably expect them to have. So what we should be aiming for isn’t that every comment is either correct or fact-checked, especially remembering that Twitter’s potential as a news source is on actively developing or rapidly changing stories, where initial details can easily wind up being wrong. Instead, what we should be shooting for with Twitter as a news source is a place that rapidly self-corrects in order to arrive at accurate details faster than what we’ve traditionally seen from other channels. 

Tool #1 – Community notes

Twitter’s community notes feature has been around for a while now, but the basic idea behind them is this: users can leave comments on tweets to add additional information and context. Users then rate the notes, and notes marked helpful are appended to the tweet. Per Twitter:

Community Notes do not represent Twitter’s viewpoint and cannot be edited or modified by our teams.

In other words, community notes represents a fact-checking service, on Twitter, powered by users, that Twitter doesn’t directly control. Although Twitter can remove notes that violate the terms of service, and it’s possible to request a manual review if you think community notes on your tweet don’t add context or additional information, it appears the policy under Musk is going to be that things have to be clearly and unambiguously violating policies to get pulled. 

This is already live on Twitter, but is also something that can potentially be expanded. Particularly with regards to tool #2.

Tool #2 – Tweet metadata

Twitter has access to a lot of data about the tweets people post, as well as anything attached to that tweet (like photographs), although that can vary based on how users set their privacy settings. Obviously, Twitter knows when people tweet, but if you give Twitter access to your location, they know where a user was when they posted the tweet in question. That means there are instances where Twitter can tell that a tweet was posted from the place where a story is developing, during the period when it is developing. With access to photo/video metadata, Twitter can confirm the same about pictures and videos being posted. There’s likely also useful information to be gleaned from any other media that you can attach to a tweet, but right now the most common attachments are visual so let’s focus on them. 

Tool #3 – Context alerts on tweets

If you saw people talking about Covid-19 on Twitter during the pandemic, then you probably saw some sort of alert on Tweets directing people to government sites as a source of information, or maybe even saying that the claims in the tweet constituted misinformation. We still don’t want Twitter in the fact-checking business, but what interests me about this is the ability to automatically insert alerts on tweets. Instead of using this tool as a fact-checking system (which you shouldn’t trust Twitter to do, it has no expertise in anything other than running a web application where millions of people post stuff all day), it can be used to flag tweets coming from people who are tweeting about a developing story from the relevant time and place (using the tweet metadata we just discussed). This gives us a signal of quality – tweets coming from people who are actually where something is happening, as its happening are more likely to be correct and up-to-date than tweets from people who aren’t there, and thus are getting their information second-, third-, or even later-hand. 

Twitter’s still not declaring that some tweets are “Official Truth” (TM), just that “This tweet came from {location} on {date and time}.” They can add the same message to any photos, even if the tweet isn’t from on-site: “These photos were taken on {date and time} at {location}.” People reading the site can do with that information what they want, but calling attention to tweets that have first-hand content in them is going to create an incentive to share those tweets because they’re more likely to be accurate and unfiltered. Likewise, given tweets on developing situations, people will likely prefer tweets with first-hand accounts over other tweets – something you can search for (they may need to tweak the UI to make this easier, but it’s entirely possible to do). 

Putting it all together

So now we’re looking at a Twitter that’s automatically flagging on-location, at-time tweets and media for actively evolving situations, with community notes as an update/fact-check/correction mechanism. In theory, an app that serves as a reliable means of surfacing live, on-location reports on major stories would be valuable. We also say that journalism is valuable, and just look at how well that industry is doing these days. But this is finally the part where you may be able to get some value out of the updated Twitter blue. Part of the “this tweet came from the location of a major story as it’s developing” message could be that it’s limited to verified users only. So now if you want to contribute to live reporting, you needed to be a verified user.

For the sake of promoting reliability as much as possible, let’s say that users had to have been verified for some set period of time before their tweets get tagged. This prevents random people signing up for accounts, paying for verification, and getting promoted as a first-hand source. Even with manual review, there’s still too much potential for opportunistic shenanigannery if you allow verification followed by immediate flagging as a reliable witness. You can also use user metatdata as a potential signal of helpfulness in community notes. The fact that this also incentivizes signing up for verification early if you have any interest in being flagged as a trustworthy tweeter doesn’t hurt Twitter, who needs to start making money anyways. 

Where Twitter isn’t good with news

Before we get all excited over the idea of Twitter as a crowd-sourced news channel, it’s worth remembering that it does have limits. This is only valuable during the relatively small time window when a story is rapidly changing. Once that window closes, these tools and flagging mechanisms aren’t adding much because there’s nothing more to tweet about. It’s also not very useful for developing stories that don’t have a specific location component (think the Snowden revelations). There’s a lot of stuff that’s actively coming out, but it’s not geographically bound so there’s no value from tagging comments. 

Another area where Twitter isn’t ideal is for longer-form updates. The Twitter documents are interesting, but it’s a massive tweet thread to follow. 280 characters and a few tweets threaded together is great for quick updates because they’re generally limited to just the new piece of information. Massive information dumps are hard to read on Twitter itself – generally they’re better served by a blog or Substack post that then gets linked into a Tweet. 

There’s also the issue of people using a VPN to spoof their location and pose as someone on-site. The issue may end up being the deal-breaker for this whole idea. After all, if detecting when someone was using a VPN to fake their location was a straightforward issue to solve, streaming services would have implemented it already. This is another area where tweet metadata (in this case, a history of tweet metadata), image/video metadata (which wouldn’t be spoofed by a VPN), and verification can help assure people of accuracy. If a verified Twitter user is “posting” from somewhere significantly farther away from where they live, that they don’t have a history of tweeting from, then that’s an indication they may not actually be tweeting first-hand. This kind of dragnet obviously also catches any professional reporter who travels to cover developing stories, but for an initial implementation, I think limiting these “on-location” signals to people who Twitter has high confidence lives and works there is the safe choice. They can always come back later and add to their “signal detection” and “spam filtering” for on-location tweets in order to accommodate that particular edge case. 

These ideas are all built around the idea that on-the-scene tweets have a higher probability of being accurate and useful than other tweets, that would have to get their information from some other source and thus be second-hand at best. It’s entirely possible for these tweets to still be wrong (at best) to intentionally false (at worst). And that’s ignoring the fact that with an actively-developing situation, there’s a lot of unknowns and things change quickly. That’s why I stated earlier that I wanted something that could rapidly self-correct. Twitter declaring a source of a tweet “The Truth” (TM) only works if they’re right every single time. That’s why the community notes feature is so important – it’s a correction mechanism that lives on the original tweet so everything sits in context. 

Elon Musk is clearly optimistic about Twitter as a collective knowledge-share and media application. In some aspects, he’s right – Twitter can take an otherwise unknown story and be the amplification needed to give it the significance it deserves, sort of like how the video of the Ahmaud Arbery murder going viral online finally drove prosecutors to charge the McMichaels and Bryan with the homicide. But there are some scenarios, like actively-developing stories, that play right into Twitter’s wheelhouse. With tools it already has, Twitter also has the ability to highlight first-hand posts to help boost posts from the people actually there. Is this valuable enough to get people to start paying Twitter $8 a month? I have no idea. I do think it has enough potential that it’s worth trying to find a business model to support it. 

 Posted by at 11:45 AM