Custom feeds and election defense scenarios

Custom feeds in the ATmosphere are great for getting the word out about news and calls to actions, and keeping people informed about the situation. As I'll discuss below, there are several important election defense scenarios where they can be useful. Here are some of the ones I've set up so far.

The different kinds of feeds are discussed in more detail below.

ATmosphere social networks like Blacksky, Bluesky, etc are all relatively small ... but you can follow the links and see the contents of these feeds even if you don't have an ATmosphere account. So the information is potentially useful to everybody. And, these networks are big enough that in the US as a whole, and at least some states and cities, there's enough of a critical mass of non-profits, state / county / city government agencies and officials, and media organizations for this approach to work.

Freedom Trainers' Election Defense Scenarios has several situations of where the custom feeds could be very useful. For example, the section on Disinformation Flood: Wrong Place, Time, Day discusses the risk of "that allies of good will hear rumors of ICE at polls and quickly pas\ them along on social media, thinking they’re doing people a service when they’re just spreading falsehoods intended to keep people from polls."

A few other Freedom Trainers' scenarios where the various feeds are also likely to be useful in relate to getting the word out rapidly. 

In their discussion of responses, Election Defense Scenarios focuses a lot on "phone-tree"-like diffusion of information in Signal groups, group chats, email lists etc. Those are very important channels, and much more controllable than social media. Still, social media can complement these mechanism in several ways:

The goal of a disinfo-resistant feed like Election Defense is to give people information and narrative framing they can be fairly confident in -- and to ensure that any incorrect information is taken down fairly quickly and replaced with accurate information.

These feeds start with relatively-small number of fairly trustworthy sources, refined by simple manual curation (adding or removing specific posts to the feed). Back in 2020 I worked with Shireen Mitchell of Stop Online Violence Against Women on a project using a variant of this approach for a Twitter-based news aggregator, and in the thousands of posts during the week surrounding the election I only saw one dodgy post get through. By contrast, Twitter search or trending hashtags had a lot of speculations and rumors disguised as facts, inaccurate information, and even some flat-out disinformation.

As discussed below, an obvious potential next step to further improve the disinfo resistance of these feeds would be to integrate moderation services (aka labelers).

Feeds with posts from a much broader list of sources will often feature news that doesn't make it to disinfo-resistant feeds, which have a much more limited set of sources. The downside, though, is that with a broader list there's much more of a chance that somebody is unintentionally amplifying inaccurate info, reporting a rumor as fact, combining accurate facts with misleading framing, or intentionally spreading disinformation.

Filtered feeds can still be useful in several ways:

US Election Chatter doesn't attempt to filter sources at all (other than leaving out posts from accounts on blocklists like Blacksky's Green List). As a result, there's likely to be a lot more inaccurate information and disinformation than a filtered feed -- as well as plenty of shitposting and scamming. Unfiltered feeds are still potentially useful, for example to identify what disinfo is circulating (and what sources are spreading it). On the other hand maybe it's so noisy that it shouldn't even be associated with this account ... at the very least, a warning needs to be prominently displayed..

Bluesky's own "trending" topic-specific feeds on voting (generated on the fly when there's a lot of discussion of the topic) are also in practice unfiltered. While their algorithm selects the most popular posts, this doesn't necessarily help so much in practice: many popular resistance grifters sometimes post dodgy stuff (hey it gets the clicks). Similarly, third-party feeds (like Bluetrends' Voting Restrictions or Voting News) are also unfiltered.

Activity is likely to ramp up significantly as we get closer to Election Day, so more geographically focused feeds are useful complements to nationally-focused feeds, which may get overwhelming.  WA Election Defense has a limited number of sources, so is fairly disinfo-resistant; AZ election defense has a somewhat larger number of sources (at least for now) including a lot of media, and I don't know much about the landscape there, so I'd classify it as a filtered feed.

While it's possible for an outsider to create a basic feed (as I did with AZ election defense, starting with the Secretary of State and a couple of County Recorder offices I found with some quick searching, and leveraging existing lists of Arizona media and voting rights groups), deeper local knowledge is likely to lead to a better feed. WA Election Defense, for example, intentionally leaves out a few local media outlets that I know enough about to view them as untrustworthy in an election context, 

Ideally, these will be driven by people or groups involved in different states' election protection coalitions, or state-specific "nerve centers" (as discussed in Act Free to Be Free). Not only do they have the best understanding of the landscape, they're the best equipped to make the call in situations where urgent information needs to be highlighted (for example by pinning posts).

In principle, feeds focused on specific counties or cities could also be useful. In practice, though, it's an open question whether there are enough trustworthy accounts in any of the key locations to make this worthwhile.

Voter suppression and challenges to the election are likely everywhere in the country -- and especially likely in states where there's a pattern of it, including Arizona, Michigan, Pennsylvania, North Carolina, Georgia, and Ohio. So, more state-specific feeds would be good!

In the short term, I'm planning on working directly with people and groups in a few more early-adopter states to get feeds in place -- if you're interested, please let me know.

After that, depending on the level of interest, it might make sense to develop training materials. Using the Graze feed building platform, reusable "custom nodes" can help here, but even so it's a multi-step project that's not all that accessible to inexperienced people (and this is not a scenario where using a simple prompt and relying on AI feed generation makes sense). If we get to that point, it will certainly be a good problem to have ... but still something we'll need to think through.

The more credible sources there are for election-related information in the Atmosphere as a whole, the more likely it is that people will see accurate information. That's true in general, and it's true for custom feeds as well.

One thing that would help here is tailored outreach and onboarding material that people in various locations can use to approach key officials and organizations. Since most are already posting to Facebook and/or Xitter, this includes helping them leverage cross-posting.

Community-run moderation services like Blacksky and Medsky are another distinctive ATmosphere technology, providing additional levels of moderation by tagging posts and profiles with arbitrary labels. Blacksky's labels, for example, include Digital Blackface (profiles pretending to be Black people, a common disinformation technique) and Synthetic Media as well as White supremacy or Antiblack rhetoric, Misogynoir, and others. These aren't things that we people to be seeing on our feeds, so it would be great to leverage the existing effort to filter them out.

One challenge doing this is that Graze (the feed-building I've been using for the feeds so far) doesn't currently support integrating labelers. So, either the software needs to be extended, or we'd need to shift to another feed-building infrastructure. So this is still a "Details TBD".

Once we have an infrastructure in place that supports integrating labelers, Medsky Moderation's disinfo labels highlight another opportunity: new labelers that focus on election-related disinfo. Doing this effectively will require a trained team of people (as well as some software support). It doesn't necessarily have to be that large a team; there's a lot of value just from focusing on the relatively-small number of posts on the disinfo-resistant, and maybe media-centric, feeds. Still, it's not yet clear whether the rescuers to do this will emerge. We shall see.

TBD! I'm circulating the ideas and examples now to gauge interest in a couple of dimensions:

So please let me know what you think!