I think people are missing something with Bluesky's "consumer choice" model of moderation. A lot of harmful content has harm beyond *you* seeing it. 1/

There are plenty of examples. Child porn is exhibit A, of course. But the same is true for scams; propaganda; promotion of hate along racial, gender, and religious lines; doxing; glorification of violence (especially hate-based violence); promoting targeted abuse; and disinformation. 2/

That kind of content causes societal harm even if it's only seen by people who want to see it. But that's not what will happen. No automated moderation is perfect; I still have dozens of examples of Twitter nazis saying egregiously racist things in ways that aren't algorithmically detectable. And worse, there's something pretty fucked up about getting people onto a platform and saying, "Hey, don't like racist shit? Just pay a monthly fee to get that labeled for you." 3/

Beyond that list, there's a whole lot of "conspiracy to" content. Even if I don't personally see a bunch of neo-Nazis organizing their new hate campaign, torch rally, synagogue vandalism, kristallnacht, or pogrom, I don't want to contribute content to a service where that stuff exists, whether or not I personally see it. I hope you don't either. 4/

I get that both libertarian and consumerist dogmas can make "we'll just label the bad stuff" sound reasonable. And maybe that can work for matters of taste, like people who don't want to see nipples or swears. But it doesn't solve problems of safety. That's a platform responsibility. 5/5

@williampietri I want to wade in because I—I hope obviously!—agree about the content points, but some of the surrounding things are live questions for me, and for me really interesting if anxiety-laced ones.

Please forgive me for nesting this stuff in your replies—it seemed more polite than pseudo-QTing.

[1/x]

@williampietri But first, we're talking about future apps on ATProto here, right? Because Bluesky the platform/app runs vanilla central mods, boots nazis, harassers, etc., with ~usual levels of accuracy.

But the protocol is unbundling things we usually bundle, like moderation services, identity management, data storage, and indexing* and idek what else. So the things that happen inside the central opaque services are potentially scattered around in full* view.

(* I know nothing)

[2/x]

@williampietri The "we label and let you hide hate speech" thing makes my whole nervous system turn inside out—but also labeling is a backend thing that *could* permit multiple layers of action on top. Maybe in usefully cross-cutting ways.

And on Mastodon, we're all already using a protocol/app combo that also powers nazi servers/harassment nodes/CSAM distribution—we just don't see it much bc fedi has done a lot of gradual institution-building that walls it off from mainstream views.

[3/x]

@kissane I’m confused to hiding the hate speech. I’ve seen several people attribute “hiding” key word filtering is “hiding” muting is hiding as is blocking. I’m just unsure why it’s viewed differently on Atproto? I’m obviously missing something as several people have made this point. Mind helping me understand?

@Damon I think the troubling idea is having ONLY user-level muting/hiding moderation, which means being unable to boot bad actors “out” of a “place” in the way a server mod here can.

To me, a widely used labeling-and-hiding moderation layer run by, say, a civil society org with credibility in specific areas could function in much the same way as server blocs on fedi that defederate from asshole servers.

But there are still differences and I don’t know how they’ll play out.

@kissane Its not just user level but server level particularly for illegal content. Other PDS can be blacklisted. But, yes there’s a lot that needs to be clarified. I wasn’t happy about them opening up without more effective tooling and clarification around the tooling

@Damon By “user-level” I mean “actions taken by a user,” rather than a server mod (or weird semi-equivalent).

My *personal* reaction to “just label everything and let people choose what to see” is viscerally negative, but I think it’s more complicated that than along multiple axes.