This is, more or less, for the benefit of the community at large. Not everyone pays deep attention to the inner workings of Stack Overflow, and it may be disconcerting to some to see moderators are considering a strike (see also Moderation Strike: Stack Overflow, Inc. cannot consistently ignore, mistreat, and malign its volunteers). I figured I would share things from my perspective (as an elected moderator), and not from the broader moderator team (not all of whom are considering a strike). Other moderators may add their thoughts as they see fit.
I should start by saying that the Code of Conduct change really hasn't played any role. The rollout has been known and well discussed amongst all Stack Exchange (SE) moderators for some time now and Community Managers had a community discussion in advance. There has been some speculation this played a role, but the SE moderators I know have not really been bothered by it in any significant way. In fact, that rollout stands in stark contrast to what has happened in the last week.
A history of Artificial Intelligence policy
In late November of 2022, a company called OpenAI launched what we now know as ChatGPT
On Monday, OpenAI announced a new model in the GPT-3 family of AI-powered large language models, text-davinci-003, that reportedly improves on its predecessors by handling more complex instructions and producing longer-form content.
That, combined with the also recently launched AI image generator called Midjourney, sparked what could only be called a frenzy of new uses for AI-generated content. ChatGPT, obviously, was far more impactful to Stack Overflow, because it could answer coding questions. People began to openly game the system as a result
So I started a new stackoverflow account and I am plugging random questions without answers into https://chat.openai.com/chat and pasting the answer.So far, after 9 answers in 1.5 hours it has 1 accepted and 3 upvotes and a reputation of 62...
6 accepted answers, of the 26, 11 upvotes, 5 downvotes. I am not checking the answers in any way. I'll give it a rest for now and see how we do tomorrow...😄
That latter comment (about "not checking the answers in any way") exemplifies the problems the moderation team was starting to see in droves. People were coming out of the woodwork and, in some cases, becoming coding "wizards" instantly. ChatGPT was printing reputation for some people, and that was the point. Thus we crafted a new policy for the site
The primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce. There are also many people trying out ChatGPT to create answers, without the expertise or willingness to verify that the answer is correct prior to posting. Because such answers are so easy to produce, a large number of people are posting a lot of answers. The volume of these answers (thousands) and the fact that the answers often require a detailed read by someone with at least some subject matter expertise in order to determine that the answer is actually bad has effectively swamped our volunteer-based quality curation infrastructure.
This was not crafted solely by moderators. We directly involved the Community Management Team, who were not only supportive, but helped us in many ways to get this new policy driven home to the community at large. These efforts included sharing posting behavior analysis and creating a new Help Center page
Stack Overflow is a community built upon trust. The community trusts that users are submitting answers that reflect what they actually know to be accurate and that they and their peers have the knowledge and skill set to verify and validate those answers. The system relies on users to verify and validate contributions by other users with the tools we offer, including responsible use of upvotes and downvotes. Currently, contributions generated by GPT most often do not meet these standards and therefore are not contributing to a trustworthy environment. This trust is broken when users copy and paste information into answers without validating that the answer provided by GPT is correct, ensuring that the sources used in the answer are properly cited (a service GPT does not provide), and verifying that the answer provided by GPT clearly and concisely answers the question asked.
The broader policy across the network has been that each SE site got to decide its own course of action (here's a list of all site ChatGPT policies). SE declined to make a global policy. Stack Overflow's position has not changed since that time.
Ground Conditions
The moderation team uses many userscripts (JavaScript scripts written and maintained by fellow moderators, and running under tools like TamperMonkey) to aid in moderation. Sometimes we need a tool and we write one. Other times we extend the existing site functionality. One of them lets us craft our own "stock" moderator messages. We wrote one to inform and/or suspend users caught using ChatGPT.
Another userscript allowed us to run posts against AI detectors like Huggingface, which we and other community members used to detect ChatGPT posts. This system turns out to be important to the issue at hand.
If community members flagged a post, and moderators agreed, the user would get a mod message and/or suspended. The volume we handle every day in some cases meant a user would get a suspension for as few as one suspected post. In hindsight, this isn't ideal (and I won't say moderation mistakes weren't made). And, of course, many users would merely dispute the findings without much in the way of evidence (which is a lot of message replies we get). In some cases, those replies would be ignored, commonly because the reply is rude/abusive and evaluated to be frivolous, but in rare cases because the reply was missed (the tooling for moderators in this area is ... poor, particularly as SO's volume). This, in turn, led them to the Contact Us form, and Community Managers (which is the official way to appeal any moderator action).
That brings us to this week.
What we have here is a failure to communicate
The normal method to deal with an issue like this is Community Managers will reach out to us in back channels to question the action(s) taken and/or tell us we need to do something different. We can, and do, reverse course when asked. We're all adults, and the atmosphere in most moderator channels is collegial. We are, at the end of the day, volunteers, not employees.
On Monday (a major US holiday), moderators were informed, via pinned chat messages in various moderator rooms (not a normal method), to view a post in the Moderator Team that instructed all moderators to stop using AI detectors (as outlined above) in taking moderation actions. Had it stopped there, it probably would have not been as controversial as it has been. It went considerably further than that, which meant that moderators were all but prohibited from removing content we suspected was AI-generated. Unfortunately, as the Team is private, I cannot share details here.
The fact that it was private was a problem in of itself, in that we were being told to enact a policy that was not available to the broader community. This also undercut any previously supported community policies without any discussion or even the ability to communicate the policy change. After some sites started making public Meta posts out of necessity, Philippe (Vice President of Community) posted about it on Meta Stack Exchange
In order to help mitigate the issue, we've asked moderators to apply a very strict standard of evidence to determining whether a post is AI-authored when deciding to suspend a user. This standard of evidence excludes the use of moderators' best guesses based on users' writing styles and behavioral indicators, because we could not validate that these indicators are actually successfully identifying AI-generated posts when they are written. This standard would exclude most suspensions issued to date.
This differs greatly from the Teams guidance (which, again, we're not allowed to publicly share given its delivery method). To say we're confused would be an understatement. Worse is that CMs have come into the SO mod room to tell us we were violating the Teams version of rules. The only consistent part of all the guidance we've gotten could be summed up as "Delete/suspend for AI posts, just don't say it's for AI posts". In other words, we basically have to vacate the ChatGPT rule, but without saying we're vacating the ChatGPT rule.
On Thursday, Philippe held a meeting in Teacher's Lounge. He admitted the rule should not have been rolled out like that and explained what has precipitated this urgent change (I cannot comment on this point so please do not ask). Their reasons make sense, and explain why SE Staff (himself included) is currently stressed. But he offered no guidance to resolve the split in guidance, and Community Manager comments after he left seemed to reinforce the Teams version, which is far more restrictive on moderators.
On top of all this was a concern that another proposed (and deeply controversial) change that could be highly impactful to the community at large (and drastically increase moderator work) would be rolled out on top of all this. Per Thursday's meeting, this second change was postponed. It's unclear whether it is permanently shelved or merely delayed to a less stressful time. At this time, given the postponement, we cannot discuss this change, since there are no public announcements for it. It might be moot for this, but it still lurks in the back of the minds of most moderators. It is not, however, part of the current proposed strike.
What do striking moderators want?
There's a number of issues at play. Some of them merely need an acknowledgement and public repudiation. Others need open comment from SE Staff so we have clarity in what we're expected to do.
The ChatGPT rule was a community-crafted and highly supported rule. Staff should not make large scale changes to these rules solely using back channels (chat rooms and Mod Team) that are never meant for public consumption.
Moderators were being asked to handle flags based on a rule that was not public. Indeed, everything on the site still indicates the old policy is still in effect. As such, we were still getting flags that we could not handle consistent with that policy. The community would have noticed that difference eventually. The few moderators who braved posting about the policy change on their Metas before the public announcement were risking their diamonds and/or access to the Mod Team by doing so. That is deeply unfair to moderators. We have had to go to great lengths to not leak private communications, which has compounded our frustration.
Philippe's post poured fuel on that fire with things like
Through no fault of moderators' own, we also suspect that there have been biases for or against residents of specific countries as a potential result of the heuristics being applied to these posts.
This sounds a great deal like an accusation of racism. Nobody wants to be accused of that, inadvertent or not. The private Teams post has a better explanation (with metrics) that avoids that conclusion but, again, it's private. Again, I strongly suspect Staff is merely distracted with larger issues, but that doesn't make this any more acceptable. Moreover, this is just one example of where accusations have been levied at moderators in vague ways that are incredibly hard to take as good-faith discussion on the part of Staff. At best, their words are ill-considered. At worst, they reflect an adversarial view of moderators and the community.
Staff has been maddeningly vague on policy (as I've noted, public and private policy differ) and providing instances of what they feel was done incorrectly. They want us to stop using AI detectors as our sole source of action. That's fine and good, but what do we do with someone who posts a dozen (or more) long-format answers in two hours? They're clearly not writing those posts themselves, just bulk answering questions as fast as they can copy-paste them back and forth. Under no policy is this acceptable, but how are we supposed to deal with that? Do we use the new plagiarism tool without direct evidence it was plagiarism? That seems like the wrong tool for the job. Internal guidance suggests removing them as "low quality", but they aren't "low quality" (which is kinda the point of ChatGPT, in that its goal is to make high quality content on-demand). This has resulted in moderators not feeling comfortable in removing these posts and/or issuing suspensions that explicitly say they are violating the ChatGPT prohibition. This is a de-facto reversal of the ChatGPT policy as things currently stand.
In conclusion
SE Staff has come a long way from the great Monica debacle of three years ago. In fact, they've been doing some very good things in this vein (I have repeatedly sung the praises of the Staging Ground project, which has had excellent community-to-staff engagement). It's greatly disappointing that we have, in the space of a week, regressed to a point where Staff is firmly at odds with the community that represents their product. None of the moderation team likes having to threaten a moderation strike to get answers, but here we are.
Moderators involved in the strike have already slowed our flag handling. The official strike starts Monday, Jun 5, unless SE Staff can come back and deal with this in a fair and just manner.
