Turn over moderation duties to AI

[annk]

The issue with moderation on this forum is not necessarily that it is performed by humans but rather that:

1) often a _single_ human decides (collective discussion is referenced but I do not see this reflected in moderator messaging).
2) there is no means to contest a moderator action : they decide, the decision is final, even if questionable or based on clearly incorrect assessment or an overly broad ‘rule’. There is a fine line between moderation and censorship.

...... Even with a positive response score (which would denote positive contribution to this community) on my profile, I will likely get banned for this inflammatory content (which would prove the point).

These points are incorrect.

1) A single moderator decides only in cases that are very clear. For example, when a spammer is posting, a moderator won't wait to consult others before taking action. A clear case of insults (such as "you're a loser" or similar) might be handled by a single moderator, but often also these are discussed. The reason for the discussion in these cases is not because the rule involved is unclear, but rather because in cases where a user already has been moderated for violations, discussion is often useful to determine how much the moderation should be escalated.

Even in cases where one moderator acts, everything that moderator did is visible to all other mods and admins. Moderators document everything they do, but even if they didn't, the forum software leaves clear traces.

You don't see the collective discussion referenced in the moderator message because of the way the forum software works. The semi-automatic moderator messages need to be sent by a individual. The moderator who has the time to do it takes care of the administrative housekeeping.

2) There is a very clearly explained way to contest moderation.

https://macrumors.zendesk.com/hc/en-us/articles/201295026-What-if-I-disagree-with-moderation-of-my-posts

https://macrumors.zendesk.com/hc/en-us/articles/201260827-How-are-moderation-errors-handled

Users contact us about moderation when they don't understand or agree with it, and moderation is overturned and apologies issued when we see that a mistake has been made. Luckily, that's not often - exactly because of the discussion that happens before most moderation is done. This channel to contact us about moderation exists because we're human and we know that mistakes can happen. When users contact us to question moderation, the admins do a review of the moderation that was done by the moderators.

The line between moderation and censorship is clear because users on this private site agree to rules as a condition of membership. Users can disagree with each other, post critical comments about Apple or other companies, questions editorial decisions made here (in the appropriate forum section), etc. We don't censor opinion; we simply expect users to follow the rules they agree to. Comments are only moderated if they break rules.

Finally, you won't be suspended for your comment, which doesn't break any rules. I do however think it's important to correct the misinformation in your post.

 

[za9ra22]

The issue with moderation on this forum is not necessarily that it is performed by humans but rather that:

1) often a _single_ human decides (collective discussion is referenced but I do not see this reflected in moderator messaging).
2) there is no means to contest a moderator action : they decide, the decision is final, even if questionable or based on clearly incorrect assessment or an overly broad ‘rule’. There is a fine line between moderation and censorship.

The whole mechanism seems arbitrary and I have found moderators here to be very sensitive… jokes have been made about an AI being unable to distinguish between jokes and serious statements but the same could be said of some moderators

Sites like StackOverflow have a community moderation platform which makes assessment of any post a matter of multiple site users, thus making the entire process much more objective.

Even with a positive response score (which would denote positive contribution to this community) on my profile, I will likely get banned for this inflammatory content (which would prove the point).

In theory at least, moderator action is actually performed by 'group decision', rather than an individual, so that point doesn't seem to apply.

Also, since the function of moderators is to control the 'tone' of the site, their decisions really can't be subject to debate, and neither can they be free from a certain amount of censorship - by the nature of things, removing an 'offending' post is censorship sin action by default.

One assumes, from your logic, that if you are not banned for your post on this subject, it also proves a point - just a different one.

 

[adib]

A perfect role for artificial intelligence; discussion forums moderation. That would be much better than the way it works now, with fallible humans making decisions based on their fallible judgements.

Ever heard of Microsoft's Tay?

 

[cmcbhi]

Why not?
What downsides would there be?

Another good thing; it would be real-time. Posts breaking rules would not be first posted and then reported/deleted. Much tidier and less frustration.

I guess you don’t know much about AI.

 

[turbineseaplane]

Along the lines of moderation and AI

Do we think it might be better at weeding out potential bias from moderators?

It's only natural, as moderators are human, and I've noticed (or at least have some strong suspicions about it) that personal bias on what gets moderated or not definitely can creep into the decision making process.

I wonder if AI might be more balanced and simply apply the rules of the forum re: off topic and attacks and better deal with when political terms and memes and attacks sneak into posts?

 

[Apple fanboy]

Along the lines of moderation and AI

Do we think it might be better at weeding out bias from moderators?

It's only natural, as moderators are human, and I've noticed (or at least have some strong suspicions about it) that personal bias on what gets moderated or not definitely can creep into the decision making process.

I wonder if AI might be more balanced and simply apply the rules of the forum re: off topic and attacks and better deal with when political terms and memes and attacks sneak into posts?

No thanks. AI is not going to be less biased. It’s biased based on the coding it’s written with.

Plus I think AI will not understand context of a post better than humans.

 

[turbineseaplane]

No thanks. AI is not going to be less biased. It’s biased based on the coding it’s written with.

Plus I think AI will not understand context of a post better than humans.

A fair concern re: coding, I hear you on that

What I wonder is if an AI could be tuned better though -- if it just has exact rules to work off, it'd be better to not entertain context... or at least really dial down how much that is considered

Bias of humans is even worse because it the human understands context just fine, but then purposely, even if just subconsciously, refrains or alters their behavior from simply applying rules of "post is off topic" (as an example) as the human in question is "ok with what was said, even if it technically should be moderated", etc.

Bias is really pernicious that way

Know what I mean?

An example that appears somewhat frequently due to the current climate of things are political terms. There really is no context where things like "woke" or "snowflake" or "maga" etc should be allowed on the forum as they are inflammatory and derogatory ways of talking about things and all their usage does is inflame discussions and solidify polarization....

In short -- those type of terms, from all political angles, just pisses people off and people using those terms know exactly what they are doing.

 

[Apple fanboy]

A fair concern re: coding, I hear you on that

What I wonder is if an AI could be tuned better though -- if it just has exact rules to work off, it'd be better to not entertain context... or at least really dial down how much that is considered

Bias of humans is even worse because it the human understands context just fine, but then purposely, even if just subconsciously, refrains or alters their behavior from simply applying rules of "post is off topic" (as an example) as the human in question is "ok with what was said, even if it technically should be moderated", etc.

Bias is really pernicious that way

Know what I mean?

An example that appears somewhat frequently due to the current climate of things are political terms. There really is no context where things like "woke" or "snowflake" or "maga" etc should be allowed on the forum as they are inflammatory and derogatory ways of talking about things and all their usage does is inflame discussions and solidify polarization....

In short -- those type of terms, from all political angles, just pisses people off and people using those terms know exactly what they are doing.

I understand what you are saying. But AI would moderate your post as it uses those derogatory terms. Where as the human moderators we have now will not (see how context DOES matter a lot!).

I think the system we have now is a lot better.

 

[turbineseaplane]

Where as the human moderators we have now will not (see how context DOES matter a lot!).

That's where I disagree

Allowing things like inflammatory terms to not be moderated is not good ... no matter the context of it

That's why I gave that example above ... we don't want things that should be moderated out to "not be", as it just furthers the polarization on topics

I just consider derogatory terms, in any context, to be off topic and only serving to irritate people

Again - people know what they are doing when they use such terms
It's to irritate people -- on purpose

 

[turbineseaplane]

@Apple fanboy

Let me come at it from a different angle
What context would you say it's appropriate and within forum rules to use the term "woke"?

(just randomly picking "woke" -- could use a lot of terms here as an example -- moderators, please bare with me as I explore this briefly to understand the other users perspective)

 

[bousozoku]

@Apple fanboy

Let me come at it from a different angle
What context would you say it's appropriate and within forum rules to use the term "woke"?

(just randomly picking "woke" -- could use a lot of terms here as an example -- moderators, please bare with me as I explore this briefly to understand the other users perspective)

"She woke up, got dressed, and we took her to lunch on Mother's Day."

I saw too much of the other meaning the other day on the iPad advert threads. I didn't see anyone moderating any post that contained "woke", though.

 

[turbineseaplane]

"She woke up, got dressed, and we took her to lunch on Mother's Day."

I saw too much of the other meaning the other day on the iPad advert threads. I didn't see anyone moderating any post that contained "woke", though.

ha!
Well done -- excellent example ... I should have thought of that one straight away

I do think, though, AI could be very easily trained to identify that usage vs "calling behavior woke", which of course is the negative connotation with political undertones.

I mean .. I think AI could? I would hope?
lol

 

[turbineseaplane]

I saw too much of the other meaning the other day on the iPad advert threads. I didn't see anyone moderating any post that contained "woke", though.

Yep - lots of various terms getting thrown around that I just don't understand why were allowed.

It wasn't from any political side .. not a right vs left thing...
To me it should be a "let's not increase polarization" angle

When people resort to those terms - they are doing it only as a shortcut and a shot at "the other side", whatever side that might be

I thought that was half the reason (or more) that PRSI went away?

 

[Apple fanboy]

@Apple fanboy

Let me come at it from a different angle
What context would you say it's appropriate and within forum rules to use the term "woke"?

(just randomly picking "woke" -- could use a lot of terms here as an example -- moderators, please bare with me as I explore this briefly to understand the other users perspective)

This one.

A fair concern re: coding, I hear you on that

What I wonder is if an AI could be tuned better though -- if it just has exact rules to work off, it'd be better to not entertain context... or at least really dial down how much that is considered

Bias of humans is even worse because it the human understands context just fine, but then purposely, even if just subconsciously, refrains or alters their behavior from simply applying rules of "post is off topic" (as an example) as the human in question is "ok with what was said, even if it technically should be moderated", etc.

Bias is really pernicious that way

Know what I mean?

An example that appears somewhat frequently due to the current climate of things are political terms. There really is no context where things like "woke" or "snowflake" or "maga" etc should be allowed on the forum as they are inflammatory and derogatory ways of talking about things and all their usage does is inflame discussions and solidify polarization....

In short -- those type of terms, from all political angles, just pisses people off and people using those terms know exactly what they are doing.

I think AI would not spot you were using an example rather than calling anyone anything.

 

[turbineseaplane]

This one.

I think AI would not spot you were using an example rather than calling anyone anything.

Hmm. Interesting.
I don't know what my opinion is

I think LLM based AI would spot differences -- that's where the whole "large language model" part comes in.

Absorb a TON of examples and it should be really good at picking out exactly the differences in something like our example situation.

Creation of context from scratch is where hiccups usually occur -- but analyzing text against it's huge dataset of examples is precisely what AI LLMs are really good at.

I could see AI moderation being at least very useful for broadly deciding what should be in a "penalty box" that could get human review. As we know, right now, a huge issue is just sheer volume. Moderators only see what gets reported ... whereas AI could scan it all and find things currently slipping through completely. This of course is a different issue than moderator bias -- which is a different problem than sheer volume.

 

[bousozoku]

ha!
Well done -- excellent example ... I should have thought of that one straight away

I do think, though, AI could be very easily trained to identify that usage vs "calling behavior woke", which of course is the negative connotation with political undertones.

I mean .. I think AI could? I would hope?
lol

AI could be trained, but by the time it got it right, there would be a whole new generation of moderators.

The thing is, every country has its own quirks, and every language with it. e.g. in Spanish, there are two words for hanger. The one, "gancho", also means a female body part, so "percha" is also available.

"Git" means idiot in English, but is used by friends, and that goes for "baka" in Japanese". They're not always used in serious insults.

I suspect that an AI moderator for Japanese would be easier than one for western languages. I can't imagine trying to train one for Serbian. Those people are the best at cursing in the world.

The use of "woke" in the iPad advert takedown thread was along political lines, I'm sure. I replied to someone "So, doing the right thing and being considerate is "woke"?

 

[turbineseaplane]

The use of "woke" in the iPad advert takedown thread was along political lines, I'm sure. I replied to someone "So, doing the right thing and being considerate is "woke"?

Yeah -- the terms being thrown around in that thread were basically all political oriented and used to take shots.

I'm not sure why any of them were allowed as I'm sure many were reported

It's kind of what led me to wonder about moderator bias ... as perhaps some folks looking at those reported posts were "ok with it"? I guess?

 

[bousozoku]

Yeah -- the terms being thrown around in that thread were basically all political oriented and used to take shots.

I'm not sure why any of them were allowed as I'm sure many were reported

It's kind of what led me to wonder about moderator bias ... as perhaps some folks looking at those reported posts were "ok with it"? I guess?

I guess. As the younger crowd takes over, they don't find offense offensive in the same way, which makes me wonder where the world will be in 25 years. It makes me think that people my age should have been better at parenting.

I didn't report any of the posts, even those where I was being insulted, just because I was laughing too hard to report anything.

In real life, I have people complaining about my Spanish, even though there English isn't apparently as good as my Spanish. I've just made it into a big joke.

I'd love to see AI handle something substantial. It will probably turn out like the Japanese-to-English translators.

 

[icanhazmac]

Along the lines of moderation and AI

Do we think it might be better at weeding out potential bias from moderators?

Absolutely not. Did you ever consider that you believe there is "bias" in moderation because you are yourself overly sensitive towards certain words or topics and view them as inherently bad? As an example of this, there are over 1 million members on MR yet you are the only one here claiming that there is bias in moderation because "woke" is being tossed around in what you feel is a negative or derogatory way.

What happens when the folks teaching the AI are themselves "biased" in one way or another? * Cough * Google Gemini * Cough *.

It has been covered many times, moderation is a team effort here at MR, if you are not happy with a team's decision could it not be because you are on the extreme side of the bell curve for what you feel is appropriate discourse? If 0 = Wild wild west, zero moderation and 100 = a thousand pages of strict rules that will not be broken, you might be in the 90s where the vast majority might live in the 30-70 range.

"She woke up, got dressed, and we took her to lunch on Mother's Day."

+1, there are plenty of plain language examples that I do not feel AI is capable of dealing with.

I woke up on the wrong side of the bed.
I woke to find it raining.
Your text message woke me up.
I woke the computer by moving the mouse.

AI could be very easily trained to identify that usage vs "calling behavior woke", which of course is the negative connotation with political undertones.

I disagree.

First off, while I get what you are referring to with this example I don't believe adding to the list of words we cannot say is constructive. Can you not imagine a conversation where "woke", in the context you are citing, can be discussed without "negative connotations or political undertones"?

My .02 is, if I am posting on a forum it is to engage with other human beings (hopefully, but I fear more and more are bots etc.) and I don't want to speak to other humans through a digital filter/censor. In that endeavor we will, see/read/hear things we might not feel are appropriate, that is the cost of "unfiltered" communication. I will gladly accept the occasional horrible post versus what you propose.

I don't understand why anyone would value conversation that has been filtered to the extent that some here want an AI to do. Yes, as a civilization we set some ground rules for civil behavior but like everything else this is a sliding scale. On a public street corner all bets are off, you can and will see and hear just about everything. At a place like MR there are a set of rules that are agreed upon, by joining the community, for civil discourse. The only challenges we have here are when individuals are not happy with the rules and want either more restrictive or less restrictive moderation of discussions.

Personally I err on the side of less restricted conversation. I'm not afraid of "4 letter words" or even bad or horrible statements. I don't know who to attribute this to but I have heard it many times "the best way to counter bad ideas is with better ideas". In other words, allow bad or even horrible ideas but counter them.

That being said, if terms are clearly being used to derail a thread, I didn't read your example as I am not sure which thread it is, moderation could be warranted. Ex: I didn't go to an iPad thread to witness a discussion on the proper use of "woke". I agree that if we are talking iPads I don't need 1 page of that and 10 pages of members bickering about "woke" whatever.

 

[Apple fanboy]

Yeah -- the terms being thrown around in that thread were basically all political oriented and used to take shots.

I'm not sure why any of them were allowed as I'm sure many were reported

It's kind of what led me to wonder about moderator bias ... as perhaps some folks looking at those reported posts were "ok with it"? I guess?

I’ve not read the thread, but when you say you are sure they were reported, did you report them? That’s the best way to help the mods. Report posts that break the rules.

 

[turbineseaplane]

As the younger crowd takes over, they don't find offense offensive in the same way, which makes me wonder where the world will be in 25 years. It makes me think that people my age should have been better at parenting.

I feel the same way

I’m not sure where we went wrong, but I expected better results than have come about (generally)

 

[arefbe]

A perfect role for artificial intelligence; discussion forums moderation. That would be much better than the way it works now, with fallible humans making decisions based on their fallible judgements. I'm sure it would be much more capable of weeding out chatbot posts aswell. It would be both more fair and more predictible, I think.

In time, we might consider it for our judicial systems too.

Your sarcasm is noted…

 

[Nermal]

The thing is, every country has its own quirks, and every language with it.

Indeed. In New Zealand you might have a "wetback", which is a type of fireplace with plumbing in the back of it: it heats your water in winter. I know of someone who went on Facebook asking for assistance with one and got an instant ban for using "offensive" language. He had no idea (and neither did I, until hearing this story) that the word has a completely different meaning in the US.

 

[turbineseaplane]

Indeed. In New Zealand you might have a "wetback", which is a type of fireplace with plumbing in the back of it: it heats your water in winter. I know of someone who went on Facebook asking for assistance with one and got an instant ban for using "offensive" language. He had no idea (and neither did I, until hearing this story) that the word has a completely different meaning in the US.

Again though, AI gets trained on huge data sets of language.
That very much can be localized

And on an international site like this, I'm not sure regional terms that could be interpreted to be offensive elsewhere would (or should) be allowed anyhow.

I feel like folks are trying to find reasons why AI couldn't work at all, rather than looking at all the ways it very likely could.

This is just forum moderation -- not mission critical stuff.
I think there is real potential in these use cases for at least some AI usage

 

[steve123]

Well well well. OpenAI announced GPT-4o today at their Spring Update Event. The OP's suggestion might be a big step closer to reality. If you have not seen it yet, watch the recording of the livestream on the OpenAI website, it is well worth it. GPT-4o is very impressive technology and from what they demonstrated could very likely be used for moderation.

https://openai.com/index/spring-update/