Twitch introducing AutoMod, a self learning moderation tool

[NovusBal]

Source: http://www.kitguru.net/channel/generaltech/matthew-wilson/twitchs-automod-uses-machine-learning-to-help-police-chat/

 

Quote

 

The announcement goes into detail on how streamers can customise AutoMod to suit their needs: “Streamers can configure AutoMod by selecting one of four levels. These levels affect how aggressively AutoMod holds back messages under four categories: identity, sexual language, aggressive speech, and profanity.”

“When AutoMod flags a message, it lets the original sender know that moderators will review the message’s content before it’s sent to chat. Meanwhile the message is shown to moderators, giving them a chance to accept or reject the message.”

 

It seems that Twitch plans to introduce AutoMod, a tool to help streamers on the Twitch platform to not remove just simple insults and swearing, but specifically removing any insults or abusive chat aimed towards people. How this is going to manage that we simply don't know yet, however the most concerning aspect is how this may do this.

 

It does allow for users to have their moderators approve messages being held back by the AutoMod, this at least allows for some moderation. I do think this is at least something, however, for large streamers there won't be enough moderators to hold back the proverbial zerg of profanity that tends to be in twitch chat, especially if each flagged comments needs approved. 

 

My own opinions is that, although this could be developed to be useful, I do not enjoy chat filters or tools of this area, twitch chat is not a pleasant place, but the chat itself is most of the reason a lot of users may stay around to certain channels, to have their input on the situation, even if it is washed over by the sea of comments.

 

What is your own opinion? Would you use this? Let me know in the comments!

[MValuableP]

9 minutes ago, NovusBal said:

Source: http://www.kitguru.net/channel/generaltech/matthew-wilson/twitchs-automod-uses-machine-learning-to-help-police-chat/

 

It seems that Twitch plans to introduce AutoMod, a tool to help streamers on the Twitch platform to not remove just simple insults and swearing, but specifically removing any insults or abusive chat aimed towards people. How this is going to manage that we simply don't know yet, however the most concerning aspect is how this may do this.

 

It does allow for users to have their moderators approve messages being held back by the AutoMod, this at least allows for some moderation. I do think this is at least something, however, for large streamers there won't be enough moderators to hold back the proverbial zerg of profanity that tends to be in twitch chat, especially if each flagged comments needs approved. 

 

My own opinions is that, although this could be developed to be useful, I do not enjoy chat filters or tools of this area, twitch chat is not a pleasant place, but the chat itself is most of the reason a lot of users may stay around to certain channels, to have their input on the situation, even if it is washed over by the sea of comments.

 

What is your own opinion? Would you use this? Let me know in the comments!

There are already loads of possibilities to do this now without this "AutoMod". Nightbot can do these things too, but I only see this particular thing working for bigger streamers who will have a packed chat, and then you'd need some auto verification of which messages (that are flagged) should be deleted. But good thing that Twitch are acknowledging that their that is... aids.

[zMeul]

it will end up banning everyone before they could finish the word they were typing 

 

[sof006]

Because this could end well....

[tlink]

not a bad idea really, if Channel owners don't like it they can disable it so nothing wrong with it.

[FPSwithaWacomTablet]

For this to be implemented properly, one needs to have solved the entirety of the Internet... 

[AshleyAshes]

10 hours ago, sof006 said:

Because this could end well....

Twitch chat is hell, it can't possibly end any worse than it currently already is.

[sof006]

33 minutes ago, AshleyAshes said:

Twitch chat is hell, it can't possibly end any worse than it currently already is.

It depends which twitch streamers you watch lol. I am watching a stream right now and the chat is fine.

[Fetzie]

3 hours ago, sof006 said:

It depends which twitch streamers you watch lol. I am watching a stream right now and the chat is fine.

It's no problem on streams where you can actually read the chat. But human moderators can't cope with twitch chat when you've got 2000+ people using it (even if you turn on slow mode) because by the time you've clicked on the "ban" symbol you're banning somebody 30 posts further down the chat.

[Thony]

RIP twitch chat freedom. 

[Misanthrope]

23 hours ago, tlink said:

not a bad idea really, if Channel owners don't like it they can disable it so nothing wrong with it.

Actually yes: a really bad idea cause of this:

On 12/13/2016 at 4:15 AM, zMeul said:

it will end up banning everyone before they could finish the word they were typing 

 

 

[tlink]

1 minute ago, Misanthrope said:

Actually yes: a really bad idea cause of this:

 

it can't talk, it just flags stuff. thats all it does, flag things it thinks the mods will not like, if mods do OK it it still gets trough. automod is powerless without mods.

[Misanthrope]

2 minutes ago, tlink said:

it can't talk, it just flags stuff. thats all it does, flag things it thinks the mods will not like, if mods do OK it it still gets trough. automod is powerless without mods.

That's not the fucking point: It can be easily taught to ban silly things or have a specific political inclination or bias, etc.

[tlink]

1 minute ago, Misanthrope said:

That's not the fucking point: It can be easily taught to ban silly things or have a specific political inclination or bias, etc.

if the mods themselves are biassed, automod will be biassed. but thats not automod's designflaw. and no it can't ban, so it can't be thought to ban anything.

[Misanthrope]

2 minutes ago, tlink said:

if the mods themselves are biassed, automod will be biassed. but thats not automod's designflaw. and no it can't ban, so it can't be thought to ban anything.

On a per channel level there's not as much potential for abuse but I think this will have some overall AI otherwise it's fairly fucking useless: easily defeated by raids and such.

[tlink]

1 minute ago, Misanthrope said:

On a per channel level there's not as much potential for abuse but I think this will have some overall AI otherwise it's fairly fucking useless.

even then automod's biass still goes trough the real moderaters so any biass gets filtered out by them.

[Misanthrope]

Just now, tlink said:

even then automod's biass still goes trough the real moderaters so any biass gets filtered out by them.

So without real moderators, which is the point since they can't handle big chats, it's useless? Like what the fuck does this thing does then? Learn when certain things are not cool and then tell mods that might miss them, can't keep up or are simply not fucking there cause the streamer isn't big enough or doesn't has enough moderators anyway?

 

Again I haven't seen a single fucking use for this shit.

[tlink]

2 minutes ago, Misanthrope said:

So without real moderators, which is the point since they can't handle big chats, it's useless? Like what the fuck does this thing does then? Learn when certain things are not cool and then tell mods that might miss them, can't keep up or are simply not fucking there cause the streamer isn't big enough or doesn't has enough moderators anyway?

 

Again I haven't seen a single fucking use for this shit.

its meant to help the real moderators filter more accuratly, not replace them. that why its only capability is to flag messages and not actually do anything about them. i can also search the web myself by typing in random urls vagely representing what i want but googles AI can do that way quicker for me and organize it in such a way that i can easily deal with it. AI's are good at handling bulk info, twitch chat often is exactly that. its up to the mods to decide what to do with its end result.

[Misanthrope]

Just now, tlink said:

its meant to help the real moderators filter more accuratly, not replace them. that why its only capability is to flag messages and not actually do anything about them.

Right so it's either gonna be taught to flaw silly shit or biased shit or it's gonna be useless since it can't learn enough by being on just one channel so any raid can defeat it.

 

So it's either abused or it doesn't works: it doesn't fucking works. AI never works it's just not sophisticated enough yet to ignore trolls.

[tlink]

Just now, Misanthrope said:

Right so it's either gonna be taught to flaw silly shit or biased shit or it's gonna be useless since it can't learn enough by being on just one channel so any raid can defeat it.

 

So it's either abused or it doesn't works: it doesn't fucking works. AI never works it's just not sophisticated enough yet to ignore trolls.

you can say that all you want but it has no basis, we both don't know how it learns yet. all i know is that its an AI that learns on moderator's behaviour. if thats on a per channel basis we don't know. it could very well be that it access the database of per channel moderator actions and tests its own prediction algorithm against that until it gets to a certain accuracy level.

[Misanthrope]

Just now, tlink said:

you can say that all you want but it has no basis

This is the basis:

 

 

Nobody has managed and Twitch isn't bigger than some of the other companies that tried and failed.

[Space Reptile]

my prediction of what the bot will say after half year of exposure

Quote

kappa pogchamp babyrage kappa kappa

 

[tlink]

9 minutes ago, Misanthrope said:

This is the basis:

 

 

Nobody has managed and Twitch isn't bigger than some of the other companies that tried and failed.

you're cherry picking data to suit your narrative. just because one situation fails doesn´t mean that every situation is going to fail. automod isn´t some ai thats going to communicate with us, just like google search prediction isn´t communicating with us. im not even sure if calling automod an AI first is accurate as its more machine learning than AI. tay was an AI first and machine learner second. automod is a completely different ballpark.

also this:

Quote

“AI has become so pervasive in our lives we don’t come to recognise that it’s powering a lot of things,” she added. “You probably use it dozens of times a day without knowing it.”

 

http://www.wired.co.uk/article/machine-learning-ai-explained

[Misanthrope]

2 minutes ago, tlink said:

you're cherry picking data to suit your narrative.

No we pointed at a popular example on a much wider scale with people with much more resources failing at filtering out trolls. You're dismissing a relevant case (without even explaining why it isn't relevant or wouldn't apply) for the sake of your narrative. There's nothing further to discuss.

[tlink]

Just now, Misanthrope said:

No we pointed at a popular example on a much wider scale with people with much more resources failing at filtering out trolls. You're dismissing a relevant case for the sake of your narrative. There's nothing further to discuss.

well then prove that its the case on a much wider scale, because until you do that its just cherry picking. i can point you to googles search prediction as a comeback but again thats cherrypicking.