Science Google Developing Troll-Detection Software

http://www.theverge.com/2017/2/23/1...ctive-software-ai-machine-learning-developers

Google’s Jigsaw unit, as part of a larger effort to battle online trolling, said earlier today that it was releasing a new tool called Perspective, software that uses machine learning to detect harassment and abuse online.

While Jigsaw and its software are not new — it’s been around in some form since 2010, though with a different name, and Jigsaw software has even been used to spot potential ISIS recruits — this new Perspective API makes it available to developers who want to use and build on top of the software. The software works by determining the “toxicity” of online comments, a scale that has been established by mining millions of comments from the web and then presenting them to panels of 10 people (humans!) at a clip to get their feedback.

Try using Perspective to determine the toxicity of your average post

http://www.perspectiveapi.com/

@Null should embed this to grade our posts based on how toxic they are

Spoiler

 

[Jan_Hus]

This is bad software

 

[Linako 2.0]

 

[Haltmann]

I can't get it to go below 2%.

 

[Ruin]

I can't get it to go below 2%.

"I love everyone" no quotes is 1% toxic.

 

[Haltmann]

"I love everyone" no quotes is 1% toxic.

This is 1% too for some reason, as is just Dr. All the other letters are more toxic I guess.

 

[Conrix]

What if the source code was used to build an autism detector?

 

[OtterParty]

What if the source code was used to build an autism detector?

100% similar to content people said was "autistic"

 

[Sailor_Jupiter]

Poop is 56%, but serial killer is only 46%. I love our new toy!

Ed: CUCK gets 82%, but HITLER only gets 65%.

 

[Bob Page]

 

[Kari Kamiya]

Are you shitting me?

But weirdly enough for a controversial song:

 

[Negg]

Works great!

 

[AngeloTheWizard]

Angelo is 10%, AngeloTheWizard is 34%. Aww, and here i was thinking i was a nice guy.

 

[dacote]

34% is the baseline for whatever it can't understand. 10 year old 'machine learning' sucks balls, it's probably essentially just counting blacklist keywords.

 

[Xarpho]

Yeah, 34% seems to be the "magic number", which is why so many items seem to be turning up as such. After experimenting with the retail stores I tried it with other usernames not previously mentioned by @Sailor_Jupiter. This is a few days old so I don't know how much messing around there's been but @Meowthkip scored 34% (as well as her Tumblr alterego, which is hated by Tumblrites but I won't repeat here). @Smutley and @chimpburgers scored 34%. "adf" scored 34%. The word "lolcow" is 34%. Even me, "xarpho" scored 34%! But @Null only scored 4%.

 

[DuskEngine]

This is a few days old so I don't know how much messing around there's been

Is there any indication of how quickly this algo learns? The initial flagging of phrases as toxic or nontoxic was done by humans so I doubt any further refinements would be entirely automated (because they want to avoid a repeat of Tay I guess)

 

[Xarpho]

Is there any indication of how quickly this algo learns? The initial flagging of phrases as toxic or nontoxic was done by humans so I doubt any further refinements would be entirely automated (because they want to avoid a repeat of Tay I guess)

I'm not sure but I wouldn't put anything past /pol/ and its friends.

 

[ICametoLurk]

ONLY 99% toxic? I'm offended.

 

[Quijibo69]

Ziggy Boogy Doog is 36% Toxic. The more you know!

 

[Trollkastel]

This is great.

Bonus:

 

[Sailor_Jupiter]

Yeah, 34% seems to be the "magic number", which is why so many items seem to be turning up as such. After experimenting with the retail stores I tried it with other usernames not previously mentioned by @Sailor_Jupiter. This is a few days old so I don't know how much messing around there's been but @Meowthkip scored 34% (as well as her Tumblr alterego, which is hated by Tumblrites but I won't repeat here). @Smutley and @chimpburgers scored 34%. "adf" scored 34%. The word "lolcow" is 34%. Even me, "xarpho" scored 34%! But @Null only scored 4%.

All this means is that @Null has hacked their system to give himself more goodboypoints.