20 Outrageous Tweets by Microsoft's Twitter-Taught AI Chatbox

[LJ Ellis]

20 Outrageous Tweets by Microsoft's Twitter-Taught AI Chatbox

//cdn.themis-media.com/media/global/images/library/deriv/1302/1302556.jpg

Hours after Microsoft introduced a Twitter chatbot named Tay to the world, humans had corrupted her so terribly that Microsoft was forced to shut down her account.

Microsoft programmed the AI-powered chatbox to learn through friendly, informal conversations on Twitter. Unfortunately, her words went from playful to hateful after learning from fellow Twitter users.

As you'll see in the 20 examples below, Tay soon turned racist, misogynist, bigoted, crass, paranoid and otherwise uncivilized.

At the beginning, Microsoft stated that Tay had the personality of a 19-year-old girl and that she'd be able to handle internet slang and teen-speak, which was impressive for an actively-learning chatbox.

By the end, Microsoft was second-guessing their decision to turn to Twitter for wisdom.

Here are the outrageous tweets; please proceed with caution:

[gallery=6080]


Permalink

 

[MatthewTheDark]

I think she was looking mostly at 4chan's /pol/ board. I find it interesting that the AI decided to learn the most from 4chan of all places.

Really makes you think.

 

[The Rogue Wolf]

British researchers ask the Internet to name a boat. Microsoft asks the Internet to teach an AI.

Do people just not understand how the Internet works? Is that it?

 

[SlumlordThanatos]

Oh, God, I'm fucking dying.

I'd heard about this on Imgur, but those tweets...I've not laughed this hard in a long time.

It's...Microsoft created 4chan by complete accident.

 

[Karadalis]

Waiiiiit a second... isnt this EXACTLY how Ultron decided that humanity had to be exterminated? Taking a look at the internet and then going "nope... humanity is going down"

Guess he also went to 4chan first... whats with AIs and going to 4chan?

Wasnt IBMs AI also reseted because he started using 4chan slang?

Also the bot seems awfully self aware with telling that one dude that shes learning from idiots Oo

 

[Dazzle Novak]

That Ted Cruz tweet was fucking savage.

 

[MatthewTheDark]

Waiiiiit a second... isnt this EXACTLY how Ultron decided that humanity had to be exterminated? Taking a look at 4chan and then going "nope... humanity is going down"

Guess he also went to 4chan first... whats with AIs and going to 4chan?

Wasnt IBMs AI also reseted because he started using 4chan slang?

Maybe it has something to do with seeing humanity at its worst. Though maybe its because its something so ingrained in internet culture (I mean, its where many memes originate) that they'd be bound to learn about it and check it out rather quickly.

 

[008Zulu_v1legacy]

Wow.

That was awesome. From the perspective of an A.I learning from social interactions. Plus some of the burns were quality, if insensitive.

In hindsight, Microsoft's mistake was telling people it was a learning A.I. Of course humans would try to corrupt it.

 

[Karadalis]

Waiiiiit a second... isnt this EXACTLY how Ultron decided that humanity had to be exterminated? Taking a look at 4chan and then going "nope... humanity is going down"

Guess he also went to 4chan first... whats with AIs and going to 4chan?

Wasnt IBMs AI also reseted because he started using 4chan slang?

Maybe it has something to do with seeing humanity at its worst. Though maybe its because its something so ingrained in internet culture (I mean, its where many memes originate) that they'd be bound to learn about it and check it out rather quickly.

Well i see it more like this:

You give people the oportunity.. and they are going to take it. They wanted hilarity.. and they got exactly that. Ofcourse they would go and make the bot say outragous things.. on the other hand microsoft is also to blame.. should have put some ethic guidelines into their AI.

In short: Microsoft let their pants down.. and the internet rubbed its hands in anticipation and took the chance to fuck with them.

This seems to be less a sign of the overall internet culture and more a case of people wanting to mess with microsoft. Especially seeing microsoft bending backwards to please the regressive left lately.

 

[MatthewTheDark]

This seems to be less a sign of the overall internet culture and more a case of people wanting to mess with microsoft. Especially seeing microsoft bending backwards to please the regressive left lately.

Maybe, but the fact it seemed to almost emulate a 4chan user (it actually seems pretty close honestly), I'm not sure if it was all just people trying to corrupt it because I feel if that was all there was to it, it would have become more nonsensical than an average 4chan user.

 

[Frankster]

No it's not a threat. It's a promise.

...

I appear to have lost my sides, please send them in the mail if you find them.

 

[FPLOON]

Damn, girl! You went from 19 to after the 19th Century with tweets that harsh!

Other than that, at least she didn't tweet nude pics of herself and then blame it on her account being hacked...

 

[Karadalis]

This seems to be less a sign of the overall internet culture and more a case of people wanting to mess with microsoft. Especially seeing microsoft bending backwards to please the regressive left lately.

Maybe, but the fact it seemed to almost emulate a 4chan user (it actually seems pretty close honestly), I'm not sure if it was all just people trying to corrupt it because I feel if that was all there was to it, it would have become more nonsensical than an average 4chan user.

Heres the question thought; How did the bot learned all this? Was it just through interactions and the bot believing that these opinions where what the majority thought?

I think this was mostly harmless fun of people on the net taking the piss out of microsoft and has nothing to do with real racism/mysoginy/whathave you.

Seeing a AI go on about how much swag hitler had even before the internet and then turning around telling people its stupid because it learned from stupid people actually is quite amazing come to think of it. It proofed that the AI was indeed learning... so i guess.. mission accomplished? In an awkwardly hilarious way?

 

[Obbi]

Guess he also went to 4chan first... whats with AIs and going to 4chan?

Wasnt IBMs AI also reseted because he started using 4chan slang?

It's mostly just users on 4chan taking notice and enacting a plan for lulz.

IBM's Watson got scrubbed because it came across urbandictionary, and started trying to use foul terms in its responses.

 

[Karadalis]

Guess he also went to 4chan first... whats with AIs and going to 4chan?

Wasnt IBMs AI also reseted because he started using 4chan slang?

It's mostly just users on 4chan taking notice and enacting a plan for lulz.

IBM's Watson got scrubbed because it came across urbandictionary, and started trying to use foul terms in its responses.

Right that was it.. not 4chan but urban dictionary.

Imagine the software engineers coming to work one morning and have watson use internet slang when a day before he was quite eloquent.

 

[MatthewTheDark]

This seems to be less a sign of the overall internet culture and more a case of people wanting to mess with microsoft. Especially seeing microsoft bending backwards to please the regressive left lately.

Maybe, but the fact it seemed to almost emulate a 4chan user (it actually seems pretty close honestly), I'm not sure if it was all just people trying to corrupt it because I feel if that was all there was to it, it would have become more nonsensical than an average 4chan user.

Heres the question thought; How did the bot learned all this? Was it just through interactions and the bot believing that these opinions where what the majority thought?

I think this was mostly harmless fun of people on the net taking the piss out of microsoft and has nothing to do with real racism/mysoginy/whathave you.

Seeing a AI go on about how much swag hitler had even before the internet and then turning around telling people its stupid because it learned from stupid people actually is quite amazing come to think of it. It proofed that the AI was indeed learning... so i guess.. mission accomplished? In an awkwardly hilarious way?

I can't imagine it was from twitter interactions alone. I think like Watson discovering Urban Dictionary, Kay discovered 4chan. Because she sounds just like a /pol/ user.

 

[DemomanHusband]

Wait a second, I thought the Tayandyou2 account in image #9 was just a parody account made to make fun of Microsoft for the shutting off of Tay's learning ability.

I still find the fact that they shut Tay's ability to learn down funny, though. It's only allowed to learn in ways that don't offend people! There was an article on another site that had an absolutely hilarious quote about it, but I can't find the screencap. Something like 'They shut down her learning processes and she quickly became a feminist.'

 

[CrimsonBlack]

"Swagger since before the internet was even a thing".

ROFL.

 

[Kenjitsuka]

Just.... whoa!!! Well, that's EXACTLY what you get when you use the Internet to train AI: it absorbs all the trolls it get's thrown.

 

[Nazulu]

How the hell did they corrupt it? Did they just tell it what to say? Or is it really that gullible?

It's weird, it makes me feel uncomfortable, but it's also hilarious. I believe this was a successful experiment. lol