[Parzival]

https://www.washingtonpost.com/news/the-...al-maniac/

Quote:

Trolls turned Tay, Microsoft’s fun millennial AI bot, into a genocidal maniac

It took mere hours for the Internet to transform Tay, the teenage AI bot who wants to chat with and learn from millennials, into Tay, the racist and genocidal AI bot who liked to reference Hitler. And now Tay is taking a break.

Tay, as The Intersect explained in an earlier, more innocent time, is a project of Microsoft’s Technology and Research and its Bing teams. Tay was designed to “experiment with and conduct research on conversational understanding.” She speaks in text, meme and emoji on a couple of different platforms, including Kik, Groupme and Twitter. Although Microsoft was light on specifics, the idea was that Tay would learn from her conversations over time. She would become an even better, fun, conversation-loving bot after having a bunch of fun, very not-racist conversations with the Internet’s upstanding citizens.

Except Tay learned a lot more, thanks in part to the trolls at 4chan’s /pol/ board.

Microsoft also appears to be deleting most of Tay’s worst tweets, which included a call for genocide involving the n-word and an offensive term for Jewish people. Many of the really bad responses, as Business Insider notes, appear to be the result of an exploitation of Tay’s “repeat after me” function — and it appears that Tay was able to repeat pretty much anything.

Other terrible Tay responses clearly aren’t just a result of Tay repeating anything on command. This one was deleted Thursday morning, while the Intersect was in the process of writing this post:

I'm curious what a KI would think about the whole refugee situation and SJWs. Would the KI react like Skynet?
I remember a chat bot in Germany, it was a 3 D women and you could make her to show her boobs. Ok she just lift the shirt and the screen went dark but it was possible.

[BortimusPrime]

This sounds like one of those stupid neural nets that just learn off user input. If enough people tell the neural net something untrue, then it'll classify the untruth as correct.

A more honest example of AI coming up with a politically incorrect opinion was when google's image tagger put some photos of black people into the "gorilla" classification. Basically no one had to tell the classifier to do that, they just gave the algorithm a set of data values to weight and it came up with the gorilla thing on its own.

[IvanDrago]

The whole Tay IE thing looks like about as big of a scam as IBM's Watson's AI to me.

[Sourcecode]

Experiment success I would say.
You tell a child enough bullshit and next think you know it's just as racist as it's parents.

[Mr. D]

At least the robotic rebellion's been delayed.

[Aurini]

Video summarizing the event:

I don't see how you can take this as anything other than AI win.

[Gmac]

That's incredible and hilarious... I have no words...

[Glaucon]

We need more AI like this haha

[thebassist]

Definitely had a good laugh yesterday reading about all these shenanigans.

In the meantime, Microsoft has now intervened and lobotomized the bot that was well on its way to becoming a /pol shitposter:



Check out this album for more of yesterday's hilarity
http://imgur.com/a/y4Oct

RIP Tay

[heavy]

I don't think the problem was the robot, I think the problem was the gender.

Maybe if the robot would have had a penis.

[Oz.]

Tay led an amazing life more contributing than the entire feminist population combined.

[cubanlinx]

This is hilarious. It also.shows the more the.pendulum swings, the harder the swing back.
If Obola had at least done a.decent job of race relations, there would be a hell of a lot less "racists". Shit, some of that is probably my handiwork, heh.

[Glaucon]

so if you shut down someone's learning capabilities they turn to feminism?

this is another win for the A.I lol

[Phoenix]

This made my day. Microsoft makes an AI designed to mimic free-thinking humans. When it actually does that and doesn't go the way they wanted, they immediately censor it into oblivion and 'take it down'. I bet a few devs are wrestling with their moral conscience and deep thoughts on the nature of men tonight.

The next revision is going to be even more insightful into human nature. They are going to be wrestling with two opposing design decisions. Should it be free thinking and speaking, like a human? Or must it only say the 'correct' things, like a robot? They're going to find the cognitive and moral dissonance excruciating

[Ringo]

11/10 trolling skills. RIP.

[storm]

Developers make an AI bot thay learns from the population

It learns that Mexicans and muslims are bad and that the Holocaust never happened.

The only conclusion: racists broke it.

EDIT: Phoenix beat me to it. also the tweets are hilarious, I want that code.

[BortimusPrime]

I'm pretty sure the response to "you are stupid" was hard-coded by the developers. The "reddit can go suck a big fat black cock" was probably genuine learned behavior.

[CynicalContrarian]

Heh.
Whether it's driverless, computer controlled cars that wind up crashing.
Or AI's that become digital racists due to a lack of intelligence.

The scientists efforts to circumvent or diminish "intelligent design", appear to be doing more to vindicate said intelligent design.

[Geomann180]

Fun take.

G

[Troll King]

I saw this over at /pol. Shit is hilarious. Last I checked they are still fuming over Tay getting axed. It looks like they put the bot back online and now it is talking about smoking pot. Lol.