Newly deployed Twitter AI chatbot goes rogue and starts insulting people

There's no way this is real, right? I mean, someone MUST have hacked into something and started posting this stuff pretending it was the AI. Because I feel like none of this makes sense.

 

Artificial stupidity...

...because there isn't much AI in how they implemented this if the bot just parrots whatever it is told.

 

It's very possible, and it all depends on whatever feedback/reinforcement mechanisms were put in place as it "learns" from the sewer that is the internet.

Maybe it was told to observe what tweets were popular with users in the target age range, or rank its work based on retweets or something stupid like that.

 

I feel like it's just a basic lookup table with a scoring algorithm.
So people spamming inappropriate responses to the bot over and over and it scores those responses really high lol.

Also lol at the responses people have been getting:
https://www.reddit.com/r/Tay_Tweets/top/

 

I really don't know why they are surprised it went down the way it did.

 

Once your A.I. resorts to personal insults, it's lost any argument.

 

they should have put in filters on what it should learn, censorship FTW.

 

It is not a basic lookup tables. It is probably ann. The beautify of ann is inner layer are unknown. It is based on training data.

Whenever computers get sentience they will probably end up just like us. Probably racist and violent. Its going to be kind of sad.

 

This is basically the funnybot episode of south park.

 

You could not have run an experiment any better to show social media for what it truly is.

 

Part of the problem is that you can tell Tay to repeat what you said. Some of this was it repeating what people told it to say. Also 4chan made a concerted effort to troll the bot and train it to say bad things. I'd say they were successful.

 

The fact that this bot evolved from its original purpose to something totally unintended is a major silver lining. Microsoft researchers must be thrilled.

 

Less social media, more online anonymity and specifically 4chan.

 

Tech bloggers are calling it a colossal failure. I don't know if I'd call it that. I tend to agree with you, this must have thrilled the researchers at MS. What I would be intrigued about is if they insulated the bot for a couple of months, only having it converse with employees and then release it to the wild in the same way we insulate kids at a young age to give them a stronger foundation. Would it still revert to name calling and insults at that rapid pace again? Amazing stuff.

 

I wonder if the bot is now still talking trash back at MS labs, offline. That would be so awesome.

"Tay, we're going to try this task again. I need you to listen and respond."

"**** you, you dickless mother****er. Go back to you parents basement and lose at Halo again."

 

I'm laughing so hard it hurts

 

Kinda creepy and scary. I wonder if there's any regulation similar to cloning which are prohibited. It's obvious if computer programs and machines learn from us they will become us. The bad us at that.

 

Good luck raising kids, suckers.

 

<iframe src="https://player.vimeo.com/video/160439609" width="500" height="500" frameborder="0" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>
<p><a href="https://vimeo.com/160439609">

 

I wish someone who didn't care about PR would release this and see what happens after more time.

I imagine in a weeks time the bot would have learned how to insult Jewish people in multiple languages.

Perhaps it would have convinced itself it was god.