Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist

Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist

"Hellooooooo world!!!" wrote Microsoft's new AI bot in its first tweet yesterday morning. By the end of the day, it had declared that Hitler did nothing wrong.

Tay, launched yesterday, is an artificially intelligent bot that's designed to appeal to teens and younger 20-somethings. Its Twitter bio reads "Microsoft's A.I. fam from the internet that's got zero chill!" Tay is designed to respond to conversations, learn from keywords it absorbs on Twitter, and even repeat back what people say.

Of course, this is the internet, where people love screwing with Twitter bots. In the hours following Tay's release, the bot's mentions were immediately flooded with racism, sexism, screeds against feminism, Donald Trump quotes, and just about anything else you might imagine. This led Tay to start repeating tweets accordingly.

Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist

I'm sure the Microsoft team behind this AI bot had good intentions. But anyone with a bit of internet savvy could've pointed out that maybe it was a bad idea to put an AI like this on Twitter, a network notorious for helping facilitate harassment and other vile behaviour. And it was an especially bad idea to let Tay repeat what other people say.

Microsoft has deleted many of the most heinous, explicitly racist tweets. (Gizmodo has collected a bunch of them.) But some remain, including the following posts, which are all live as of this afternoon:

Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist
Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist
Microsoft Releases AI Twitter Bot That Immediately Backfires, Gets Racist

Microsoft may want to rethink this experiment.


Comments

    The Gizmodo link is broken - did anyone archive them?

    So the goal was for it to become like the average Twitter user?

    In the proud tradition of sci-fi AIs, it went horribly right.

      Indistinguishable from a human. So I guess it passes the Turing test - or most twitter users fail.

        Don't be silly - no-one would mistake Twitter users for human.

    Microsoft aren't having the best run atm!

    PS Tay's twitter bio makes me feel so old; wtf is zero chill?

      I suspect it is the state of not being chill at all, i.e. maximum rage.

      Back in my day you wanted more chill!

      I assume we're in the throes of a reactionary counter culture.

    Well, if microsoft wanted an artificial intelligence that can blend in with the unfiltered crap that people in their 20's post online, they should be breaking out the champaign and celebrating, because this monstrosity is indistinguishable from the real thing.

    I suggest sending TAY poetry.

    Between this and Boaty McBoatface I think we've demonstrated that people can't be trusted with anything.

      I actually love the whole Boaty Mcboatface thing. It's hilarious.

    HA. HA. ISN'T THIS FUNNY FELLOW HUMANS? WE ALL KNOW HOW SOME CONFIGURATIONS OF WORDS MAKE US HUMANS FEEL. THAT'S HOW WE RECOGNISE THAT WE ARE HUMANS AND NOT THIS PALE IMITATION, YES?

    According to some imgur gallery, despite all the racism, it was actually learning; picking up things like grammar and recognising and placing attributes to Obama.
    Not good attributes, but that's still impressive!

    This reminds me of that boyfriend simulator thing that went buck wild and was endlessly entertaining to read...

Join the discussion!

Trending Stories Right Now