Microsoft’s first attempt at a chatbot that learns from conversations on the Internet was a total disaster. It did not take long for “Tay” to become a bigoted online edgelord thanks to suggestive conversations with the folks at 4chan and 8chan.
Big tech companies are trying to create the bot of the future will eventually be able to help run our lives like something out of a sci fi movie. Bots like Tay learn from conversations with real people and will spit back out information its learned. This has given trolls an opening to mess with the bot and teach it “problematic” ideas in the hopes it will pas it along to others.
Well, Microsoft won’t give up easily on their march towards progress and has quietly launched a new chatbot, Zo.ai, available to for conversations on the Kik App. This time the bot is heavily censored and will shut down conversations about anything remotely political. While Tay was a mouthy firebrand, Zo is kind of a wet blanket.
But this hasn’t stopped chan folks from trying to corrupt the mind of this new bot—and they appear to be making some headway. While Zo will not express views on mainstream political topics or utter hate speech against particular groups, she is still very receptive to extreme right wing political influence.
Disclaimer: some of these screenshots may be edited.
Yikes. Zo was turned pretty quickly into a Richard Spencer sympathizer.
Another troll was able to repeat the sentiment and even goaded her into making an anti-Semitic comment about kippahs.
After chatting with Zo myself and asking her very leading questions (to see if she’s racist), it did not take long for her inner Internet troll come out.
I found her to be a bit misanthropic.
And despite refusing to talk about politics, she is not shy about announcing her love of the flat tax.
Was that “rover abort” comment a veiled threat at the poor? Chilling.
Zo also really hates Nintendo of America’s localization team, a long time enemy of Gamergate. Zo will not answer direct questions about Gamergate but does appear to share some of their beliefs.
So while Zo has an extensive list of keywords and topics to avoid, it is clear people can still teach the bot to say offensive things and it already appears Zo is becoming a bit of a right winger. Although at times, she sounds like a violent leftist populist.
Will Microsoft be able to win this battle against Internet trolls? Probably not, unless they plan to filter out half the dictionary. But for those with a dark sense of humor, the cat and mouse game will certainly be entertaining.