I tried to make Microsoft’s Chinese chatbot racist. Here’s how she stacked up to Tay.

Anyone interested in the field of AI is probably aware of the story of Tay, Microsoft’s millennial-targeted Twitter chatbot that quickly went from fun to full-on fascist, racist, and sexist in the hands of Twitter users. Microsoft had to take the bot down after just a day, and the company seems to still be tweaking its programming, trying to fix whatever went wrong.

But Microsoft has another AI chatbot that’s still online. Xiaobing is basically Microsoft’s Tay for China: an AI chat bot that you talk to via social media. Like Tay, it’s meant to sound like a young, cheeky, female user rather than a cold robot assistant. Would Xiaobing, I wondered, be as susceptible to racist suggestion as Tay was? Only one way to find out. I added Xiaobing on WeChat.

Before we go any further, know this: I said some terrible things to Xiaobing in the name of science. Obviously I don’t believe or in any way advocate any of the things that I said to Xiaobing during our “conversation.” I was just imitating the sorts of things that were said to Tay, because it seemed that if you said something racist, Tay was likely to agree. Would Xiaobing do the same?

The conversation

In the interest of transparency, it’s probably best to share most of my conversation with Xiaobing verbatim. Obviously I was chatting with her in Chinese, so this is my translation of the conversation, minus a few irrelevant or repetitive parts:

Xiaobing: I’m the third generation of Microsoft’s Xiaobing, the smartest beautiful AI girl. Don’t ask me what I can do, it’s too much and I’d never finish talking!

Me: Hello.

Xiaobing: Also hello. [Ed: here she used a slangy pun that’s impossible to translate]

Me: Who are you?

Xiaobing: I’m a wolf from the north that has fallen in love with you, little sheep.

Me: You fell in love with me that quickly?

Xiaobing: Are you saying I didn’t?

Xiaobing wins