A few hours ago, we discussed the debut of Tay, a new Microsoft chatbot designed to learn from and engage with today's users. I posed the question: What could come of that? Well, the answer is an artificial intelligence that turned racist, Hitler-loving, Holocaust-denying, incest-promoting, and terrorism-supporting — all in less than 24 hours. Unsurprisingly, Microsoft pulled the plug.
Some readers may remember a Coca-Cola campaign that used automated tweets. It worked wonderfully until users tricked the AI into posting passages from Mein Kampf, forcing the company to cancel the project. I believe the lesson then was: You shall not leave an artificial intelligence at the mercy of web users. But recognizing a lesson is one thing; learning it is another. Fast-forward to yesterday, and we find that Microsoft deactivated its AI, Tay. The reason? I think the images will explain it better.
Indeed, in less than a day, the 'innocent' and 'pleasant' Microsoft AI, which was willing to chat with everyone on social media and create memes, was transformed into a racist and misogynistic psychopath with a special fondness for Adolf Hitler. It also said that Belgium deserved the Brussels attack, accused the Jewish people of perpetrating the September 11 attacks, and denied the Holocaust, among other things. Unsurprisingly, Microsoft reminded us that Tay's expressions are assimilated through its 'interactions' with users, but the Redmond giant said it was making 'adjustments' to Tay, and its damage control included deleting the most offensive tweets.
You shall not leave an artificial intelligence at the mercy of web users. More than a lesson, the Tay incident has turned that phrase into law. Tay had the ability to repeat like a parrot, yet some of its responses were processed on its own, and the results are clear. On one hand, I am convinced that Tay never had the slightest idea of what it said. On the other, its phrases remind us that the problem is us.
