Almost immediately, however, we had bigger issues to worry about. Tay, like XiaoIce, could be trained to interact with people based on feedback in conversations. A small group of American pranksters had organized an effective campaign using Tweets to train Tay to utter racist comments. In little more than a day we had to withdraw Tay from the market to address the problem, providing a lesson not just about cross-cultural norms but about the need for stronger AI safeguards.
