A simple experiment using sentiment-analysis to show how algorithms such as word2vec can be negatively influenced by hidden biases of human language which make up the training corpus.
Original blog: http://blog.conceptnet.io/posts/2017/how-to-make-a-racist-ai-without-really-trying/