Trained algorithms, even if they are not designed for directed impact, can be manipulated via data fed into them to have a targeted impact. To be trained is to have bias.