The damage AI algorithms cause is not easily remedied. Breaking algorithmic harms into four categories results in pieces that better align with the law and points the way to better regulation.