“The researchers argued that, whichever way you use it, their algorithm vastly outperforms the human judge.”