When we use machine learning for really important things like recruitment and health, we need to be immensely cautious and rationally sceptical of the results that we get.