How can you assert that it will evaluate everyone equally when biases are a well documented deficiency of various flavours of AI?