Wait, so he asked it to generate the opposite of a man who’s been pictured smiling and is considered traditionally handsome (Marlon Brando), and it’s somehow shocking that the opposite of that is a face of a frowning woman with ugly features?

I mean all the training data that gets fed to this thing just results in the replication of pre-existing biases of the society that data was collected from, filtered through the people collecting and classifying the data.