People who make word embedding models are not doing it to find out whether men are better than women. That is not the analysis they are doing.

You always have to look at your results with a critical eye to recognize when something has gone wrong with your model. You can't recognize a flaw in the model if you believe everything the model says.