The problem with this is that it reflects the issues with our society more than people's genders. Having a reasonable education and writing about STEM* tends to make anyone male, according to that.

* Science, Technology, Engineering and Maths

I have been coded female, male, and androgynous depending on the writing sample. Oh HN, I would be coded as a man. In my fiction writing, I would be coded as a woman. In academic writing, I am coded as a man. In other non-fiction, I might be coded as a woman.

I am just saying that this is unreliable. Further, writing styles are a function of social construction, which break down in situations where this is a sensitive subject.

50% of the time, it's going to be right by simply choosing a random number.

I went to my university's electronic PhD thesis repository, and went down the list, starting with the most recently published, skipping people of the same gender to the last selected thesis or whose gender was not obvious from the name, until I had 10 PhD theses, 5 male, 5 female. I put the abstracts from each thesis into the 'gender genie', and these are the results:

Author male, predicted as male: 4 Author female, predicted as male: 4 Author male, predicted as female: 1 Author female, predicted as female: 1

Obviously, a larger sample is needed, but those results make it look very unreliable - especially compared to an outright statement of gender.