When you envision a nurse, a woman most likely pops into your mind. If you imagine an accomplished executive, on the other hand, it's quite likely you're thinking about a man.

It's not just you, though. The machine learning algorithms that target ads at us, prune our search results, or sort resumes for recruiters are all plagued by gendered stereotypes.

Algorithms that model natural language transform words into vectors, and similar words should be near each other in this vector space. Unfortunately, our models have learned to capture the biases present in the real-life data on which we train them. In word embedding space, for example, the relationship between "he" and "she" mirrors that of "programmer" and "homemaker". When we train our machine learning models on embeddings like these, a recruiter searching for "programmers" will leave female resumes at the bottom of the pile.

The following visualization allows you to define a category and see how words in that category relate to gender, by projecting them onto the axis representing gender in word embedding space. It is intended to encourage you to think critically about the tools you use and to consider carefully before treating anything as a black box. Feel free to explore the categories, or create your own.

Additional parameters