The disproportionate often irresponsive use of Machine learning Algorithms risks to emphasise stereotypes already present in data. Whenever data reflects biases of the broader society, the learning algorithm captures and learns from these stereotypes.
On this website we visualise how text embeddings trained on Google News represent gender.
She is an engineer, while he is a nurse.
Here you can see what gender connections the algorithm does.
In green words which are considered to be
female
, in purple
male
.