Bradley Merrill Thompson, Member of the Firm in the Health Care & Life Sciences practice, in the firm’s Washington, DC, office, authored an article in The Journal of Robotics, Artificial Intelligence & Law, titled “Unpacking Averages: Searching for Bias in Word Embeddings Trained on Food and Drug Administration Regulatory Documents.”
Following is an excerpt (see below to download the full version in PDF format):
Often when we talk about bias in word embeddings, we are talking about such things as bias against race or sex. This article talks about bias a little bit more generally to explore attitudes we have that are manifest in the words we use about any number of topics.
Bias Evaluation Using Sentiment Analysis
There are many different ways to evaluate potential bias in word embeddings, but I did not want to do a survey article where I talked briefly about all of them. Instead, I thought I would pick just one approach for illustration. The one I picked is perhaps the simplest, which is an evaluation of the word embeddings using a model for positive versus negative sentiment. In other words, I am looking to see whether particular word embeddings have a largely positive or negative connotation.
If words that should be regarded similarly have significantly different sentiments or connotations, that would be evidence of bias. In other words, if the word “Black” as an adjective for people has a largely negative connotation while the word “white” as an adjective for people has a largely positive connotation, that would be some evidence that the embeddings, trained on what people have written, have absorbed from that training data a bias against Black people.
However, I am not going to use race as my example in the analysis below. For one thing, race is rarely discussed in the documents that I am going to examine—Food and Drug Administration (FDA) documents—apart from a handful of documents specifically on race. I will leave you to draw your own conclusions from that. Instead, I am going to look for bias in other topics.