The core natural language models have been trained on the world’s knowledge, with some content in the training data going back as far as centuries and including multiple biases “inherited” from previous generations.
While more recent content is also available in the data set, there is a risk that the product created with the help of AI may be biased against women, people of color, and other groups. Given the scale and speed of AI deployment, this can have large-scale effects.
Thanks to advances in machine learning and AI, the world is now entering a phase where AI capabilities will be available and widely used by businesses in many industries: e-commerce, ticket sales, publishing, media, marketing, customer service, healthcare, finance, logistics, and more.
Our goal is to create a set of resources to help anyone working with AI identify potential bias in content and flag it to businesses and individuals using AI in their work.
The dictionary is a list of terms that are connected to different types of bias and can be used to help identify and flag bias in content. The DEI dictionary will be available for anyone for free on the internet through Github and other channels to then be plugged into their workflows.
Join us in this collaborative effort by contributing in the following ways: