Weblog: Google’s Project Respect fights back against toxic speech
Machine learning and crowd-sourced knowledge helps cut abuse of LGBTQ+
Project Respect: designed for publishers and developers to help them host safer and more inclusive conversations online.
Continuing on from its work on Perspective, an AI tool developed to detect and combat online abuse and harassment, Google has created Project Respect to combat toxic speech aimed at the LGBTQ+ community.
Project Respect is a combination of machine learning and crowd-sourced knowledge: Google aims to train its system with a dataset compromised of statements and identity labels written by people who identify as lesbian, gay, queer, bisexual, non-binary and transgender. This way there is a wealth of data teaching the algorithm language associated with identity labels used in a normal manner in order to better detect when someone is using it in a negative, abusive way.
“To train these machine learning models, we need a wide variety of statements from as many sources as possible. That’s where you come in. You can be part of making this technology better by writing a positive identity statement below,” says Google.
Project Respect is designed for publishers and developers to help them host safer and more inclusive conversations online.