A Google project dedicated to ending online harassment, Jigsaw, has launched the Perspective API. The new API will allow Jigsaw to partner with the developer world by providing Jigsaw tools to third parties. Jigsaw has been working on its anti-harassment tools for over a year. The tools fit within a broader initiative, the Conversation AI, which uses machine learning to programmatically detect harassment and other abusive speech.
Jigsaw and Google have partnered with major players that host and moderate online conversations. Wikipedia, the Guardian, the New York Times, and the Economist are all exploring how to protect contributors against bullying and online attacks. The goal is to determine when a conversation moves beyond honest and constructive into a toxic environment. Google and its partners fully realize the downsides that exist in unintentionally curbing speech, and solicits active feedback from third party experiments to strike a correct balance.
Through machine learning, Perspective creates a "score" that anticipates the impact particular speech may have on a conversation. For instance, "you are not a nice person" might receive an 8% similarity to toxic phrases; whereas, "you are a nasty woman" receives a 92% similarity score. In addition to the API, Google and Jigsaw are open sourcing projects, experiments, models, and research data to better determine the strengths and weaknesses of using AI and machine learning to moderate online conversations. Sign up here for API access or to learn more.
Playing with the bounds of censorship, Google and partners want to ensure intentions are clear. read through the project's values, goals, and upcoming experiments at GitHub. Check out TensorFlow and Cloud ML Platform to dig into the machine learning and AI driving Perspective.