Fake news. We can’t define it, but we know it when we see it. In this escalating online propaganda war nefarious clickbaiters have until now avoided the eyes of the platform they used. Twitter recently announced it was going to begin work on AI to curb hate speech on its service. Jigsaw, a Google partner, has offered Perspective AI. It offers an AI perspective on whether someone is being a troll or not.
Currently Wikipedia and the New York Times are implementing the Perspective API to help moderating comments and user activity. The AI itself is in need of some trianing. Researchers from the University of Washington fooled the AI by misspelling the words to fool the parsing process.
False positives abound too. Calling someone not an idiot still triggers a toxicity score. This is an early iteration however. Perspective as an ideal service is really aiming for a human like understanding of conversation. Presently Perspective makes moderators tasks easier by flagging more content that might previously have been missed.
Perspective AI is still learning
As more time to train the AI unfolds we’ll see better results. Human idioms, memes and t r i c k s l i ke t h i s evolve fast. CJ Adams, Product Manager for Perspective, sent a statement to Ars Technica rightly acknowledging the difficulty of the task and welcoming collaboration.
“It’s great to see research like this. Online toxicity is a difficult problem, and Perspective was developed to support exploration of how ML can be used to help discussion. We welcome academic researchers to join our research efforts on Github and explore how we can collaborate together to identify shortcomings of existing models and find ways to improve them.
Perspective is still a very early-stage technology, and as these researchers rightly point out, it will only detect patterns that are similar to examples of toxicity it has seen before. We have more details on this challenge and others on the Conversation AI research page. The API allows users and researchers to submit corrections like these directly, which will then be used to improve the model and ensure it can to understand more forms of toxic language, and evolve as new forms emerge over time.”