In December 2018, I traveled to Montreal for the NeurIPS (Neural Information Processing Systems) conference for my job at ATB Financial. For those who aren’t in the industry, this conference is the largest conference in the world for artificial intelligence research.
While I was there to present a paper I had collaborated on with my colleagues in the workshop on challenges and opportunities for AI in financial services, I also had the opportunity to check out other workshops and sessions. I was thrilled to find the workshop on AI for social good.
While the highlight was indisputably the fact that I caught a surprise performance by Yo-Yo Ma (!!!), I also enjoyed perusing the posters in the room, and one particular poster caught my eye. It was a description of the collaboration between Element AI and Amnesty International, and how they had come together to collect data on how often women in politics and journalism experience abusive and problematic tweets on Twitter. They called it Troll Patrol.
I was so inspired by this work, because of my volunteer work in the field of women in politics, and I was eager to dive in and think about how I could help–and how I could perhaps turn this amazing research into a practical tool for encouraging more women to run for political office. And here is the most important lesson of this article: take vacations. Because while I was relaxing by the pool in Mexico a few weeks later, I had an idea: what if we could take this idea and build a Twitter bot that could detect abusive and problematic tweets directed towards women during a certain election, or in a certain jurisdiction, and then send out a positive tweet (or “positivitweet” as my friend at SAM desk said) for every bad tweet it detected?
So we did it. I found an amazing collaborator who is in the know about machine learning methods and who was willing to help us build the bot, and we deployed it during Alberta’s spring 2019 provincial election. This news story does a great job of covering the system and showing what it can do, and features my collaborator and I explaining it.
Our next step is to gather some feedback, refine the bot and deploy it again during Canada’s fall 2019 federal election. I’d love to hear what you think!