Machine Learning by Communities, for Communities

When was the last time you thought about that blank text field where members of your community can leave comments? That text field and blinking cursor are the closest we have to pauses between human interaction on the internet. In this episode, Perspective’s product manager, Cj Adams, encourages us to think about how we might innovate that text field and blinking cursor in hopes of having more inclusive, difficult, and natural conversations.

Cj also explains how Perspective can help. Its API has a variety of ways that can be implemented, all with the goal of perceiving the impact a comment might have on a conversation. But Cj also explains that machine learning is is not flawless, and he reminds us that the humans responsible for training it are what encourages its actual biases. So, just like with any other tool that you consider for your community, think about how you can implement it with your community in mind and not as the be all, end all solution for creating better conversations.

Cj also shares:

  • How Perspective creates a conversation around moderation
  • Why Perspective is a tool for communities small and large
  • What machine learning does when it’s “really stupid”

Big Quotes

Using machine learning to encourage community-minded conversations: “People can be a little too focused on how good is the machine learning, or what is the exact technology behind it, [but] a lot of it comes down to what’s the actual user experience? What does that feel like to type something in? That idea of just giving someone a moment when they submit [a comment] to give them some feedback instead of moderating after the fact is a really powerful thing. … It’s a little bit more like when you’re in a human conversation.” –@adamscj

The bias in teaching machines: “There are three main types of machine learning. Supervised learning, unsupervised learning, and then reinforcement learning. What we are talking about here is supervised learning, where you put in a bunch of training data and machine learning is trying to identify patterns in that data and then be able to, in this case, classify new examples to say what this does or doesn’t look like. One of the things that you have to recognize is that it’s going to be as smart or as dumb as what it learns from. … Machine learning is only as right as whatever the data was that it was trained on.” –@adamscj

Addressing the challenge of an empty text box and a blinking cursor: “In a lot of forums, we have an empty white box with a blinking cursor, and that’s how we talk to other humans. … Is there anything we could do here that’s a little more creative, a structure that might facilitate less toxicity, that might facilitate people to understand the humanity of the people they’re talking to? I don’t know what that answer is there. I am excited by trying to figure out how we might nudge people in a direction towards understanding people they disagree with, listening to people, learning and sharing their views in a way that fosters understanding and fosters people being able to keep talking even when they disagree.” –@adamscj

About Cj Adams

Cj Adams is a product manager at Jigsaw, part of Alphabet (Google), focused on building technology to make people safer around the world.  Since 2015 he has been the product manager of Perspective, an API that helps communities and platforms use machine learning to protect voices in conversation.

Previously he led Project Shield, a free DDoS mitigation service for news organizations and before Google, he helped build a national confidential hotline for victims of human trafficking at a non-profit called Polaris

Related Links


Your Thoughts

If you have any thoughts on this episode that you’d like to share, please leave me a comment, send me an email or a tweet. If you enjoy the show, we would be so grateful if you spread the word and supported Community Signal on Patreon.

Leave a Reply

Your email address will not be published. Required fields are marked *