In the wake of recent acts of extreme brutality and injustice and mass protests, we’re examining our role in perpetuating systems of inequality. We are responsible for our impact as a tech company, as a news reader, and, acutely, as a developer of machine learning algorithms for Leo, your AI research assistant.
Artificial intelligence and machine learning are powerful tools that allow Leo to read thousands of articles published every day and prioritize a top selection based on the topics, organizations, and trends that matter to you. However, if not designed intentionally, these tools run the risk of reinforcing harmful cultural biases.
Bias sneaks into machine learning algorithms by way of incomplete or imbalanced training data. Without realizing it, we miss or overrepresent certain variables and the algorithm learns the wrong information, often with dangerous outcomes.
In the case of Leo, we risk introducing bias when teaching him broad topics such as “leadership.” Leo learns these topics by finding common themes in sets of articles curated by the Feedly team. For the topic “leadership,” Leo might pick out themes like strong management skills and building a supportive team culture. However, if more articles about male leaders than female are published or added to the training set, Leo might also learn that being male is a quality of leadership. Tracking which themes Leo learns is an essential part of topic modeling that helps prevent us from reinforcing our biases or those of the article author or publisher.
It’s on us as developers to be deliberate and transparent about the way we account for bias in our training process. With that in mind, we’re excited to share what we’re working on to reduce bias at the most crucial stage: the training data
Break down silos
Collaboration among folks from diverse backgrounds helps us account for our blind spots. However, to make that collaboration possible, we need an accessible tool. The new topic modeler is that tool — designed so that anyone in the Feedly community can help curate a dataset to train Leo about topics they’re passionate about.
The topic modeler takes advantage of the Feedly UI we know and love to allow multiple users to search for articles for the training set and review Leo’s learning progress. Our goal is to connect with experts in a variety of fields to build robust topics that represent our entire community — not just the engineering team.