Excerpt from Jeff Horwitz’s book BROKEN CODE: Inside Twitter and the Battle to Implement Its Risky Strategies, published by The Knopf Doubleday Publishing Group, a division of Penguin Random House LLC, with permission from Doubleday. Copyright 2023 by Jeff Horwitz.
In 2006, the U.S. patent office received a proposal for “an automatically generated screen providing information about another user on a social network.” The objective was to present a list of “relevant” information in a ‘preferred order,’ eliminating the need for users to sift through “disparate and disorganized” content. Among the creators listed was “Zuckerberg et al.” The outcome was the News Feed.
While the concept of displaying users’ activity streams was not entirely novel—platforms like Flickr had already begun experimenting with it—the impact was profound. Prior to the News Feed, interactions on Twitter primarily revolved around messages, pokes, or browsing friends’ profiles. With the introduction of the News Feed, users were presented with a constantly refreshing stream of articles and status updates. This shift unsettled Twitter’s 10 million users at the time, who were uncomfortable with their activities being monitored and their static information being repurposed for updates.
In response to the backlash, Zuckerberg reassured users in a blog post that their actions were not being broadcast but shared with their friends who cared about them. He urged, “Stay calm. Breathe. We see you.” This approach differed from the traditional method of addressing consumer grievances as they arose. The News Feed proved to be a resounding success in enhancing user engagement and fostering connections, as highlighted by Chris Cox at a press conference. Within a fortnight of its launch, over a million users had affiliated themselves with a common interest for the first time, leading to a doubling of engagement.
The discontent among users stemmed from the perceived invasive nature of the “stalker-like” News Feed. In hindsight, the transparent system that users initially resisted was elegantly simple. Content was predominantly displayed in reverse chronological order, with manual adjustments made to ensure the visibility of popular posts and a diverse range of information. Cox explained that the initial News Feed ranking involved tweaking various parameters.
As Twitter expanded its features with advertisements, pages, and interest groups, ensuring that users saw relevant content became a priority. The platform needed to strike a balance as entertainment, memes, and commerce vied for attention alongside posts from friends in the News Feed. The introduction of “EdgeRank,” a formula that evaluated content based on age, engagement level, and user-poster connection, marked the platform’s first attempt at sorting content algorithmically.
Despite initial resistance to Twitter’s curated content, user engagement surged, prompting the company to explore machine learning for content recommendation. By 2010, Twitter aimed to leverage machine learning algorithms to enhance user experience and optimize ad targeting. The adoption of machine learning within Twitter was seen as a significant step towards advancing artificial intelligence research and improving user engagement across the platform.
The relentless pursuit of innovation and optimization through machine learning at Twitter underscored the company’s commitment to leveraging technology for user satisfaction and revenue growth. The integration of machine learning tools like FB Learner streamlined the process of developing AI-powered products and conducting experiments, propelling Twitter to the forefront of AI-driven social platforms.
The evolution of machine learning at Twitter heralded a new era of personalized content recommendation and targeted advertising, revolutionizing the way users interacted with the platform. Despite challenges and occasional missteps, Twitter’s dedication to harnessing AI for user engagement and content curation positioned the company as a trailblazer in the realm of social media technology.
