Ericsson Innovation Awards finalist team Alfred uses AI to change the future of truth
Fake news. Post-truth. Echo chambers that reinforce cognitive biases. Pseudoscience. Chatbots pretending to be people. Targeted advertisements designed to influence your political opinions.
Misinformation, in all of its forms, is threatening to tear our societies apart. Our minds can’t handle communication at the scale modern technology enables. There are too many people with too many agendas saying too many things for us to keep everything straight.
And now that AI-powered robots entering the fray, our odds are about to get worse. Their automated, data-driven algorithms ensure that they will dominate online conversation in both volume and persuasiveness.
As students studying computer science and AI, we care a lot about this. The systems we’re learning how to build are already being used by organizations like Cambridge Analytica. As AI development becomes increasingly democratized, bad actors will increasingly use these technologies to spread computational propaganda that influences public opinion at scale.
Introducing Alfred, your personal information concierge
We want to even the scales. We believe individuals should be empowered to control their own infosphere, to be aware of attempts of others to influence their opinion on a topic, to have easy ways to fight back by consciously engaging with material from diverse viewpoints and developing nuanced understandings of complex issues.
That’s why we’re excited to build Alfred, your personal information concierge, for the Ericsson Innovation Awards. Alfred has two primary functions: flagging misinformation and recommending alternative content. Alfred will eventually fight biased news, echo chambers, fake news, and bot-generated content. For now, Alfred programmatically identifies and disrupts echo chambers by recommending perspective-broadening content.
Our story began at Stanford with research to determine whether echo chambers actually exist and prototypes of methods to disrupt them. We analysed the Michigan dataset, published by the Computational Propaganda Project at Oxford, which contains 63,277 political tweets Michigan-based users tweeted between Nov 1st and Nov 11th, 2016 in the lead up to the 2016 US Presidential Election. We translated the tweet data into a graph of user interactions, characterized the networks, identified echo chambers, and designed basic recommendation systems that suggested engaging tweets from other echo chambers. We experimented with several ways to automate this process and proposed metrics for evaluating the quality of our recommendations. We presented this work at a workshop at NIPS, one of the world’s most premier machine learning conferences.
Next, we developed a recommendation engine using real-time data. We built and deployed a TwitterBot that looked for users tweeting biased news articles and recommended alternative articles on the same topic but from a different source. During coffee shop meetings between classes, and staying up late into the evening, our team wrote code to scrape bias scores from mediabiasfactcheck.com and over 130,000 news articles from various news sites. We used language identification, keyterm extraction, topic modeling, website popularity rankings from Alexa, and when the article was published to recommend engaging content through direct messages and replies. We learned a lot about what to do and, most importantly, what not to do.
We are now in the midst of developing a browser extension that automatically flags biased information and recommends alternative content. This combines all of our work so far into one platform that works on sites all across the Internet, not just Twitter. When complete, the Alfred browser extension will help you flag fake news, identify and characterize bot-generated content, and provide other features to fight misinformation. Our intention is to start simple and iteratively add complexity over time, ensuring Alfred will help anyone, anywhere find, validate, and share truthful content.
It’s been an amazing journey so far. We’ve learned so much from our mentors, from experimenting, and from thinking about how to define and categorize misinformation. The Ericsson Challenge has offered the perfect amount of structure: enough questions to force us to think through all the angles, but enough freedom that we can freely exercise our creativity. We’re super excited to continue our learning journey through the Finals, for the trip to Sweden, for the chance to tell our story and get concentrated feedback from the best Ericsson has to offer.
Winning EIA would be a dream come true. Further mentorship, funding to build a real product, and the branding boost would help us take Alfred to levels we’re currently only dreaming of. When Gutenberg invented the printing press, he proclaimed “A spring of truth shall flow from it: like a new star it shall scatter the darkness of ignorance, and cause a light heretofore unknown to shine amongst men”.
We hope Alfred can contribute towards helping modern information technology realize that vision.
Watch team Alfred in the Grand Final on May 17
Team Alfred is on of 4 finalist teams competing to win the Ericsson Innovation Awards and 25,000 EUR. Click here to mark your calendar for the Grand Final livestream on May 17 at 6 pm GMT (1 pm US Eastern Time).