As a collaboration between Columbia Journalism School and Stanford University’s School of Engineering, the Brown Institute for Media Innovation awards its “Magic Grants” to projects that work between the two disciplines, creating new forms of media and new ways to serve the public interest.
This was a milestone year for the Magic Grant program, which received a record number of applications, nearly doubling the count from last year. There were so many great ideas, judges had a particularly hard time selecting the winners. Not surprisingly, many of this year’s submissions had to do with AI and the ways in which AI-inspired technologies could meet the current and unique demands of journalism and media.
A total of 10 Magic Grants and 3 smaller Magic “Seed” Grants were awarded. This year, the projects span a wide range of areas – one team is developing an AI-assisted platform to analyze cross-lingual, country-specific perspectives on global news stories. Another is creating a discussion layer on top of arXiv, the largest open access preprint repository. And another is looking at the social costs of AI through the lens of new forms of automation in customer service.
“It is exciting to see projects addressing a variety of societal issues via media, journalism, and innovative applications of new technologies,” said Maneesh Agrawala, professor of computer science at Stanford Engineering and the West Coast Director of the Brown Institute.
Established in 2012 with a generous gift from the famed Cosmopolitan Magazine editor Helen Gurley Brown, the Brown Institute honors the memory of Helen’s late husband, David Brown, a successful filmmaker and alumnus of both the Columbia Journalism School and Stanford University. The institute is committed to fostering unique interdisciplinary collaborations, sparking the “magic” that arises from combining diverse perspectives and expertise.
“While the winning teams may have very different topics, they’ve all proposed working across disciplines,” said Mark Hansen, the East Coast Director of the Brown Institute. “To date, we’ve funded over 325 people through our Magic Grant program – the most gratifying part of my job is watching the teams in each cohort grow by learning from each other. Where will these new grantees take us?”
We warmly congratulate our Magic Grant winners. Supporting your groundbreaking work is an honor, and we can’t wait to see what you produce in the year ahead! Congratulations! And here’s to the magic you’ll create!
—
2024-2025 MAGIC GRANTS
alphaXiv: Open Research Discussion Directly On Top of arXiv
Rehaan Ahmad, MS Candidate in CS (Stanford); Raj Palleti, MS Candidate in CS (Stanford); Daniel Zamoshchin, MS Candidate in CS (Stanford); Daniel Kim, MS Candidate in CS (Stanford); Advised by Professors Sebastian Thrun and Michael Bernstein (Stanford)
alphaXiv is a discussion platform for students and researchers to ask questions and exchange ideas directly on top of papers from arXiv, the largest open access preprint repository with nearly 2.4 million academic papers. Early researchers often find it hard to understand new research papers, and it’s even harder for them to reach authors and academics to ask questions. With alphaXiv, the team hopes that anyone from distinguished researchers to curious students can interact with authors and partake in high-quality discussions directly on top of papers, connecting scholars from all over the world.
Cross-Lingual Multi-Perspective News (Bicoastal)
Jialiang Xu, MS Candidate in CS (Stanford); Tiffany Weiyang Le, MS ’24 Journalism (Columbia); Advised by Professors Ty Lawson (Columbia), Jennifer Pan (Stanford), and Monica S. Lam (Stanford)
Researchers from Columbia and Stanford are creating the first AI-assisted platform to deliver breaking news from around the world while analyzing the differences in their reporting points of view. The analysis draws from local news outlets in each country’s native language. This platform gives consumers, journalists, and social scientists better access to on-the-ground perspectives in foreign countries as well as an understanding of how international events are portrayed across the globe.
Business Monitor to Empower Worker Stories
Max Siegelbaum, MS ’16 Journalism (Columbia); Lam Thuy Vo, MS ’08 Journalism (Columbia)
The Business Monitor is an accountability-focused database built from datasets created by federal, city, and state agencies in New York aimed at telling the stories of workplace abuse. This project builds on Documented’s Wage Theft Monitor, New York State’s first searchable list of confirmed cases of wage theft.
DataTalk: All Documents and Data, All at Once, All Verified (Bicoastal)
Shicheng Liu, PhD Candidate in CS (Stanford); Eryn Davis, MS ’24 Journalism (Columbia); Sajid Omar Farook, MS Candidate in CS (Stanford); Leah Harrison, MS ’93 Journalism (Columbia); Advised by Professors Serdar Tumgoren, Cheryl Phillips, and Monica S. Lam (Stanford)
Investigative journalism often relies on the ability to mine diverse data sets, with both structured and unstructured forms. In collaboration with Stanford’s Big Local News initiative, the Stanford Department of Computer Science, and Columbia Journalism School, DataTalk: All Documents and Data, All at Once, All Verified aims to develop trustworthy conversational agents for journalists to uncover insights from such hybrid data sources using natural-language queries. Building on a novel programming language, SUQL (Structured and Unstructured Query Language), the project will expand the current research prototype into a full development framework to enable non-AI experts to quickly deploy tools to probe complex datasets and fact-check results to produce groundbreaking stories.
African History from the Bottom Up with LLM-Augmented Agents
Sina J. Semnani, PhD Candidate in CS (Stanford); Kwame Nyarkoh-Ocran, MS Candidate in CS (Stanford); Ashley Celestine Kamdom Tamdjo, BS Candidate (Stanford); Chidubem Nwosu, BA/BS Candidate; Advised by Professors Robin P. Chapdelaine (Stanford), Monica S. Lam (Stanford), Trevor R. Getz (SF State)
African History from the Bottom Up with LLM-Augmented Agents is a joint project between the Center for African Studies and the Department of Computer Science at Stanford University. Together, the team will create HistoryChat, an AI tool that enables scholars, students, and teachers of history to interact with and study historical corpora. The first subject of this project is The African Times, a unique, historically significant newspaper produced by members of the transatlantic African diaspora in the late 19th century. HistoryChat will be openly accessible so that it can be applied to any historical corpus, addressing the intellectual gap in the history of other underrepresented communities.
Improper Conduct
Kristen Lombardi, Director of Columbia Journalism Investigations; Smaranda Muresan, Sr. Researcher at the Data Science Institute at Columbia; Jeffrey Fagan, Professor at Columbia Law School
The team will use an original dataset of prosecutorial misconduct cases in Ohio as a test case for developing artificial intelligence tools based on large language models (LLMs), in a collaborative human-LLM framework, to automate, update, and lay the foundation for a first-of-its-kind public database tracking improper behavior by prosecutors during criminal trials. Through these AI tools, they will leverage the legwork of our investigative reporters to publish a recent repository of these Ohio cases – a prototype of what the team hopes will become an interactive database enabling watchdog journalists, legal researchers, and others to identify patterns of improper and at times illegal behavior by prosecutors in this state and beyond. Along the way, they will produce a user study to assess the utility of the human-LLM collaboration framework and models, as well as a guidebook to help journalists replicate this database in states with online court dockets similar to Ohio’s.
HairFlow – Bridging Representation Gaps in Image to 3D Hair Generation
Sarah Jobalia, PhD Candidate in CS (Stanford); Yitong Deng, PhD Candidate in CS (Stanford); Advised by Professor Ron Fedkiw (Stanford)
Hair is one of the physical traits that is most often used to express personality and identity. Despite this, building 3D hairstyles for virtual characters today relies on a multitude of complicated tools, making it one of the most difficult parts of the character creation pipeline. HairFlow aims to make generating hair easy, allowing users to create an artist-ready 3D hair groom from just a few cellphone-quality photos. Additionally, HairFlow is the first system to incorporate an understanding of diversity within hair, creating accurate models for any hair type regardless of curl pattern or texture. This system will allow artists of any technical background to generate characters with a diversity of hairstyles, making character creation more accessible and representative.
Justice Delayed
Shalaka Shinde, MS ’23 Journalism (Columbia); Tazbia Fatima, Dual MS ’23 Journalism and CS (Columbia)
Justice Delayed will be an essential database that will catalog long-pending criminal cases across 650+ district courts in India. The data, and the software system it will fuel, will equip journalists from remote parts of the country with the tools to scrutinize every aspect of the criminal justice system within their respective regions. By extracting and organizing data from district-level courts, the project aims to pinpoint cases contributing to India’s staggering backlog of nearly 50 million court cases, primarily concentrated in district-level jurisdictions.
Measuring Silence on Social Media
Dorothy Zhao, PhD Candidate in CS (Stanford); Jordan Troutman, PhD Candidate in CS (Stanford); Advised by Professors Michael Bernstein and Diyi Yang (Stanford)
As a part of Measuring Silence on Social Media, the team will conduct a broad census of what perspectives are being systematically under- or over-represented across various online communities. To do so, they propose a novel human + AI pipeline that leverages large language models and survey methods to measure this silencing effect.
Digital Archives of Refugees’ Self-Representation
Pamela Martinez, MFA Candidate (Stanford); Artem Arzym, MS Candidate (Stanford); Advised by Professor Natalia Almada (Stanford)
Digital Archives of Refugees’ Self-Representation is a digital archive of self-represented audiovisual media that attempts to diversify journalistic practices by recentering the agency of those traditionally excluded from humanitarian discourse pertinent to the U.S.-Mexico border. The team will be investigating how migrants use social media to share their stories, uncover which groups of migrants don’t, and work with translation and transcription software and local community organizations to increase marginalized communities’ accessibility to storytelling technology. In that process, they investigate what long-term consent of data can mean in a seemingly permanent digital space for individuals in legally precarious circumstances whose needs can change.
2024-25 MAGIC “SEED” GRANTS
This Call May Be Recorded
Professor Habiba Nosheen (Columbia) and Theo Balcomb, Creator of The Daily
This Call May Be Recorded is an investigative podcast by Emmy and Peabody Award-winning journalist Habiba Nosheen and Theo Balcomb, also a Peabody Award-winning journalist, and the creator of The Daily from The New York Times. The podcast will go inside the world of people’s most haunting customer service problems and will untangle one thorny quandary at a time with the goal to help listeners navigate the often maddening and rapidly evolving use of AI by companies instead of human beings to deal with customers. The team will craft an easy-breezy romp through that moment of collective angst as they come up against chatbots, a fun-filled journey that helps the little people win in little ways, especially when they’re going up against AI, and offer just a bit of catharsis for anyone whose nightmares have been haunted by hold music. The project aims to make today’s often inhospitable world of customer service just a little safer to navigate. The goal is to be the voice of the people who feel unheard and along the way win some well-deserved victories for those who don’t have the time, money, or power to advocate for themselves in our increasingly automated world.
pollfinder.ai
Professor Dhrumil Mehta (Columbia); Aisvarya Chandrasekar, Researcher (Columbia); Ken Miura, MS Candidate in CS (Columbia)
During election cycles, journalists are inundated with data from surveys asking Americans who they plan to vote for and, perhaps more important, how they feel about the issues at stake. But given the sheer volume of polls and the myriad formats in which pollsters release them, it is hard for newsrooms to keep up and for journalists to know which polls to reference. pollfinder.ai aims to use large language models to help newsrooms collect and organize both horserace and issue polls so that journalists can write stories that are informed by an up-to-date aggregation of public opinion polls that can provide a more complete picture of what Americans think about a given topic.
Lethal Italy
Natasha Caragnano, MA ’24 Journalism (Columbia); Sacha Biazzo, MA ’24 Journalism (Columbia)
Lethal Italy confronts the devastating environmental impact of industrial pollution in southern Italy, using groundbreaking data and multimedia storytelling to raise global awareness and enforce sustainable practices.