George Gerbner Postdoctoral Fellow Baird Howland Analyzes the News With AI

Baird Howland arrived at the Annenberg School for Communication five years ago, driven by big questions about the media’s influence on Americans’ political views. Why do some stories, like Hillary Clinton’s email controversy or Russia’s election interference, stick in the public consciousness, while others fade away? How do Americans’ particular media diets lead to different levels of political engagement and support? Is “fake news” as prominent as it seems?

A former data scientist, Howland hoped to use data to look for answers in his doctoral work. “I wanted to think about and critique media systems and their role in politics, but do so in a data-driven, computational way,” he says. “With its interdisciplinary focus, Annenberg was the perfect place to be.” 

He joined Stevens University Professor Duncan Watts’s Computational Social Science Lab (CSSLab) and dove into investigating Americans’ media diets. With Watts, he analyzed the online and television news consumption of hundreds of thousands of Americans to investigate the phenomenon of political echo chambers, finding that cable TV news contributes to more partisan segregation in news audiences than any combination of far-right or far-left sites circulating on the internet. Howland also helped create the lab’s Mapping the (Political) Information Ecosystem Dashboard: a set of four interactive data visualizations that map where Americans get their news and the extent of political “echo chambers” in the U.S.

The release of ChatGPT and the large language model (LLM) it uses was transformative in his research. “That made a whole new world of things possible,” he says. “I was always trying to apply literary theory ideas to mass communications within a big data approach, and for a while, it felt impossible.” He almost gave up. “But then, when these fancy AI models, which can reason about implied meaning, came out, it made the problem I was working on tractable,” he says.

For his dissertation, he used LLMs to track narratives in political discourse, such as the surge of news stories about immigration in the lead-up to the 2016 presidential election. His main finding was that over the past decade, right-wing news outlets were more effective storytellers than mainstream media. Now, as the George Gerbner Postdoctoral Fellow at Annenberg, Howland is expanding the work he started as a student: looking at which stories are covered in the news and how.

With Annenberg doctoral candidate and CSSLab member Timothy Dorr, he has been analyzing media coverage of the conflict in Gaza, using the LLM GPT-5 to sift through tens of thousands of articles about the conflict from U.S. and U.K. news outlets.

“There's been a lot of claims that the media coverage of Israel and Gaza has been skewed, effectively sanitizing a genocidal military campaign,” he says. “We’re taking advantage of the massive data sets of news articles that we have in the lab and comparing how violence was depicted in the mainstream media when the victim has an Israeli identity versus a Palestinian identity.”

Howland has also been analyzing the digital messaging of the current Trump administration compared to that of previous presidential administrations, in collaboration with CSSLab data scientist Grace Jennings.

“Whitehouse.gov has a digital briefing room, and each administration posts news releases about their activities,” he says. “We created a dataset that includes the current administration's version of the website, the Biden administration's version, and the first Trump administration's version. We're looking to explicitly compare the current administration's messaging on the official White House website to the Biden administration and the rest of the mainstream media.”

The George Gerbner Fellowship, named in honor of the school’s second dean, is awarded in alternate years to a graduate of Penn’s Annenberg School for Communication or USC’s Annenberg School for Communication and Journalism. The faculty of the partner school selects the recipient from a group of applicants, who is then in residence at Penn for the duration of the fellowship.

Howland is glad he has this time to further his research.

“In this time of rising authoritarianism and ethno-nationalism, the power of political narratives and the media ecosystem around them demands careful scrutiny now more than ever,” he says.