Growing up in Taiwan, Herbert Chang '18 saw firsthand how technology and politics could be tightly intertwined.
"Because of its contentious history with China, Taiwan receives a ton of misinformation, but also boasts a high freedom-of-speech index. It's a particularly interesting space," he says, "to study how technology and democratic values come together."
Now, as an assistant professor in the Quantitative Social Science Program, Chang studies the role of technology and social networks in politics and human relationships, with a special interest in the impact of artificial intelligence.
This fall, working with a PhD student and five undergraduates as part of his lab and new course, AI and Machine Learning for Social Science, Chang led two groundbreaking studies that examine the role of AI and social networks in political engagement online. The resulting papers are under review by prestigious journals.
"This is the first U.S. election that coincides with the release of sophisticated generative AI," Chang says. This technology is being used to create images and memes, which are then uploaded to social networks like Twitter, Instagram, and TikTok. "This is really important to understand—especially as an analysis unbiased by the results of the elections," he says.
In the first study, the team investigated how AI-generated content influences political discussions on Instagram. With LLM-based tools, they collected more than 500,000 Instagram images and found that AI-generated content alone does not increase engagement among users—but when AI-generated content is combined with a meme, it's more likely to go viral. In other words, when an AI-generated image is overlaid with text or given some kind of cultural context, it is much more effective at stirring up discussion.
"Generative AI and the meme format produces a synergistic effect," Chang says.
"Our paper definitely challenges assumptions that AI will dominate online discourse," says Mingyue Zha '27, who took part in the research with fellow students Maya Magee '25, Sean Noh '28, Ben Shaman '26, Tracy Weener '26, and Guarini PhD candidate Chiyu Wei. "However, it also highlights the potential of meme content that is AI-generated, which does synergize and increase social media engagement."
One implication is that thanks to platforms like Instagram, where mass exposure isn't limited to politicians and legacy media, political discourse is shifting away from traditional media. "There's more opportunity for 'ground up' discussion by organizations who aren't MSNBC or Fox News," Chang says.
Another (and perhaps more sinister) implication is that AI is actually "mediating our content and information quite significantly, especially in this political context," Chang says.
Thanks to AI-powered meme-generating tools like MidJourney, users can simply type in general prompts about what kind of meme they'd like to create, and the system automatically generates one for them. But users don't have much ability to tweak the resulting meme, meaning the AI technology is ultimately responsible for it—and, therefore, for shaping the increased chatter around it.
The researchers found that about 15% of memes were generated by AI. Additionally, when considering synthetic memes versus normal memes over time, the biggest difference occurs during Donald Trump's conviction.
The second ongoing study offers a large-scale, systematic analysis of the influence of TikTok and Instagram on political discourse. The team collected more than 800,000 Instagram posts and 1 million TikTok posts curated from election-related key words, and has begun analyzing divergences across the two platforms.
Understanding their differences is important because while Instagram and TikTok are the two largest visual social media platforms, they operate in different ways. Instagram content is driven by social networks, meaning a user will be exposed to content by their online "friends" or followed accounts. TikTok, meanwhile, is algorithm driven, so users are exposed to content based on machine learning—technology chooses what they see based on a variety of datasets and past online behaviors.
"Pragmatically, this means that if you're a politician or you're trying to start a social movement, your strategies on the two platforms would be different," Chang says. "On Instagram, you might want to leverage the social network and work with big players to gain solidarity across different audiences. But that might not be effective on TikTok, where you'd want to invest more broadly in individual micro-influencers."
The team is observing that Instagram is used more for sharing breaking news from established media accounts, whereas TikTok has more in-depth commentary from individuals.
"My students are incredibly talented and come from diverse backgrounds—from pre-med to computer science to QSS," Chang says. "They are also perfect collaborators. As frequent users of these platforms, they have incredible insight into how these platforms work and strategies for generating visibility."
"We all worked on different parts of the project based on our experiences and interest, so it was super rewarding to see how everything came together to create new insights that show how this election is taking place over social media," Noh says. "It's a constant learning process, and it's been so cool to be studying something so important and relevant while it's actually happening."
This year's presidential election feels especially significant to Zha, who recently obtained her U.S. citizenship after living in the country for the past 10 years.
"Not only is this the first U.S. election where I'm able to vote, it's also one of the most pivotal elections of our lifetimes, especially since many issues on the ballot will have huge ramifications on the future of my generation," she says.
"One of the most rewarding aspects of being part of this research has been gaining a deeper understanding of how people discuss and engage with electoral issues, as well as how social media platforms influence and mediate this discourse."
Zha also found inspiration in fellow students. "We're able to support each other's work and challenge each other, achieving far more as a team than we could alone," she says. "Herbert's mentorship and encouragement have been instrumental in making this collaborative environment possible."
Chang will continue live-tracking the U.S. presidential election using a dataset curated for his course this fall. "I want students to know that their projects at Dartmouth are directly relevant to the world right now," he says.
At the end of the day, the irony that he wouldn't be able to conduct this research without machine learning, and that there also wouldn't be any research to conduct without AI, isn't lost on Chang.
"AI can be used to generate fake content or misinformation, but it can also be used as a form of civic engagement, and to help identify malicious trends," he says. "It's a technological arms race I'm excited to continue navigating with my students."
