Newswise — Eunji Kim caught Chris Wiggins’ attention when she started talking about memes at an event on AI and democracy.
“A string of fellow technologists — including myself — got up and made various technical claims about how technology is useful for understanding the information ecosystem,” says Wiggins, an associate professor of applied mathematics at Columbia Engineering. “Then Professor Kim got up and said, in academic terms, none of you has any idea how Americans actually interact with information.”
Now, the professors are teaching “Persuasion at Scale: Causal Inference, Machine Learning, and Evidence-Based Understanding of the Information Environment”, a new course at the intersection of data and political science. Both are members of the Columbia University Data Science Institute (DSI).
“It’s very common for researchers in computational social science to use big data to draw conclusions about society,” says Kim, an assistant professor in political science who uses quantitative methods to study political communication and public opinion in American politics. “But if you don’t consider political context and meaning that influences your data, the analysis will be incomplete and your conclusions could be wrong.”
In their new course, Kim and Wiggins aim to give students the tools necessary to rigorously analyze the impact of political communication—from campaigns and advertisements to partisan media and social media.
“Persuasion is happening at scale on information platforms,” Wiggins says, "so we now have the chance to understand this question statistically.”
Machine learning — a branch of AI — will be a pillar of the course. Students will learn how this technology underlies the content recommendation and moderation systems that drive information platforms. They will also use machine learning techniques to interpret datasets.
Equipping students to challenge conventional wisdom
Persuasion at Scale pairs an examination of the research literature on political persuasion with a survey of the statistical methods necessary to make sense of complex datasets. Students will use real-world data to quantify the effects of partisan media, social media, advertising, and political campaigns while taking a historical view on the development of persuasion architectures.
“When we actually bring data to these questions and look at them objectively, we sometimes find that conventional wisdom isn’t supported — or that it’s wrong,” Kim says. For example, op-eds, blog posts, and cable news monologues often assume that Americans are living in partisan echo chambers, with half the country watching Fox News and the rest watching MSNBC.
“If you look at actual behavior-level data, the extent to which echo chambers exist is very limited because most people do not consume news to begin with,” Kim says. “Consumption of news content is very low relative to other media, like sports or entertainment.”
Another counterintuitive finding is that political campaigns — even those that spend hundreds of millions of dollars — often don’t have a substantial impact on voter choices.
“There's a lot of discrepancy between what people believe versus what empirical social science has been discovering,” Kim says.
Students will learn to employ causal inference techniques to distinguish between correlation and causation in real-world datasets.
An opportunity for interdisciplinary collaboration
The course, which is offered under the Provost’s Cross-Disciplinary Frontiers Initiative, is open to undergraduates across the University. The professors hope to offer roughly 70 students from diverse backgrounds the chance to collaborate and learn from each other.
“Engineering students don't often take classes in political science, and our own social science students do not often take many math classes,” Kim says. “These types of classes are critical for them to learn how to fix the many complex problems facing our society.”
For Wiggins, the course is an opportunity for students to bring quantitative rigor to a domain that’s often understood through conventional wisdom and unjustified assumptions.
“I think it's useful to zoom out and see how persuasion — whether it's political persuasion or marketing — has some universal aspects that we can understand using mathematics,” he says. “By combining that context with the language of probability, we hope to enable students to look past inflammatory anecdotes in order to think methodologically and historically.”
RSS