Kenny Peng, a Siegel Fellow and a computer science PhD student at Cornell Tech, wants to know how algorithms can help us actually understand the world. He looks at the messy side of social media algorithms, like polarization and misinformation, and asks a bigger question: In a world where AI answers almost anything, how do we help people figure out the right questions to ask?
Consider New York City’s high school application process: eighth graders can choose from more than 800 programs. That sounds like an extraordinary range of possibilities. But for students who don’t know what to look for, or which questions to ask, more information can create another kind of disadvantage.
To address these problems, Peng wants to develop a “scientific theory of AI” that peels back the hood of large language models. By studying what happens inside these models, his research works to uncover not only how AI arrives at an answer, but also uncover patterns in human behavior, reasoning, and decision-making. What makes a social media post engaging? What kinds of math problems give students trouble? What can the internal workings of a chess-playing algorithm reveal about strategy?
Ultimately, Peng’s work asks us to look beyond the chatbot as the default way we experience AI. What if language models could help us make connections across the enormous amount of knowledge already out there, rather than simply answer questions for us? Peng sees an opportunity to bring some of the curiosity and discovery of the early internet back into how we explore information, while giving people more control over how they navigate human knowledge.
