Vanderbilt Peabody faculty receive NSF grant to study how conversation shapes memory, with applications for AI and education

By Jenna Somers

In daily conversations, most people are not aware of the complex brain processes taking place that make their conversations possible. However, understanding these processes could improve how AI communicates with people and how students learn at school.

A new study aims to shed light on these processes, specifically on how people remember what they talked about in conversation. The study is led by principal investigators Sarah Brown-Schmidt, professor of psychology and human development, and Deon Benton, assistant professor of psychology and human development. They are supported by a three-year, $446,272 grant from the National Science Foundation.

This study builds on Brown-Schmidt's career of human behavior experimental research that examines the cognitive mechanisms underlying conversation. In addition to existing behavioral data, the study will include more than 100 new behavioral research participants.

Using this behavioral data on how people remember information from conversation, Benton will create a computational model that attempts to clarify the processes that underlie human memory during conversation. The study seeks to answer a fundamental question about memory: how does what we say in conversation influence what we take away from that experience in the future?

"We have discovered that people tend to remember best what they say themselves in conversation and have a comparatively worse memory for what was said to them," Brown-Schmidt said. "We also find that the words we choose influence what is remembered. For example, if your friend points out a cute dog on campus and remarks on its 'fluffy ears,' adding that extra bit of detail will make that dog- and particularly its ears- memorable. We can then build computational or artificial neural network models that are inspired by the brain that either emphasize or de-emphasize phrases like 'fluffy ears' to evaluate what impact that has on the model's ability to replicate human performance. The chief advantage of this is that it can give us concrete ideas about how memory works."

Enhancing conversational AI through human cognition

Beyond advancing basic science research, this study could influence the development of more sophisticated conversational AI systems. The study's computational modeling will create a framework for improving AI conversational agents by implementing into these agents insights about the role of conversation and language on human memory, which could potentially strengthen the agents' understanding of humans and therefore, their responses to prompts.

"If you are engaging in a dialogue with a conversational agent, like an LLM, we can make predictions about what information people are likely to remember from that exchange," Brown-Schmidt said. "If the programmers designing LLMs knew that human users are not going to have a very good memory for certain aspects of the conversation, they could improve how the AI interacts with humans to account for weaknesses in memory."

A core part of the project is training graduate students in both experimental and computational modeling techniques. Together, the team's interdisciplinary expertise in behavioral research and computational methods could be just what tech employers need to enhance user experience with AI conversational agents.

Educational applications and memory retention

By improving understanding of what people do and do not remember in conversation, this study could strengthen pedagogical practices in the classroom. For example, rather than an instructor lecturing to a class and students passively listening, asking students to restate the information in their own voice could result in better retention.

Looking ahead

As the study progresses over the next three years, its discoveries could lay the groundwork for new, evidence-based technological advancements and educational practices that are designed to take advantage of new discoveries in the science of language and memory. This collaboration underscores Vanderbilt's commitment to training the next generation of scientists who can work at the intersection of psychology, computational modeling, and technology-skills increasingly vital in an AI-driven world.