Sudipta Sarangi: "Insights from Two Experiments: Using Role Models and Using ChatGPT"

On The December 4, 2023

11.00 - GATE Seminar

MILC building - 35 rue Raulin, Lyon 7
Room 308 - 3rd floor

© Clarisse Croset (Unsplash)
© Clarisse Croset (Unsplash)

Sudipta Sarangi – a Collegium - Lyon Institute for Advanced Studies fellow this academic year – will present: "Insights from Two Experiments: Using Role Models and Using ChatGPT"during the GATE seminar

“In this talk I will provide insights from two different experiments. One experiment involves a Large Language Model (LLM) making decisions which is currently being promoted as a big decision-making aid. The second experiment is a light touch intervention that involves the use of role models to nudge girls towards STEM domains (Science, Technology, Engineering, Mathematics).

The first experiment examines how ChatGPT plays the Little Treasures of Game Theory a la Goeree and Holt (2001). In other words, it explores how an LLM trained on human text behaves in strategic settings.

In the second experiment, which is a Randomized Controlled Trial (RCT), female engineering students act as role-models for high school students in Peru. We find that a brief 20-minute interaction with the role models led to an increase in preferences towards engineering in a select group, but context is key”.