Mónica Capra on "Harnessing Intelligent Tools to Advance Behavioral Research"
Title: "Harnessing Intelligent Tools to Advance Behavioral Research"
November 27th, 2023, 16:00 CET (GMT+1)
Registration: click here
Abstract
Intelligent technologies have the potential to revolutionize behavioral research. I present two examples of research done with collaborators that show how we could utilize new technologies to extract information for behavioral prediction and experiments. The first example uses machine learning (ML) to classify the acoustic features of the voice extracted from natural, high-stakes interactions to predict cooperation and deception. Our results show that vocal cues alone have a high level of predictive accuracy. Voice is easy to collect and contains a wealth of information about intent. The findings have implications for the law, business, the design of bots, and the collection and evaluation of survey responses. The second example uses large language models (LLMs) to create “agents” that make choices in hypothetical scenarios that mimic experiments carried out in the field. LLMs like GPT-4 soak up latent social knowledge by training on texts and ethnographies. Thus, we can create artificial agents with decision-making profiles modeled after a specific culture. As a proof-of-concept, we created LLM agents that are representative of the Hadza tribesmen of Tanzania to test for the endowment effect and for behaviors in the ultimatum game. We find that our digital tribespeople exhibit behaviors that align with observations in the field. Our results suggest the potential of LLMs to extract latent preferences of non-WEIRD populations that are hard to study and for the use of LLMs for simulation and piloting of expensive interventions in the field.