About the project
Large Language Models (LLM) are becoming ubiquitous. New applications are introduced every week, the great majority of which rely on untested assumptions about human-LLM interaction. This project aims to explore how people actually interact with LLMs.
Taking an experimental approach, the AICOM project will investigate how we engage in conversations with LLMs, focusing on the human side of the interaction.
The project will shed light on the communicative expectations and biases we bring to our conversations with LLMs – whether brought in from ‘ordinary’ communication or unique to LLM interactions.
Purpose
The project aims to find out whether we interpret the ‘utterances’ of LLMs in the same way as we understand each another. We will investigate three distinct groups with different communication needs and styles:
- Typically developing children
- Individuals on the autism spectrum
- Neurotypical adults
For each group, we will investigate how they interact with LLMs: Do they trust them, how do they learn from them, and how different is this interaction compared to communication with real people?
A second research focus is how LLMs are affected by human communication tactics. We can induce LLMs to ‘accept’ false information by presupposing it in what we say to them. We will investigate whether it is possible to enhance LLMs’ ability to react more appropriately.
