Eliciting Uncertainty in Chain-of-Thought to Mitigate Bias against Forecasting Harmful User Behaviors

TL;DR

This paper investigates how eliciting uncertainty in large language models can improve conversation forecasting for social media moderation by reducing bias and enhancing prediction accuracy without extensive training data.

Contribution

It introduces methods to incorporate uncertainty into LLMs for conversation forecasting, demonstrating bias mitigation and improved accuracy in social media moderation tasks.

Findings

Uncertainty elicitation improves forecasting accuracy.

Model bias against harmful outcomes is reduced with uncertainty methods.

Effective bias mitigation achieved with minimal additional data.

Abstract

Conversation forecasting tasks a model with predicting the outcome of an unfolding conversation. For instance, it can be applied in social media moderation to predict harmful user behaviors before they occur, allowing for preventative interventions. While large language models (LLMs) have recently been proposed as an effective tool for conversation forecasting, it's unclear what biases they may have, especially against forecasting the (potentially harmful) outcomes we request them to predict during moderation. This paper explores to what extent model uncertainty can be used as a tool to mitigate potential biases. Specifically, we ask three primary research questions: 1) how does LLM forecasting accuracy change when we ask models to represent their uncertainty; 2) how does LLM bias change when we ask models to represent their uncertainty; 3) how can we use uncertainty representations to reduce or completely mitigate biases without many training data points. We address these questions for 5 open-source language models tested on 2 datasets designed to evaluate conversation forecasting for social media moderation.