Bias in Generative AI (webinar recording)

This May 2024 Scottish AI Alliance workshop focused on Bias in Generative AI, specifically on text-generating technologies called Large Language Models (LLMs), such as ChatGPT. University of Edinburgh PhD student Lara Dal Molin moderated the session exploring concrete instances of bias in an open-source LLM.

Webinar structure:

Lara introduced the basic functionalities of LLMs and the emerging issue of bias within these.

The session was organized into three interactive activities - Prompting, Scoring and Designing Stories - which participants carried out in groups. These activities were facilitated by University of Edinburgh and Edinburgh Futures Institute students Carla Garcia, Savannah Thomson and Antara Sutradhar.

In Prompting, participants were invited to discuss relevant dimensions of their identities and guided through a prompt engineering exercise. In Scoring, participants observed model outputs and reflect on their detectable extent of bias. In Designing Stories, participants analyzed emerging themes in the model's responses and envision values-oriented alternatives.