In Conversation: Niamh Daly discusses her research on Exploring the Impact of Avatar Race and Gender on Trust and Collaboration

1. Dr Daly, what was the primary focus of your research this summer?

The primary focus of our research is to investigate the impact of avatar race and gender on trust in the avatar and the success of collaboration with the avatar to perform a work-like task.

2. Can you summarise the main objectives or hypotheses of your study?

The advent of ChatGpt in 2022 has completely transformed how we interact with the world, and for a lot of people, how they complete their everyday work. Generative AI, like ChatGPT, can understand and generate human-like language and, as a result, can perform a variety of language-related tasks, such as answering questions, writing essays, and engaging in conversation. When most people consider LLMs, they imagine typing into the ChatGPT prompt window in a web browser. While this is the current norm, we now have the ability to embody a virtual avatar with the LLM model, allowing users to enter the metaverse, or virtual reality, via a headset and interact with the generative AI as if it were a human figure in a virtual space. The use of LLM embodied avatars is increasing at significant pace and it’s expected that most people will spend on average 1 hour per day in the metaverse by 2026. Therefore, this extremely powerful technology offers significant opportunities, but also significant challenges. It is essential that research keeps pace with technological advances and helps us to understand how best to develop and integrate the technology into everyday use. We know that trust is essential for successful collaboration. However, we also know that we don’t trust all people the same and that stereotypes and biases often impact our judgement. For instance, research suggests that if you are of the same race as your supervisor you are likely to trust them more than if you had different racial backgrounds. Equally, we know that men are often assumed to be more work-oriented and trusted as leaders, whereas women are viewed as warmer and more trustworthy in group collaborative roles. Therefore, we want to investigate whether the race or gender of avatars impacts how people interact with them, how much they trust LLM-embodied avatars and how well they collaborate with them.

3. Did you collaborate with other researchers or institutions? If so, who were your key partners?

Yes, this project is a collaborative effort. Professors Theo Lynn and Lisa Van Der Werff played a key role in securing funding for the research, and we are working closely with Assistant Professor Tim Hubbard from the University of Notre Dame to bring it to life.

4. How did the research process evolve over the summer? Were there any significant changes in your approach?

We have collected data for the first round of this study in a lab in Chicago, and will finalise the design of our next experiment based on what we find in the first study. For instance, already, we have decided to make the task that participants are asked to complete more competitive.

5. What were the key findings or outcomes of your research?

We are in the process of preliminary analysis, but we have found that participants appear to trust certain avatars more than others.

6. Are there any potential practical applications or implications of your research?

Yes, we hope that our research will 1) provide scientific evidence on optimal avatar design and 2) to contribute to diversity research in a way that is otherwise not possible. In the existing literature on diversity, people of different races and genders bring a diversity of thought, priorities, and interests – most of which may not necessarily attributable to race or gender. Therefore, it can be difficult to identify the role of race and gender in interactions. Whereas, with LLM-embodied virtual avatars, the “brain” behind the team member is exactly the same. That means that when there are ideas, interactions, and work being done, there is absolutely no difference between the avatars. Only their appearance and voice. Therefore, exploring differences in how people work with and trust in avatars of different races and gender creates a unique opportunity to contribute to our understanding of diversity in a way otherwise not accessible (Hubbard & Aguinis, 2023)

7. What’s next?

We need to finish analysing the data and then finalise the design of the second study and launch it on Prolific. We hope to do this by the end of October.