General Profile:

Hi, my name's Vishal  and I'm a first year in the CDT for Interactive Artificial Intelligence at the University of Bristol. I did my Master's in Theoretical Physics, during which I was introduced to the benefits of AI applications in high energy particle physics as part of my MSc project. This led to me applying for this PhD opportunity to better understand the fundamental concepts underlying this fascinating field. My current research interests involve modelling the growth and behaviour of mycelial systems using deep learning.

There has been a recent increase in interest surrounding table-top roleplaying games (TTRPGs) by the AI research community. This has been motivated by increased use of AI applications such as ChatGPT to support players and organisers of these games in tasks such as character creation, information retrieval for specific rules and supporting/replacing the Game Masters (GMs) who organise these games. However, such applications have faced critique for generating algorithmic and stale content which detracts from the game experience. A persistent problem with TTRPGs is also the lack of availability of players and GMs especially for new players which forms a hard barrier to entry. The PhD project will design and deploy AI agents that can tackle the problem of playing/organising TTRPGs in collaboration with human players, in situations where there is a shortage of players/GMs. The research will also investigate how to develop AI agents which can devise creative solutions to problems in the dynamic, open-ended environments of TTRPGs.

Research Project Summary:

The project will initially aim to design AI RPG players that can progress through a series of tasks in which they have to come up with creative solutions to the problems presented. It will draw on preliminary research from the field of computational narrative intelligence, which has proposed some methods for modelling AI agents that can progress through a TTRPG environment or an analogous text adventure environment. This will be combined with growing research on intrinsically motivated reinforcement learning agents, to model how AI agents balance their own goals and motivations with those of other agents (human or AI) to play TTRPGs in a collaborative, engaging and creative manner. This will involve breaking down TTRPG "campaigns" into sub-tasks such as: combat, puzzle solving and open-ended/goal-oriented dialogue scenarios, and tackling each of these in turn, initially with AI agents only, followed by human-AI studies.

The value of this research is to the domain of roleplaying games. Rather than focussing on improving the ability of existing digital assistants, the research aims to develop AI agents that can collaborate with human players in a creative manner to play or organise these games. It will also investigate creative and collaborative problem solving and decision-making under uncertainty. These attributes apply to a variety of domains such as healthcare and education. In the former, work has already been carried out on agents that can aid with cognitive behavioural therapy and with the management of chronic pain. In the latter, AI could be used to assist educators at all levels by collaborating with them to create resources that are engaging and deliver their ideas in the most receptive way for the specific audience.