Keywords: Simulation, ABM, Text
Need: The use of Large Language Models (LLMs) as Autonomous Agents is a nascent field but has already generated a large amount of interest for a variety of use-cases, new packages and corresponding workstreams. Within this framework, the language model is prompted to take on various roles, is allowed to interact with its environment, make use of a memory store, and even access tools to improve its ability at specific tasks.
This project would look to explore the use of this paradigm to model healthcare scenarios, generate complex interactions and align to non-linear outcome objectives, which can be challenging in other modelling simulations. It will look to take advantage of the underlying language model’s (or collection of models) abilities and explore pragmatic ways evaluate limitations of current techniques.
An area of particular interest is how the language model might behave when acting as an intelligence layer from different agent viewpoints within the wider simulation. Complex logs of actions and exchanges within the environment can then also be studied to gain a better understanding of how scenarios have played out, and then iterated on.
As this field that is evolving quickly, the project will be expected to build out an initial working example over a simplified pathway (such as the one described in the SynPath – Diabetes Report), which can then be built on to incorporate new emerging and promising techniques, or a more complicated pathway.
Current Knowledge/Examples & Possible Techniques/Approaches:
Related Previous Internship Projects: N/A - this is first year of the project
Enables Future Work: Builds on the team’s work in agent-based modelling (ABMs) simulations and exploration of new uses of generative AI
Outcome/Learning Objectives: Demonstaration of the usefulness and challenges of utilising LLMs as agents to model complex interactions
Datasets: Much of the work will focus on setting up appropriate scenarios and generating outputs from them, but open healthcare datasets will be used to augment where required
Desired skill set: When applying please highlight any experience around work with agent-based modelling, natural language processing, large language models, python coding experience and software development (including any coding in the open), and any other data science experience you feel relevant.
Return to list of all available projects.