Application of Generative Artificial Intelligence (GenAI) in Chemical Risk Assessment: A Study Evaluation Use Case
On this page:
The US Environmental Protection Agency (US EPA) conducts chemical risk assessments to identify the potential harm chemical exposures pose to human health and the environment. Systematic review methods are used to identify, evaluate, and integrate evidence into an assessment and derive toxicity values that can be used to support decision-making. Study evaluation is a critical step to assess whether evidence presented in a study is sufficient for derivation of toxicity values. To conduct a study evaluation, an expert reviewer will assess whether the evidence presented in a study meets predefined criteria. The outputs are judgement calls and narrative justifications within the context of respective evaluation criteria. Evaluating each of the many studies needed for a typical assessment requires tremendous resources. Advances in machine learning and other artificial intelligence technologies have the potential to dramatically reduce the required resources. Specifically, generative artificial intelligence (GenAI) models like Meta’s Llama3 or Azure OpenAI can be used to assist reviewers and potentially lower the time required to conduct study evaluation or improve the user experience of study evaluation tools. Our team will present a pilot project to perform study evaluations with assistance from GenAI that includes generating judgements with narrative justifications and a chatbot to ask questions about the document or collection of documents. The pilot will give users access to GenAI models and serve as a platform to support other use cases. The presentation will review the software architecture and our plan for performance evaluation. The views expressed in this presentation are those of the author and do not necessarily reflect the views or policies of the U.S. EPA