Formatted Title:

Artificial Intelligence: A Promising Tool for Forensic Source Identification

Background/Objectives:

Environmental forensic investigators routinely solve complex problems in high-stakes and time-sensitive situations. For example, contaminant source identification is often used to allocate cleanup costs (frequently millions or even billions of US dollars) amongst multiple potentially responsible parties. The analyses required to identify contaminant sources can be complicated by the need to account for environmental factors (e.g., mixing, weathering, and partitioning), process large datasets, resolve data quality issues, and/or combine and consider multiple lines of evidence. ChatGPT is a relatively new artificial intelligence (AI) tool with strong problem-solving capabilities. It can help process and interpret large datasets, identify patterns in data, write and execute source code, and refine its outputs based on user (i.e., human) feedback. The objective of this work is to study the strengths and weaknesses of ChatGPT as a forensic source identification tool.

Approach/Activities:

In this case study, we compare the results of a forensic source identification investigation conducted manually (i.e., by experienced environmental professionals) to the results of the same investigation conducted using the ChatGPT tool, with and without feedback from the user. In both cases, the forensic analysis includes data pre-processing, principal component analysis (PCA), and receptor modeling. The manual analysis is conducted by a forensic investigator using off-the-shelf software and scientific data analysis packages. In the AI analysis, two different procedures are used: 1) the user supplies ChatGPT with an open-ended prompt to perform the analysis independently, so that ChatGPT makes its own decisions and executes its own steps; and 2) the user supplies ChatGPT with a prompt requiring ChatGPT to confirm with the user each decision and step in its process prior to proceeding. The results of the manual and two AI investigations are compared.

Results/Lessons Learned:

This work illustrates various strengths and weaknesses of ChatGPT as an AI tool in environmental forensics. The results of this specific case study indicate that on its own, ChatGPT completes the forensics analysis with substantial errors, but with expert (i.e., human) guidance, ChatGPT replicates the results of the manual analysis. As such, multiple potential methods to ensure and control quality of the AI analysis are further explored. This preliminary work demonstrates that ChatGPT is a promising forensic investigation tool that may improve workflow efficiency, but experienced environmental professionals must still be at the helm to ensure accuracy.

Elizabeth Clyde

Affiliation

Refine Results

Artificial Intelligence: A Promising Tool for Forensic Source Identification